summaryrefslogtreecommitdiffstats
path: root/perl_checker.src
diff options
context:
space:
mode:
authorPascal Rigaux <pixel@mandriva.com>2007-04-25 15:08:17 +0000
committerPascal Rigaux <pixel@mandriva.com>2007-04-25 15:08:17 +0000
commit1a06fa7e4a300880848047118f0adba68d38348d (patch)
treee6b01d6f4feae969f9905d5245648532db254c42 /perl_checker.src
parente895f6b48826f09aeaada321d03a1d10548fc9ce (diff)
downloadperl-MDK-Common-1a06fa7e4a300880848047118f0adba68d38348d.tar
perl-MDK-Common-1a06fa7e4a300880848047118f0adba68d38348d.tar.gz
perl-MDK-Common-1a06fa7e4a300880848047118f0adba68d38348d.tar.bz2
perl-MDK-Common-1a06fa7e4a300880848047118f0adba68d38348d.tar.xz
perl-MDK-Common-1a06fa7e4a300880848047118f0adba68d38348d.zip
re-sync after the big svn loss
Diffstat (limited to 'perl_checker.src')
-rw-r--r--perl_checker.src/.cvsignore15
-rw-r--r--perl_checker.src/Makefile34
-rw-r--r--perl_checker.src/OCamlMakefile912
-rw-r--r--perl_checker.src/build.mli3
-rw-r--r--perl_checker.src/common.ml1005
-rw-r--r--perl_checker.src/common.mli276
-rw-r--r--perl_checker.src/config_file.ml40
-rw-r--r--perl_checker.src/config_file.mli6
-rw-r--r--perl_checker.src/flags.ml43
-rw-r--r--perl_checker.src/flags.mli22
-rw-r--r--perl_checker.src/global_checks.ml639
-rw-r--r--perl_checker.src/global_checks.mli26
-rw-r--r--perl_checker.src/info.ml76
-rw-r--r--perl_checker.src/info.mli17
-rw-r--r--perl_checker.src/lexer.mll1057
-rw-r--r--perl_checker.src/parser.mly500
-rw-r--r--perl_checker.src/parser_helper.ml1409
-rw-r--r--perl_checker.src/parser_helper.mli314
-rw-r--r--perl_checker.src/perl_checker.html.pl168
-rw-r--r--perl_checker.src/perl_checker.ml183
-rw-r--r--perl_checker.src/perl_checker.mli1
-rw-r--r--perl_checker.src/print.ml0
-rw-r--r--perl_checker.src/print.mli1
-rw-r--r--perl_checker.src/test/.cvsignore2
-rw-r--r--perl_checker.src/test/Makefile3
-rw-r--r--perl_checker.src/test/context.t41
-rw-r--r--perl_checker.src/test/force_layout.t23
-rw-r--r--perl_checker.src/test/method.t11
-rw-r--r--perl_checker.src/test/prototype.t23
-rw-r--r--perl_checker.src/test/read_t.pm28
-rw-r--r--perl_checker.src/test/return_value.t23
-rw-r--r--perl_checker.src/test/suggest_better.t112
-rw-r--r--perl_checker.src/test/syntax_restrictions.t70
-rwxr-xr-xperl_checker.src/test/test_it113
-rw-r--r--perl_checker.src/test/various_errors.t61
-rw-r--r--perl_checker.src/tree.ml443
-rw-r--r--perl_checker.src/tree.mli57
-rw-r--r--perl_checker.src/types.mli125
38 files changed, 0 insertions, 7882 deletions
diff --git a/perl_checker.src/.cvsignore b/perl_checker.src/.cvsignore
deleted file mode 100644
index 8c0f1f4..0000000
--- a/perl_checker.src/.cvsignore
+++ /dev/null
@@ -1,15 +0,0 @@
-._bcdi
-._d
-._ncdi
-*.cmi
-*.cmo
-*.cmx
-perl_checker
-perl_checker.html
-perl_checker_debug
-gmon.out
-lexer.ml
-parser.ml
-parser.mli
-parser.output
-build.ml
diff --git a/perl_checker.src/Makefile b/perl_checker.src/Makefile
deleted file mode 100644
index 22a45a6..0000000
--- a/perl_checker.src/Makefile
+++ /dev/null
@@ -1,34 +0,0 @@
-# OCAMLC = ocamlcp -p a
-OCAMLBCFLAGS = -w A -w e
-YFLAGS = -v
-TRASH = parser.output perl_checker.html TAGS
-RESULT = perl_checker
-BCSUFFIX = _debug
-SOURCES = types.mli build.ml common.ml flags.ml config_file.ml info.ml parser_helper.ml parser.mly lexer.mll tree.ml global_checks.ml perl_checker.ml
-LIBS = unix
-VENDORLIB = $(shell dirname `pwd`)
-DEBUG = 1
-
-default: TAGS build_ml build.ml debug-code native-code perl_checker.html
-
-build_ml:
- rm -f build.ml
- $(MAKE) build.ml
-
-build.ml:
- date '+let date = "%s"' > $@
- echo 'let fake_packages_dir = "'$(VENDORLIB)'/perl_checker_fake_packages"' >> $@
- echo 'let debugging = $(DEBUG) > 0' >> $@
-
-%.html: %.html.pl
- rm -f $@
- perl $< > $@
- chmod a-w $@
-
-tags:
- ocamltags *.ml
-
-TAGS:
- ocamltags *.ml
-
--include OCamlMakefile
diff --git a/perl_checker.src/OCamlMakefile b/perl_checker.src/OCamlMakefile
deleted file mode 100644
index 95df83f..0000000
--- a/perl_checker.src/OCamlMakefile
+++ /dev/null
@@ -1,912 +0,0 @@
-###########################################################################
-# OCamlMakefile
-# Copyright (C) 1999-2002 Markus Mottl
-#
-# For updates see:
-# http://www.oefai.at/~markus/ocaml_sources
-#
-# $Id$
-#
-###########################################################################
-
-# Set these variables to the names of the sources to be processed and
-# the result variable. Order matters during linkage!
-
-ifndef SOURCES
- SOURCES := foo.ml
-endif
-export SOURCES
-
-ifndef RES_CLIB_SUF
- RES_CLIB_SUF := _stubs
-endif
-export RES_CLIB_SUF
-
-ifndef RESULT
- RESULT := foo
-endif
-export RESULT
-
-ifndef DOC_FILES
- DOC_FILES := $(filter %.mli, $(SOURCES))
-endif
-export DOC_FILES
-
-export BCSUFFIX
-export NCSUFFIX
-
-ifndef TOPSUFFIX
- TOPSUFFIX := .top
-endif
-
-export TOPSUFFIX
-
-# Eventually set include- and library-paths, libraries to link,
-# additional compilation-, link- and ocamlyacc-flags
-# Path- and library information needs not be written with "-I" and such...
-# Define THREADS if you need it, otherwise leave it unset (same for
-# USE_CAMLP4)!
-
-export THREADS
-export USE_CAMLP4
-
-export INCDIRS
-export LIBDIRS
-export EXTLIBDIRS
-export OCAML_DEFAULT_DIRS
-export OCAML_LIB_INSTALL
-
-export LIBS
-export CLIBS
-
-export OCAMLFLAGS
-export OCAMLNCFLAGS
-export OCAMLBCFLAGS
-
-export OCAMLLDFLAGS
-export OCAMLNLDFLAGS
-export OCAMLBLDFLAGS
-
-ifndef OCAMLCPFLAGS
- OCAMLCPFLAGS := a
-endif
-
-export OCAMLCPFLAGS
-
-export YFLAGS
-export IDLFLAGS
-
-export OCAMLDOCFLAGS
-
-export DVIPSFLAGS
-
-export STATIC
-
-# Add a list of optional trash files that should be deleted by "make clean"
-export TRASH
-
-#################### variables depending on your OCaml-installation
-
-ifdef MINGW
- export MINGW
- WIN32 := 1
-endif
-ifdef MSVC
- export MSVC
- WIN32 := 1
- EXT_OBJ := obj
- EXT_LIB := lib
- ifeq ($(CC),gcc)
- # work around GNU Make default value
- ifdef THREADS
- CC := cl /MT
- else
- CC := cl
- endif
- endif
- ifeq ($(CXX),g++)
- # work around GNU Make default value
- CXX := $(CC)
- endif
- CFLAG_O := -Fo
-endif
-ifdef WIN32
- EXT_CXX := cpp
- EXE := .exe
-endif
-
-ifndef EXT_OBJ
- EXT_OBJ := o
-endif
-ifndef EXT_LIB
- EXT_LIB := a
-endif
-ifndef EXT_CXX
- EXT_CXX := cc
-endif
-ifndef EXE
- EXE := # empty
-endif
-ifndef CFLAG_O
- CFLAG_O := -o # do not delete this comment (preserves trailing whitespace)!
-endif
-
-export CC
-export CXX
-export CFLAGS
-export CXXFLAGS
-export LDFLAGS
-
-BCRESULT := $(addsuffix $(BCSUFFIX), $(RESULT))
-NCRESULT := $(addsuffix $(NCSUFFIX), $(RESULT))
-TOPRESULT := $(addsuffix $(TOPSUFFIX), $(RESULT))
-
-ifndef OCAMLC
- OCAMLC := ocamlc
-endif
-
-export OCAMLC
-
-ifndef OCAMLOPT
- OCAMLOPT := ocamlopt
-endif
-
-export OCAMLOPT
-
-ifndef OCAMLMKTOP
- OCAMLMKTOP := ocamlmktop
-endif
-
-export OCAMLMKTOP
-
-ifndef OCAMLCP
- OCAMLCP := ocamlcp
-endif
-
-export OCAMLCP
-
-ifndef OCAMLDEP
- OCAMLDEP := ocamldep
-endif
-
-export OCAMLDEP
-
-ifndef OCAMLLEX
- OCAMLLEX := ocamllex
-endif
-
-export OCAMLLEX
-
-ifndef OCAMLYACC
- OCAMLYACC := ocamlyacc
-endif
-
-export OCAMLYACC
-
-ifndef CAMELEON_REPORT
- CAMELEON_REPORT := report
-endif
-
-ifndef CAMELEON_REPORT_FLAGS
- CAMELEON_REPORT_FLAGS :=
-endif
-
-ifndef CAMELEON_ZOGGY
- CAMELEON_ZOGGY := camlp4o pa_zog.cma pr_o.cmo
-endif
-
-ifndef CAMELEON_ZOGGY_FLAGS
- CAMELEON_ZOGGY_FLAGS :=
-endif
-
-ifndef CAMLIDL
- CAMLIDL := camlidl
-endif
-
-export CAMLIDL
-
-ifndef CAMLIDLDLL
- CAMLIDLDLL := camlidldll
-endif
-
-export CAMLIDLDLL
-
-ifndef NOIDLHEADER
- MAYBE_IDL_HEADER := -header
-endif
-
-export NOIDLHEADER
-
-ifndef CAMLP4
- CAMLP4 := camlp4
-endif
-
-export CAMLP4
-
-ifndef OCAMLDOC
- OCAMLDOC := ocamldoc
-endif
-
-export OCAMLDOC
-
-ifndef LATEX
- LATEX := latex
-endif
-
-export LATEX
-
-ifndef DVIPS
- DVIPS := dvips
-endif
-
-export DVIPS
-
-ifndef PS2PDF
- PS2PDF := ps2pdf
-endif
-
-export PS2PDF
-
-ifndef OCAMLMAKEFILE
- OCAMLMAKEFILE := OCamlMakefile
-endif
-
-export OCAMLMAKEFILE
-
-ifndef OCAMLLIBPATH
- OCAMLLIBPATH := \
- $(shell $(OCAMLC) 2>/dev/null -where || echo /usr/local/lib/ocaml)
-endif
-
-export OCAMLLIBPATH
-
-ifndef OCAML_LIB_INSTALL
- OCAML_LIB_INSTALL := $(OCAMLLIBPATH)/contrib
-endif
-
-export OCAML_LIB_INSTALL
-
-###########################################################################
-
-#################### change following sections only if
-#################### you know what you are doing!
-
-# delete target files when a build command fails
-.PHONY: .DELETE_ON_ERROR
-.DELETE_ON_ERROR:
-
-# for pedants using "--warn-undefined-variables"
-export MAYBE_IDL
-export REAL_RESULT
-export CAMLIDLFLAGS
-export THREAD_FLAG
-export RES_CLIB
-export MAKEDLL
-
-SHELL := /bin/sh
-
-MLDEPDIR := ._d
-BCDIDIR := ._bcdi
-NCDIDIR := ._ncdi
-
-FILTER_EXTNS := %.mli %.ml %.mll %.mly %.idl %.c %.$(EXT_CXX) %.rep %.zog
-
-FILTERED := $(filter $(FILTER_EXTNS), $(SOURCES))
-SOURCE_DIRS := $(filter-out ./, $(sort $(dir $(FILTERED))))
-
-FILTERED_REP := $(filter %.rep, $(FILTERED))
-DEP_REP := $(FILTERED_REP:%.rep=$(MLDEPDIR)/%.d)
-AUTO_REP := $(FILTERED_REP:.rep=.ml)
-
-FILTERED_ZOG := $(filter %.zog, $(FILTERED))
-DEP_ZOG := $(FILTERED_ZOG:%.zog=$(MLDEPDIR)/%.d)
-AUTO_ZOG := $(FILTERED_ZOG:.zog=.ml)
-
-FILTERED_ML := $(filter %.ml, $(FILTERED))
-DEP_ML := $(FILTERED_ML:%.ml=$(MLDEPDIR)/%.d)
-
-FILTERED_MLI := $(filter %.mli, $(FILTERED))
-DEP_MLI := $(FILTERED_MLI:.mli=.di)
-
-FILTERED_MLL := $(filter %.mll, $(FILTERED))
-DEP_MLL := $(FILTERED_MLL:%.mll=$(MLDEPDIR)/%.d)
-AUTO_MLL := $(FILTERED_MLL:.mll=.ml)
-
-FILTERED_MLY := $(filter %.mly, $(FILTERED))
-DEP_MLY := $(FILTERED_MLY:%.mly=$(MLDEPDIR)/%.d) $(FILTERED_MLY:.mly=.di)
-AUTO_MLY := $(FILTERED_MLY:.mly=.mli) $(FILTERED_MLY:.mly=.ml)
-
-FILTERED_IDL := $(filter %.idl, $(FILTERED))
-DEP_IDL := $(FILTERED_IDL:%.idl=$(MLDEPDIR)/%.d) $(FILTERED_IDL:.idl=.di)
-C_IDL := $(FILTERED_IDL:%.idl=%_stubs.c) $(FILTERED_IDL:.idl=.h)
-OBJ_C_IDL := $(FILTERED_IDL:%.idl=%_stubs.$(EXT_OBJ))
-AUTO_IDL := $(FILTERED_IDL:.idl=.mli) $(FILTERED_IDL:.idl=.ml) $(C_IDL)
-
-FILTERED_C_CXX := $(filter %.c %.$(EXT_CXX), $(FILTERED))
-OBJ_C_CXX := $(FILTERED_C_CXX:.c=.$(EXT_OBJ))
-OBJ_C_CXX := $(OBJ_C_CXX:.$(EXT_CXX)=.$(EXT_OBJ))
-
-PRE_TARGETS += $(AUTO_MLL) $(AUTO_MLY) $(AUTO_IDL) $(AUTO_ZOG) $(AUTO_REP)
-
-ALL_DEPS := $(DEP_ML) $(DEP_MLI) $(DEP_MLL) $(DEP_MLY) $(DEP_IDL) $(DEP_ZOG) $(DEP_REP)
-
-MLDEPS := $(filter %.d, $(ALL_DEPS))
-MLIDEPS := $(filter %.di, $(ALL_DEPS))
-BCDEPIS := $(MLIDEPS:%.di=$(BCDIDIR)/%.di)
-NCDEPIS := $(MLIDEPS:%.di=$(NCDIDIR)/%.di)
-
-ALLML := $(filter %.mli %.ml %.mll %.mly %.idl %.rep %.zog, $(FILTERED))
-
-IMPLO_INTF := $(ALLML:%.mli=%.mli.__)
-IMPLO_INTF := $(foreach file, $(IMPLO_INTF), \
- $(basename $(file)).cmi $(basename $(file)).cmo)
-IMPLO_INTF := $(filter-out %.mli.cmo, $(IMPLO_INTF))
-IMPLO_INTF := $(IMPLO_INTF:%.mli.cmi=%.cmi)
-
-IMPLX_INTF := $(IMPLO_INTF:.cmo=.cmx)
-
-INTF := $(filter %.cmi, $(IMPLO_INTF))
-IMPL_CMO := $(filter %.cmo, $(IMPLO_INTF))
-IMPL_CMX := $(IMPL_CMO:.cmo=.cmx)
-
-OBJ_LINK := $(OBJ_C_IDL) $(OBJ_C_CXX)
-OBJ_FILES := $(IMPL_CMO:.cmo=.$(EXT_OBJ)) $(OBJ_LINK)
-
-EXECS := $(addsuffix $(EXE), \
- $(sort $(TOPRESULT) $(BCRESULT) $(NCRESULT)))
-ifdef WIN32
- EXECS += $(BCRESULT).dll $(NCRESULT).dll
-endif
-
-CLIB_BASE := $(RESULT)$(RES_CLIB_SUF)
-ifneq ($(strip $(OBJ_LINK)),)
- RES_CLIB := lib$(CLIB_BASE).$(EXT_LIB)
-endif
-
-ifndef MSVC
-DLLSONAME := dll$(CLIB_BASE).so
-endif
-
-NONEXECS := $(INTF) $(IMPL_CMO) $(IMPL_CMX) $(OBJ_FILES) $(PRE_TARGETS) \
- $(BCRESULT).cma $(NCRESULT).cmxa $(NCRESULT).$(EXT_LIB) \
- $(BCRESULT).cmi $(BCRESULT).cmo \
- $(NCRESULT).cmi $(NCRESULT).cmx $(NCRESULT).o \
- $(RES_CLIB)
-
-ifndef MSVC
- NONEXECS += $(DLLSONAME)
-endif
-
-ifndef LIBINSTALL_FILES
- LIBINSTALL_FILES := $(RESULT).mli $(RESULT).cmi $(RESULT).cma \
- $(RESULT).cmxa $(RESULT).a $(RES_CLIB)
-endif
-
-ifndef MSVC
- LIBINSTALL_FILES += $(DLLSONAME)
-endif
-
-export LIBINSTALL_FILES
-
-ifdef WIN32
- # some extra stuff is created while linking DLLs
- NONEXECS += $(BCRESULT).$(EXT_LIB) $(BCRESULT).exp $(NCRESULT).exp
-endif
-
-TARGETS := $(EXECS) $(NONEXECS)
-
-# handle ocamlfind
-ifdef USING_OCAMLFIND
- PACKOPT := -pack
-else
- PACKOPT := -passopt "-pack"
-endif
-
-# If there are IDL-files
-ifneq ($(strip $(FILTERED_IDL)),)
- MAYBE_IDL := -cclib -lcamlidl
-endif
-
-ifdef USE_CAMLP4
- CAMLP4PATH := \
- $(shell $(CAMLP4) -where 2>/dev/null || echo /usr/local/lib/camlp4)
- INCFLAGS := -I $(CAMLP4PATH)
- CINCFLAGS := -I$(CAMLP4PATH)
-endif
-
-INCFLAGS += $(SOURCE_DIRS:%=-I %) $(INCDIRS:%=-I %) $(OCAML_DEFAULT_DIRS:%=-I %)
-CINCFLAGS += $(SOURCE_DIRS:%=-I%) $(INCDIRS:%=-I%) $(OCAML_DEFAULT_DIRS:%=-I%)
-CLIBFLAGS += $(SOURCE_DIRS:%=-L%) $(LIBDIRS:%=-L%) \
- $(EXTLIBDIRS:%=-L%) $(EXTLIBDIRS:%=-R%) \
- $(OCAML_DEFAULT_DIRS:%=-L%)
-
-ifndef PROFILING
- INTF_OCAMLC := $(OCAMLC)
-else
- ifndef THREADS
- INTF_OCAMLC := $(OCAMLCP) -p $(OCAMLCPFLAGS)
- else
- # OCaml does not support profiling byte code
- # with threads (yet), therefore we force an error.
- ifndef REAL_OCAMLC
- $(error Profiling of multithreaded byte code not yet supported by OCaml)
- endif
- endif
-endif
-
-ifndef MSVC
- COMMON_LDFLAGS := $(LDFLAGS:%=-ccopt %) $(SOURCE_DIRS:%=-ccopt -L%) \
- $(LIBDIRS:%=-ccopt -L%) \
- $(EXTLIBDIRS:%=-ccopt -L%) $(EXTLIBDIRS:%=-ccopt -R%) \
- $(OCAML_DEFAULT_DIRS:%=-ccopt -L%)
-else
- # currenly MSVC-build ocamlc/ocamlopt cannot pass any option to C linker :-(
- COMMON_LDFLAGS :=
-endif
-
-ifndef MSVC
- CLIBS_OPTS := $(CLIBS:%=-cclib -l%)
-else
- # MSVC libraries do not have 'lib' prefix
- CLIBS_OPTS := $(CLIBS:%=-ccopt %)
-endif
-ifneq ($(strip $(OBJ_LINK)),)
- ifdef CREATE_LIB
- OBJS_LIBS := -cclib -l$(CLIB_BASE) $(CLIBS_OPTS) $(MAYBE_IDL)
- else
- OBJS_LIBS := $(OBJ_LINK) $(CLIBS_OPTS) $(MAYBE_IDL)
- endif
-else
- OBJS_LIBS := $(CLIBS_OPTS) $(MAYBE_IDL)
-endif
-
-# If we have to make byte-code
-ifndef REAL_OCAMLC
- # EXTRADEPS is added dependencies we have to insert for all
- # executable files we generate. Ideally it should be all of the
- # libraries we use, but it's hard to find the ones that get searched on
- # the path since I don't know the paths built into the compiler, so
- # just include the ones with slashes in their names.
- EXTRADEPS := $(addsuffix .cma,$(foreach i,$(LIBS),$(if $(findstring /,$(i)),$(i))))
- SPECIAL_OCAMLFLAGS := $(OCAMLBCFLAGS)
-
- REAL_OCAMLC := $(INTF_OCAMLC)
-
- REAL_IMPL := $(IMPL_CMO)
- REAL_IMPL_INTF := $(IMPLO_INTF)
- IMPL_SUF := .cmo
-
- DEPFLAGS :=
- MAKE_DEPS := $(MLDEPS) $(BCDEPIS)
-
- ifdef CREATE_LIB
- ifndef STATIC
- ifneq ($(strip $(OBJ_LINK)),)
- MAKEDLL := $(DLLSONAME)
- ALL_LDFLAGS := -dllib $(DLLSONAME)
- endif
- endif
- endif
-
- ifndef NO_CUSTOM
- ifneq "$(strip $(OBJ_LINK) $(THREADS) $(MAYBE_IDL) $(CLIBS))" ""
- ALL_LDFLAGS += -custom
- endif
- endif
-
- ALL_LDFLAGS += $(INCFLAGS) $(OCAMLLDFLAGS) $(OCAMLBLDFLAGS) \
- $(COMMON_LDFLAGS) $(LIBS:%=%.cma)
- CAMLIDLDLLFLAGS :=
-
- ifdef THREADS
- ALL_LDFLAGS := -thread $(ALL_LDFLAGS)
- ifndef CREATE_LIB
- ALL_LDFLAGS := unix.cma threads.cma $(ALL_LDFLAGS)
- endif
- THREAD_FLAG := -thread
- endif
-
-# we have to make native-code
-else
- EXTRADEPS := $(addsuffix .cmxa,$(foreach i,$(LIBS),$(if $(findstring /,$(i)),$(i))))
- ifndef PROFILING
- SPECIAL_OCAMLFLAGS := $(OCAMLNCFLAGS)
- PLDFLAGS :=
- else
- SPECIAL_OCAMLFLAGS := -p $(OCAMLNCFLAGS)
- PLDFLAGS := -p
- endif
-
- REAL_IMPL := $(IMPL_CMX)
- REAL_IMPL_INTF := $(IMPLX_INTF)
- IMPL_SUF := .cmx
-
- CFLAGS := -DNATIVE_CODE $(CFLAGS)
-
- DEPFLAGS := -native
- MAKE_DEPS := $(MLDEPS) $(NCDEPIS)
-
- ALL_LDFLAGS := $(PLDFLAGS) $(INCFLAGS) $(OCAMLLDFLAGS) \
- $(OCAMLNLDFLAGS) $(COMMON_LDFLAGS)
- CAMLIDLDLLFLAGS := -opt
-
- ifndef CREATE_LIB
- ALL_LDFLAGS += $(LIBS:%=%.cmxa)
- endif
-
- ifdef THREADS
- ALL_LDFLAGS := -thread $(ALL_LDFLAGS)
- ifndef CREATE_LIB
- ALL_LDFLAGS := unix.cmxa threads.cmxa $(ALL_LDFLAGS)
- endif
- THREAD_FLAG := -thread
- endif
-endif
-
-export MAKE_DEPS
-
-ALL_OCAMLCFLAGS := $(THREAD_FLAG) $(OCAMLFLAGS) \
- $(INCFLAGS) $(SPECIAL_OCAMLFLAGS)
-
-ifdef make_deps
- -include $(MAKE_DEPS)
- PRE_TARGETS :=
-endif
-
-###########################################################################
-# USER RULES
-
-# Call "OCamlMakefile QUIET=" to get rid of all of the @'s.
-QUIET=@
-
-# generates byte-code (default)
-byte-code: $(PRE_TARGETS)
- $(QUIET)$(MAKE) -r -f $(OCAMLMAKEFILE) $(BCRESULT) \
- REAL_RESULT="$(BCRESULT)" make_deps=yes
-bc: byte-code
-
-byte-code-nolink: $(PRE_TARGETS)
- $(QUIET)$(MAKE) -r -f $(OCAMLMAKEFILE) nolink \
- REAL_RESULT="$(BCRESULT)" make_deps=yes
-bcnl: byte-code-nolink
-
-top: $(PRE_TARGETS)
- $(QUIET)$(MAKE) -r -f $(OCAMLMAKEFILE) $(TOPRESULT) \
- REAL_RESULT="$(BCRESULT)" make_deps=yes
-
-# generates native-code
-
-native-code: $(PRE_TARGETS)
- $(QUIET)$(MAKE) -r -f $(OCAMLMAKEFILE) $(NCRESULT) \
- REAL_RESULT="$(NCRESULT)" \
- REAL_OCAMLC="$(OCAMLOPT)" \
- make_deps=yes
-nc: native-code
-
-native-code-nolink: $(PRE_TARGETS)
- $(QUIET)$(MAKE) -r -f $(OCAMLMAKEFILE) nolink \
- REAL_RESULT="$(NCRESULT)" \
- REAL_OCAMLC="$(OCAMLOPT)" \
- make_deps=yes
-ncnl: native-code-nolink
-
-# generates byte-code libraries
-byte-code-library: $(PRE_TARGETS)
- $(QUIET)$(MAKE) -r -f $(OCAMLMAKEFILE) \
- $(RES_CLIB) $(BCRESULT).cma \
- REAL_RESULT="$(BCRESULT)" \
- CREATE_LIB=yes \
- make_deps=yes
-bcl: byte-code-library
-
-# generates native-code libraries
-native-code-library: $(PRE_TARGETS)
- $(QUIET)$(MAKE) -r -f $(OCAMLMAKEFILE) \
- $(RES_CLIB) $(NCRESULT).cmxa \
- REAL_RESULT="$(NCRESULT)" \
- REAL_OCAMLC="$(OCAMLOPT)" \
- CREATE_LIB=yes \
- make_deps=yes
-ncl: native-code-library
-
-ifdef WIN32
-# generates byte-code dll
-byte-code-dll: $(PRE_TARGETS)
- $(QUIET)$(MAKE) -r -f $(OCAMLMAKEFILE) \
- $(RES_CLIB) $(BCRESULT).dll \
- REAL_RESULT="$(BCRESULT)" \
- make_deps=yes
-bcd: byte-code-dll
-
-# generates native-code dll
-native-code-dll: $(PRE_TARGETS)
- $(QUIET)$(MAKE) -r -f $(OCAMLMAKEFILE) \
- $(RES_CLIB) $(NCRESULT).dll \
- REAL_RESULT="$(NCRESULT)" \
- REAL_OCAMLC="$(OCAMLOPT)" \
- make_deps=yes
-ncd: native-code-dll
-endif
-
-# generates byte-code with debugging information
-debug-code: $(PRE_TARGETS)
- $(QUIET)$(MAKE) -r -f $(OCAMLMAKEFILE) $(BCRESULT) \
- REAL_RESULT="$(BCRESULT)" make_deps=yes \
- OCAMLFLAGS="-g $(OCAMLFLAGS)" \
- OCAMLLDFLAGS="-g $(OCAMLLDFLAGS)"
-dc: debug-code
-
-# generates byte-code libraries with debugging information
-debug-code-library: $(PRE_TARGETS)
- $(QUIET)$(MAKE) -r -f $(OCAMLMAKEFILE) \
- $(RES_CLIB) $(BCRESULT).cma \
- REAL_RESULT="$(BCRESULT)" make_deps=yes \
- CREATE_LIB=yes \
- OCAMLFLAGS="-g $(OCAMLFLAGS)" \
- OCAMLLDFLAGS="-g $(OCAMLLDFLAGS)"
-dcl: debug-code-library
-
-# generates byte-code for profiling
-profiling-byte-code: $(PRE_TARGETS)
- $(QUIET)$(MAKE) -r -f $(OCAMLMAKEFILE) $(BCRESULT) \
- REAL_RESULT="$(BCRESULT)" PROFILING="y" \
- make_deps=yes
-pbc: profiling-byte-code
-
-# generates native-code
-
-profiling-native-code: $(PRE_TARGETS)
- $(QUIET)$(MAKE) -r -f $(OCAMLMAKEFILE) $(NCRESULT) \
- REAL_RESULT="$(NCRESULT)" \
- REAL_OCAMLC="$(OCAMLOPT)" \
- PROFILING="y" \
- make_deps=yes
-pnc: profiling-native-code
-
-# generates byte-code libraries
-profiling-byte-code-library: $(PRE_TARGETS)
- $(QUIET)$(MAKE) -r -f $(OCAMLMAKEFILE) \
- $(RES_CLIB) $(BCRESULT).cma \
- REAL_RESULT="$(BCRESULT)" PROFILING="y" \
- CREATE_LIB=yes \
- make_deps=yes
-pbcl: profiling-byte-code-library
-
-# generates native-code libraries
-profiling-native-code-library: $(PRE_TARGETS)
- $(QUIET)$(MAKE) -r -f $(OCAMLMAKEFILE) \
- $(RES_CLIB) $(NCRESULT).cmxa \
- REAL_RESULT="$(NCRESULT)" PROFILING="y" \
- REAL_OCAMLC="$(OCAMLOPT)" \
- CREATE_LIB=yes \
- make_deps=yes
-pncl: profiling-native-code-library
-
-# packs byte-code objects
-pack-byte-code: $(PRE_TARGETS)
- $(QUIET)$(MAKE) -r -f $(OCAMLMAKEFILE) $(BCRESULT).cmo \
- REAL_RESULT="$(BCRESULT)" \
- PACK_LIB=yes make_deps=yes
-pabc: pack-byte-code
-
-# packs native-code objects
-pack-native-code: $(PRE_TARGETS)
- $(QUIET)$(MAKE) -r -f $(OCAMLMAKEFILE) \
- $(NCRESULT).cmx $(NCRESULT).o \
- REAL_RESULT="$(NCRESULT)" \
- REAL_OCAMLC="$(OCAMLOPT)" \
- PACK_LIB=yes make_deps=yes
-panc: pack-native-code
-
-# generates HTML-documentation
-htdoc: doc/html
-
-# generates Latex-documentation
-ladoc: doc/latex
-
-# generates PostScript-documentation
-psdoc: doc/latex/doc.ps
-
-# generates PDF-documentation
-pdfdoc: doc/latex/doc.pdf
-
-# generates all supported forms of documentation
-doc: htdoc ladoc psdoc pdfdoc
-
-###########################################################################
-# LOW LEVEL RULES
-
-$(REAL_RESULT): $(REAL_IMPL_INTF) $(OBJ_LINK) $(EXTRADEPS)
- $(REAL_OCAMLC) $(ALL_LDFLAGS) $(OBJS_LIBS) -o $@$(EXE) \
- $(REAL_IMPL)
-ifdef MSVC
-# work around the bug in ocamlc -- it should delete this file itself
- rm -f camlprim?.$(EXT_OBJ)
-endif
-
-nolink: $(REAL_IMPL_INTF) $(OBJ_LINK)
-
-ifdef WIN32
-$(REAL_RESULT).dll: $(REAL_IMPL_INTF) $(OBJ_LINK)
- $(CAMLIDLDLL) $(CAMLIDLDLLFLAGS) $(OBJ_LINK) $(CLIBS) \
- -o $@ $(REAL_IMPL)
-endif
-
-%$(TOPSUFFIX): $(REAL_IMPL_INTF) $(OBJ_LINK) $(EXTRADEPS)
- $(OCAMLMKTOP) $(ALL_LDFLAGS) $(OBJS_LIBS) -o $@$(EXE) \
- $(REAL_IMPL)
-ifdef MSVC
-# work around the bug in ocamltop -- it should delete this file itself
- rm -f camlprim?.$(EXT_OBJ)
-endif
-
-.SUFFIXES: .mli .ml .cmi .cmo .cmx .cma .cmxa .$(EXT_OBJ) \
- .mly .di .d .$(EXT_LIB) .idl .c .$(EXT_CXX) .h .so \
- .rep .zog
-ifndef MSVC
-$(DLLSONAME): $(OBJ_LINK)
- $(CC) -shared $(CINCFLAGS) $(CLIBFLAGS) \
- -o $@ $(OBJ_LINK) $(CLIBS:%=-l%)
-endif
-
-$(RESULT).cma: $(REAL_IMPL_INTF) $(MAKEDLL) $(EXTRADEPS)
- $(REAL_OCAMLC) -a $(ALL_LDFLAGS) \
- $(OBJS_LIBS) -o $@ $(OCAMLBLDFLAGS) $(REAL_IMPL)
-
-$(RESULT).cmxa $(RESULT).$(EXT_LIB): $(REAL_IMPL_INTF) $(EXTRADEPS)
- $(OCAMLOPT) -a $(ALL_LDFLAGS) $(OBJS_LIBS) \
- $(OCAMLNLDFLAGS) -o $@ $(REAL_IMPL)
-
-$(RES_CLIB): $(OBJ_LINK)
-ifndef MSVC
- ifneq ($(strip $(OBJ_LINK)),)
- ar rc $@ $(OBJ_LINK)
- ranlib $@
- endif
-else
- ifneq ($(strip $(OBJ_LINK)),)
- lib /nologo /debugtype:cv /out:$(RES_CLIB) $(OBJ_LINK)
- endif
-endif
-
-.mli.cmi: $(EXTRADEPS)
- $(QUIET)pp=`sed -n -e 's/(\*pp \([^*]*\) \*)/\1/p;q' $<`; \
- if [ -z "$$pp" ]; then \
- echo $(INTF_OCAMLC) -c $(THREAD_FLAG) \
- $(OCAMLFLAGS) $(INCFLAGS) $<; \
- $(INTF_OCAMLC) -c $(THREAD_FLAG) $(OCAMLFLAGS) \
- $(INCFLAGS) $<; \
- else \
- echo $(INTF_OCAMLC) -c -pp \"$$pp\" $(THREAD_FLAG) \
- $(OCAMLFLAGS) $(INCFLAGS) $<; \
- $(INTF_OCAMLC) -c -pp "$$pp" $(THREAD_FLAG) \
- $(OCAMLFLAGS) $(INCFLAGS) $<; \
- fi
-
-.ml.cmi .ml.$(EXT_OBJ) .ml.cmx .ml.cmo: $(EXTRADEPS)
- $(QUIET)pp=`sed -n -e 's/(\*pp \([^*]*\) \*)/\1/p;q' $<`; \
- if [ -z "$$pp" ]; then \
- echo $(REAL_OCAMLC) -c $(ALL_OCAMLCFLAGS) $<; \
- $(REAL_OCAMLC) -c $(ALL_OCAMLCFLAGS) $<; \
- else \
- echo $(REAL_OCAMLC) -c -pp \"$$pp\" \
- $(ALL_OCAMLCFLAGS) $<; \
- $(REAL_OCAMLC) -c -pp "$$pp" $(ALL_OCAMLCFLAGS) $<; \
- fi
-
-ifdef PACK_LIB
-$(REAL_RESULT).cmo $(REAL_RESULT).cmx $(REAL_RESULT).o: $(REAL_IMPL_INTF) $(OBJ_LINK) $(EXTRADEPS)
- $(REAL_OCAMLC) $(PACKOPT) $(ALL_LDFLAGS) \
- $(OBJS_LIBS) -o $@ $(REAL_IMPL)
-endif
-
-.PRECIOUS: %.ml
-%.ml: %.mll
- $(OCAMLLEX) $<
-
-.PRECIOUS: %.ml %.mli
-%.ml %.mli: %.mly
- $(OCAMLYACC) $(YFLAGS) $<
-
-.PRECIOUS: %.ml
-%.ml : %.rep
- $(CAMELEON_REPORT) $(CAMELEON_REPORT_FLAGS) -gen $<
-
-.PRECIOUS: %.ml
-%.ml : %.zog
- $(CAMELEON_ZOGGY) $(CAMELEON_ZOGGY_FLAGS) -impl $< > $@
-
-.PRECIOUS: %.ml %.mli %_stubs.c %.h
-%.ml %.mli %_stubs.c %.h: %.idl
- $(CAMLIDL) $(MAYBE_IDL_HEADER) $(IDLFLAGS) \
- $(CAMLIDLFLAGS) $<
- $(QUIET)if [ $(NOIDLHEADER) ]; then touch $*.h; fi
-
-.c.$(EXT_OBJ):
- $(CC) -c $(CFLAGS) $(CINCFLAGS) -I$(OCAMLLIBPATH) \
- $< $(CFLAG_O)$@
-
-.$(EXT_CXX).$(EXT_OBJ):
- $(CXX) -c $(CXXFLAGS) $(CINCFLAGS) -I$(OCAMLLIBPATH) \
- $< $(CFLAG_O)$@
-
-$(MLDEPDIR)/%.d: %.ml
- $(QUIET)echo making $@ from $<
- $(QUIET)if [ ! -d $(@D) ]; then mkdir -p $(@D); fi
- $(QUIET)pp=`sed -n -e 's/(\*pp \([^*]*\) \*)/\1/p;q' $<`; \
- if [ -z "$$pp" ]; then \
- $(OCAMLDEP) $(INCFLAGS) $< > $@; \
- else \
- $(OCAMLDEP) -pp "$$pp" $(INCFLAGS) $< > $@; \
- fi
-
-$(BCDIDIR)/%.di $(NCDIDIR)/%.di: %.mli
- $(QUIET)echo making $@ from $<
- $(QUIET)if [ ! -d $(@D) ]; then mkdir -p $(@D); fi
- $(QUIET)pp=`sed -n -e 's/(\*pp \([^*]*\) \*)/\1/p;q' $<`; \
- if [ -z "$$pp" ]; then \
- $(OCAMLDEP) $(DEPFLAGS) $(INCFLAGS) $< > $@; \
- else \
- $(OCAMLDEP) $(DEPFLAGS) \
- -pp "$$pp" $(INCFLAGS) $< > $@; \
- fi
-
-doc/html: $(DOC_FILES)
- rm -rf $@
- mkdir -p $@
- $(OCAMLDOC) -html -d $@ $(OCAMLDOCFLAGS) $(DOC_FILES)
-
-doc/latex: $(DOC_FILES)
- rm -rf $@
- mkdir -p $@
- $(OCAMLDOC) -latex -d $@ $(OCAMLDOCFLAGS) $(DOC_FILES) -o doc.tex
-
-doc/latex/doc.ps: doc/latex
- cd doc/latex && \
- $(LATEX) doc.tex && \
- $(LATEX) doc.tex && \
- $(DVIPS) $(DVIPSFLAGS) doc.dvi -o $(@F)
-
-doc/latex/doc.pdf: doc/latex/doc.ps
- cd doc/latex && $(PS2PDF) $(<F)
-
-###########################################################################
-# (UN)INSTALL RULES FOR LIBRARIES
-
-.PHONY: libinstall
-libinstall: all
- $(QUIET)printf "\nInstalling library to: $(OCAML_LIB_INSTALL)\n"
- -install -d $(OCAML_LIB_INSTALL)
- for i in $(LIBINSTALL_FILES); do \
- if [ -f $$i ]; then \
- install -c -m 0644 $$i $(OCAML_LIB_INSTALL); \
- fi; \
- done
- $(QUIET)printf "\nInstallation successful.\n"
-
-.PHONY: libuninstall
-libuninstall:
- $(QUIET)printf "\nUninstalling library from: $(OCAML_LIB_INSTALL)\n"
- cd $(OCAML_LIB_INSTALL); rm $(notdir $(LIBINSTALL_FILES))
- $(QUIET)printf "\nUninstallation successful.\n"
-
-###########################################################################
-# MAINTAINANCE RULES
-
-.PHONY: clean
-clean:
- rm -f $(TARGETS) $(TRASH)
- rm -rf $(BCDIDIR) $(NCDIDIR) $(MLDEPDIR)
-
-.PHONY: cleanup
-cleanup:
- rm -f $(NONEXECS) $(TRASH)
- rm -rf $(BCDIDIR) $(NCDIDIR) $(MLDEPDIR)
-
-.PHONY: clean-doc
-clean-doc:
- rm -rf doc
-
-.PHONY: nobackup
-nobackup:
- rm -f *.bak *~ *.dup
diff --git a/perl_checker.src/build.mli b/perl_checker.src/build.mli
deleted file mode 100644
index 716b843..0000000
--- a/perl_checker.src/build.mli
+++ /dev/null
@@ -1,3 +0,0 @@
-val date : string
-val fake_packages_dir : string
-val debugging : bool
diff --git a/perl_checker.src/common.ml b/perl_checker.src/common.ml
deleted file mode 100644
index dd2f6b1..0000000
--- a/perl_checker.src/common.ml
+++ /dev/null
@@ -1,1005 +0,0 @@
-open Stack
-open List
-
-exception Found
-exception Not_comparable
-exception GraphSort_circular_deps
-
-type ('a, 'b) either = Left of 'a | Right of 'b
-type ('a, 'b) or_option = Or_some of 'a | Or_error of 'b
-
-(**********************************************************************************)
-
-let internal_error s = failwith ("internal error: " ^ s)
-
-let id x = x
-let double a = a,a
-let swap (x,y) = (y,x)
-let safe_tl l = try tl l with _ -> []
-let fstfst ((e, _), _) = e
-let sndfst ((_, e), _) = e
-let fstsnd (_, (e, _)) = e
-let sndsnd (_, (_, e)) = e
-
-let fst3 (e, _, _) = e
-let snd3 (_, e, _) = e
-let ter3 (_, _, e) = e
-let sndter3 (_, a, b) = (a, b)
-
-let o f g x = f (g x)
-let curry f x y = f (x,y)
-let uncurry f (x, y) = f x y
-
-let is_int n = ceil n = n
-
-let uncons = function
- | [] -> failwith "uncons"
- | e::l -> e,l
-
-let has_env var =
- try
- let _ = Sys.getenv var in true
- with Not_found -> false
-
-let some = function
- | Some e -> e
- | None -> failwith "some"
-
-let some_or = function
- | None -> id
- | Some e -> fun _ -> e
-
-let option2l = function
- | None -> []
- | Some e -> [e]
-
-let prefer_some f a b =
- match a, b with
- | Some a, Some b -> Some (f a b)
- | None, _ -> b
- | _, None -> a
-
-let rec collect_accu f accu = function
- | [] -> accu
- | e::l -> collect_accu f (rev_append (f e) accu) l
-
-let collect f l = rev (collect_accu f [] l)
-
-let merge_some merge a b =
- match a,b with
- | None, None -> None
- | _, None -> a
- | None, _ -> b
- | Some(a), Some(b) -> Some(merge a b)
-
-let rec uniq = function
- | [] -> []
- | e::l -> if mem e l then uniq l else e :: uniq l
-
-let rec uniq_ eq = function
- | [] -> []
- | e::l ->
- try
- let _ = find (eq e) l in
- uniq_ eq l
- with Not_found -> e :: uniq_ eq l
-
-let rec non_uniq = function
- | [] -> []
- | e::l -> if mem e l then e :: non_uniq l else non_uniq l
-
-let rec member_ eq e = function
- | [] -> false
- | e'::l -> if eq e e' then true else member_ eq e l
-
-let rec find_some p = function
- | [] -> raise Not_found
- | x :: l ->
- match p x with
- | Some v -> v
- | None -> find_some p l
-
-let fold_left1 f = function
- | [] -> failwith "fold_left1"
- | e :: l -> fold_left f e l
-
-let find_index e l =
- let rec find_index_ i = function
- | [] -> raise Not_found
- | e'::l -> if e=e' then i else find_index_ (i+1) l
- in
- find_index_ 0 l
-
-let rec find_some_ p = function
- | [] -> None
- | x :: l ->
- match p x with
- | Some v -> Some v
- | None -> find_some_ p l
-
-let rec fpartition p l =
- let rec part yes no = function
- | [] -> (rev yes, rev no)
- | x :: l ->
- (match p x with
- | None -> part yes (x :: no) l
- | Some v -> part (v :: yes) no l) in
- part [] [] l
-
-let partition_either f l =
- let rec part_either left right = function
- | [] -> (rev left, rev right)
- | x :: l ->
- (match f x with
- | Left e -> part_either (e :: left) right l
- | Right e -> part_either left (e :: right) l) in
- part_either [] [] l
-
-let rec keep_best f =
- let rec partition e = function
- | [] -> e, []
- | e' :: l ->
- match f(e,e') with
- | None -> let (e'', l') = partition e l in e'', e' :: l'
- | Some e'' -> partition e'' l
- in function
- | [] -> []
- | e::l ->
- let (e', l') = partition e l in
- e' :: keep_best f l'
-
-let rec keep_bests f l =
- let rec once e unchanged = function
- | [] -> None
- | e' :: l ->
- match f(e,e') with
- | None -> once e (e' :: unchanged) l
- | Some e'' -> Some(e'', unchanged @ l)
- in
- let rec as_many_as_possible e l =
- match once e [] l with
- | None -> None
- | Some(e', l') -> Some(some_or (as_many_as_possible e' l') (e', l'))
- in
- let rec try_with e l_done l_next =
- match as_many_as_possible e l_next with
- | None -> try_with_next (e :: l_done) l_next
- | Some(e2, l_next2) ->
- match as_many_as_possible e2 l_done with
- | None -> try_with_next (e2 :: l_done) l_next2
- | Some(e3, l_done2) -> try_with e3 l_done2 l_next2
- and try_with_next l_done = function
- | [] -> rev l_done
- | e::l_next -> try_with e l_done l_next
- in
- try_with_next [] l
-
-let rec fold_right1 f = function
- | [] -> failwith "fold_right1"
- | [e] -> e
- | e::l -> f e (fold_right1 f l)
-
-let rec for_all2_ p l1 l2 =
- match (l1, l2) with
- ([], []) -> true
- | (a1::l1, a2::l2) -> p a1 a2 && for_all2_ p l1 l2
- | (_, _) -> false
-
-let rec for_all2_true p l1 l2 =
- match (l1, l2) with
- | (a1::l1, a2::l2) -> p a1 a2 && for_all2_true p l1 l2
- | (_, _) -> true
-
-let maxl l = fold_right1 max l
-
-let rec stack2list s =
- let l = ref [] in
- Stack.iter (fun e -> l := e :: !l) s ;
- !l
-
-let rec stack_exists f s =
- try
- Stack.iter (fun e -> if f e then raise Found) s ;
- false
- with Found -> true
-
-let rec queue2list q = rev (Queue.fold (fun b a -> a :: b) [] q)
-
-let rec fix_point f p =
- let p' = f p in
- if p = p' then p else fix_point f p'
-
-let rec fix_point_withenv f env p =
- let p', env' = f env p in
- if p = p' then (p, env') else fix_point_withenv f env' p'
-
-let rec fix_point_ nb f p =
- let p' = f p in
- if p = p' then p, nb else fix_point_ (nb+1) f p'
-
-let rec group_by_2 = function
- | [] -> []
- | a :: b :: l -> (a, b) :: group_by_2 l
- | _ -> failwith "group_by_2"
-
-(*
-let rec lfix_point f e =
- let e' = f(e) in
- if e = e' then e :: lfix_point f e' else [e]
-*)
-
-let fluid_let ref value f =
- let previous_val = !ref in
- ref := value ;
- let v = f() in
- ref := previous_val ;
- v
-
-let do0_withenv doit f env l =
- let r_env = ref env in
- doit (fun e -> r_env := f !r_env e) l ;
- !r_env
-
-let do0_withenv2 doit f env l =
- let r_env = ref env in
- doit (fun e e' -> r_env := f !r_env e e') l ;
- !r_env
-
-let do_withenv doit f env l =
- let r_env = ref env in
- let l' = doit (fun e ->
- let e', env' = f !r_env e in
- r_env := env' ; e'
- ) l in
- l', !r_env
-
-let do2_withenv doit f env l1 l2 =
- let r_env = ref env in
- let l' = doit (fun e1 e2 ->
- let e', env' = f !r_env e1 e2 in
- r_env := env' ; e'
- ) l1 l2 in
- l', !r_env
-
-let do_collect doit f l1 =
- let l = ref [] in
- doit (fun i t -> l := f i t @ !l) l1 ;
- !l
-
-let map_withitself f l =
- let rec map_withitself_ done_ = function
- | [] -> done_
- | e :: l ->
- let e' = f (done_ @ e :: l) e in
- map_withitself_ (done_ @ [ e' ]) l
- in map_withitself_ [] l
-
-let map_t2 f (x,y) = f x, f y
-let map_t3 f (x,y,z) = f x, f y, f z
-let map_option f = function
- | Some e -> Some (f e)
- | None -> None
-let map_optionoption f = function
- | Some e -> f e
- | None -> None
-let t2_option2option_t2 = function
- | (Some x, Some y) -> Some(x,y)
- | _ -> None
-let rec l_option2option_l = function
- | [] -> Some []
- | None :: _l -> None
- | Some e :: l -> map_option (fun l -> e :: l) (l_option2option_l l)
-let map_option_env f (e, env) = map_option f e, env
-
-let t2_to_list (a,b) = [ a ; b ]
-let t3_to_list (a,b,c) = [ a ; b ; c ]
-
-let if_some bool val_ = if bool then Some val_ else None
-
-let rec fold_left_option f val_ = function
- | [] -> Some val_
- | e::l ->
- match f val_ e with
- | None -> None
- | Some val_' -> fold_left_option f val_' l
-
-let collect_some_withenv f env l =
- let rec collect accu env = function
- | [] -> rev accu, env
- | e::l ->
- let e', env' = f env e in
- let accu' =
- match e' with
- | Some e' -> e'::accu
- | None -> accu in
- collect accu' env' l
- in collect [] env l
-
-let for_all_option_withenv remap f env l =
- let rec for_all env accu = function
- | [] -> Some(remap (rev accu)), env
- | e::l ->
- (match f env e with
- | None, env' -> None, env'
- | Some e', env' -> for_all env' (e' :: accu) l)
- in
- for_all env [] l
-
-let for_all2_option_withenv remap f env la lb =
- let rec for_all env accu = function
- | [], [] -> Some(remap (rev accu)), env
- | a::la, b::lb ->
- (match f env a b with
- | None, env' -> None, env'
- | Some ab, env' -> for_all env' (ab :: accu) (la, lb))
- | _ -> None, env
- in
- for_all env [] (la, lb)
-
-let map_or_option f = function
- | Or_some e -> Or_some (f e)
- | Or_error err -> Or_error err
-
-let map_index f l =
- let rec map_ n = function
- | [] -> []
- | e::l -> f e n :: map_ (n+1) l
- in map_ 0 l
-
-let filter_index f l =
- let rec filter_ n = function
- | [] -> []
- | e::l ->
- let l' = filter_ (n+1) l in
- if f e n then e :: l' else l'
- in filter_ 0 l
-
-let iter_index f l =
- let rec iter_ n = function
- | [] -> ()
- | e::l -> f e n ; iter_ (n+1) l
- in iter_ 0 l
-
-let map_fst f (x, y) = f x, y
-let map_snd f (x, y) = x, f y
-
-let map_withenv f env e = do_withenv map f env e
-let find_withenv f env e = do_withenv find f env e
-let filter_withenv f env e = do_withenv filter f env e
-let exists_withenv f env e = do_withenv exists f env e
-let map_t2_withenv f env e = do_withenv map_t2 f env e
-let for_all_withenv f env e = do_withenv for_all f env e
-let collect_withenv f env e = do_withenv collect f env e
-let partition_either_withenv f env e = do_withenv partition_either f env e
-
-let map2_withenv f env l1 l2 = do2_withenv map2 f env l1 l2
-let for_all2_withenv f env l1 l2 = do2_withenv for_all2 f env l1 l2
-
-let rec take n l =
- if n = 0 then []
- else match l with
- | [] -> raise Not_found
- | e::l -> e :: take (n-1) l
-let last_n n l = rev (take n (rev l))
-let last l = hd (last_n 1 l)
-
-let rec skipfirst e = function
- | [] -> []
- | e'::l when e = e' -> skipfirst e l
- | l -> l
-
-let rec removelast = function
- | [] -> failwith "removelast"
- | [_] -> []
- | e::l -> e :: removelast l
-
-let rec split_last l =
- let rec spl accu = function
- | [] -> failwith "split_last"
- | [e] -> rev accu, e
- | e::l -> spl (e :: accu) l
- in spl [] l
-
-let iter_assoc_val f l = iter (fun (_,v) -> f v) l
-let map_assoc_val f l = map (fun (k,v) -> k, f v) l
-
-let assoc_or_fail e l =
- try assoc e l with Not_found -> failwith "assoc failed"
-
-let assoc_by is_same e l =
- find_some (fun (a,b) -> if is_same e a then Some b else None) l
-
-let rec update_assoc_by is_same f e = function
- | [] -> raise Not_found
- | (a,b) :: l when is_same e a -> (a, f b) :: l
- | (a,b) :: l -> (a,b) :: update_assoc_by is_same f e l
-
-let update_assoc f e = update_assoc_by (=) f e
-
-let rec update_assoc_by_with_default default is_same f e = function
- | [] -> [ e, f default ]
- | (a,b) :: l when is_same e a -> (a, f b) :: l
- | (a,b) :: l -> (a,b) :: update_assoc_by_with_default default is_same f e l
-
-let update_all_assoc_by is_same f e l =
- map (fun (a,b) -> a, if is_same e a then f b else b) l
-
-let rec rassoc e = function
- | [] -> raise Not_found
- | (k,v) :: l -> if e = v then k else rassoc e l
-
-let rec all_assoc e = function
- | [] -> []
- | (e',v) :: l when e=e' -> v :: all_assoc e l
- | _ :: l -> all_assoc e l
-
-let rec all_assoc_by is_same e = function
- | [] -> []
- | (e',v) :: l when is_same e e' -> v :: all_assoc_by is_same e l
- | _ :: l -> all_assoc_by is_same e l
-
-let prepare_want_all_assoc l =
- map (fun n -> n, uniq (all_assoc n l)) (uniq (map fst l))
-
-let prepare_want_all_assoc_by is_same l =
- map (fun n -> n, uniq_ is_same (all_assoc_by is_same n l)) (uniq_ is_same (map fst l))
-
-let prepare_want_all_assoc_by_ is_same_a is_same_b l =
- map (fun n -> n, uniq_ is_same_b (all_assoc_by is_same_a n l)) (uniq_ is_same_a (map fst l))
-
-let rec count_uniq = function
- | [] -> []
- | e::l ->
- let has, l' = partition ((=) e) l in
- (e, length has + 1) :: count_uniq l'
-
-let rec repeat e = function
- | 0 -> []
- | n -> e :: repeat e (n-1)
-
-let rec inits = function
- | [] -> [[]]
- | e::l -> [] :: map (fun l -> e::l) (inits l)
-let rec tails = function
- | [] -> [[]]
- | (_::xs) as xxs -> xxs :: tails xs
-
-let apply f x = f x;;
-
-let rec map3 f l1 l2 l3 =
- match (l1, l2, l3) with
- ([], [], []) -> []
- | (a1::l1, a2::l2, a3::l3) -> let r = f a1 a2 a3 in r :: map3 f l1 l2 l3
- | (_, _, _) -> invalid_arg "map3"
-
-let filter2 f l1 l2 =
- split (filter f (combine l1 l2))
-
-let break_at f l =
- let rec b l1 = function
- | [] -> l1, []
- | e::l2 -> if f e then (l1, e :: l2) else b (l1 @ [e]) l2
- in b [] l
-let break v l = break_at ((=) v) l
-
-let drop_while f l = snd (break_at (fun e -> not (f e)) l)
-
-(* break_at_indice 0 [1;2] gives [], [1;2]
- break_at_indice 1 [1;2] gives [1], [2]
- *)
-let rec break_at_indice i l =
- if i = 0 then [], l else
- match l with
- | [] -> raise Not_found
- | e::l2 ->
- let a, b = break_at_indice (i-1) l2 in
- e::a, b
-
-let rev_nth e l =
- let rec rev_nth' i = function
- | [] -> raise Not_found
- | e'::_ when e'=e -> i
- | _::l -> rev_nth' (i+1) l
- in rev_nth' 0 l
-
-let rec getset_nth l i f =
- match l, i with
- | e::l', 0 -> f e :: l'
- | [], _ -> failwith "getset_nth"
- | e::l', _ -> e :: getset_nth l' (i - 1) f
-
-let set_nth l i v = getset_nth l i (fun _ -> v)
-
-let adjustModDown m n = n - (n mod m)
-let adjustModUp m n = adjustModDown m (n + m - 1)
-
-
-let hashtbl_find f h =
- let r = ref None in
- Hashtbl.iter (fun v c -> if f v c then r := Some v) h ;
- match !r with
- | Some v -> v
- | None -> raise Not_found
-
-let hashtbl_map f h = Hashtbl.iter (fun v c -> Hashtbl.replace h v (f v c)) h
-
-let hashtbl_values h = Hashtbl.fold (fun _ v l -> v :: l) h []
-let hashtbl_keys h = Hashtbl.fold (fun k _ l -> k :: l) h []
-let hashtbl_to_list h = Hashtbl.fold (fun k v l -> (k,v) :: l) h []
-
-let hashtbl_collect f h =
- rev (Hashtbl.fold (fun k v l -> rev_append (f k v) l) h [])
-
-let hashtbl_exists f h =
- try
- Hashtbl.iter (fun v c -> if f v c then raise Found) h ;
- false
- with Found -> true
-
-let memoize f =
- let hash = Hashtbl.create 16 in
- fun k ->
- try Hashtbl.find hash k
- with Not_found ->
- let v = f k in
- Hashtbl.add hash k v ; v
-
-let array_shift a = Array.sub a 1 (Array.length a - 1)
-let array_last_n n a =
- let len = Array.length a in
- Array.sub a (len - n) n
-
-let array_collect f a = Array.fold_left (fun l e -> f e @ l) [] a
-
-let rec lvector_product =
- let rec vector_product a b = match a with
- | [] -> []
- | e::l -> map (fun e' -> e :: e') b :: vector_product l b
- in function
- | [] -> []
- | [e] -> map (fun e -> [e]) e
- | e::l -> flatten (vector_product e (lvector_product l))
-
-let vector_product2 a b =
- map (function
- | [a;b] -> a,b
- | _ -> failwith "vector_product2"
- ) (lvector_product [ a ; b ])
-
-let rec transpose = function
- | [] :: _ -> []
- | ll ->
- let l, ll' = split (map (function e::l -> e,l | _ -> raise Not_found) ll) in
- l :: transpose ll'
-
-let rec range min max =
- if min >= max then [] else min :: range (min + 1) max
-
-let sum l = List.fold_left (+) 0 l
-
-let rec filter_some_with f = function
- | [] -> []
- | e :: l ->
- match f e with
- | None -> filter_some_with f l
- | Some e' -> e' :: filter_some_with f l
-
-let rec filter_some = function
- | [] -> []
- | None :: l -> filter_some l
- | Some e :: l -> e :: filter_some l
-
-let rec difference l = function
- | [] -> l
- | e::l' -> difference (filter ((<>) e) l) l'
-
-let rec difference_ eq l = function
- | [] -> l
- | e::l' ->
- let l2 = filter (fun e' -> not (eq e e')) l in
- difference_ eq l2 l'
-
-let intersection_by is_same l1 l2 = filter (fun e -> exists (is_same e) l2) l1
-
-let intersection_and_differences eq l1 l2 =
- let rec both inter l2_only = function
- | [], l2 -> inter, [], rev l2_only @ l2
- | l1, [] -> inter, l1, rev l2_only
- | l1, e2 :: l2' ->
- match partition (eq e2) l1 with
- | [], _ -> both inter (e2 :: l2_only) (l1, l2')
- | _, l1' -> both (e2 :: inter) l2_only (l1', l2')
- in both [] [] (l1, l2)
-
-let rec triangularize = function
- | [] -> []
- | e::l -> (e,l) :: triangularize l
-
-let diagonalize l =
- map_index (fun a i ->
- a, filter_index (fun _ j -> i <> j) l
- ) l
-
-let rec list_of_nonempty_sublists = function
- | [] -> []
- | e :: l ->
- let l' = list_of_nonempty_sublists l in
- [e] :: l' @ map (fun l -> e :: l) l'
-
-let rec graph_is_sorted_by eq = function
- | [] -> true
- | (_,deps) :: l ->
- for_all (fun e -> try let _ = assoc_by eq e l in false with Not_found -> true) deps && graph_is_sorted_by eq l
-
-let graph_closure_by eq graph =
- let err = ref None in
- try
- let graph_rev = collect (fun (i, l) -> map (fun e -> (e, i)) l) graph in
- let bothway = map (fun (i,l) -> i, (l, all_assoc_by eq i graph_rev)) graph in
- let closed = fold_left (fun graph j ->
- let next, prev = assoc_by eq j graph in
- let graph2 = fold_left (fun graph i ->
- if member_ eq i next then (err := Some(j,i); raise GraphSort_circular_deps) else
- update_assoc_by eq (fun (i_next,i_prev) -> i_next @ next, i_prev) i graph
- ) graph (filter (fun a -> not (eq a j)) prev) in
- let graph3 = fold_left (fun graph k ->
- if member_ eq k prev then (err := Some(j,k); raise GraphSort_circular_deps) else
- update_assoc_by eq (fun (k_next,k_prev) -> k_next, k_prev @ prev) k graph
- ) graph2 (filter (fun a -> not (eq a j)) next) in
- graph3
- ) bothway (map fst bothway) in
- Or_some (map (fun (e,(next,_)) -> e, uniq_ eq next) closed)
- with GraphSort_circular_deps ->
- Or_error (some !err)
-
-let rec graph_sort_by eq l =
- let cmp (_, deps_a) (b, _) = if member_ eq b deps_a then 1 else -1 in
- let rec sort_it = function
- | [] -> []
- | [e] -> [e]
- | e::l ->
- let l' = sort_it l in
- let gt, lt = break_at (fun ((_, deps) as e') -> deps = [] or cmp e e' = 1) l' in
- gt @ [e] @ lt
- in
- map_or_option (fun l' ->
- let l_sorted = rev (sort_it l') in
- if not (graph_is_sorted_by eq l_sorted) then internal_error "graph_sort failed" else
- l_sorted
- ) (graph_closure_by eq l)
-
-let int_sort l = sort (fun a b -> a - b) l
-
-let str_begins_with prefix s =
- String.sub s 0 (min (String.length s) (String.length prefix)) = prefix
-
-let rec strstr s subs =
- let len_s, len_subs = String.length s, String.length subs in
- let rec rec_ i =
- let i' = String.index_from s i subs.[0] in
- if i' + len_subs <= len_s then
- if String.sub s i' len_subs = subs then
- i'
- else
- rec_ (i' + 1)
- else
- raise Not_found
- in
- rec_ 0
-
-let str_contains s subs =
- try
- let _ = strstr s subs in true
- with Not_found -> false
-
-let str_ends_with s suffix =
- let len = min (String.length s) (String.length suffix) in
- String.sub s (String.length s - len) len = suffix
-
-let chop = function
- | "" -> ""
- | s -> String.sub s 0 (String.length s - 1)
-
-let chomps s =
- let i = ref (String.length s - 1) in
- while !i >= 0 && (s.[!i] = ' ' || s.[!i] = '\t') do decr i done ;
- String.sub s 0 (!i+1)
-
-let rec times e = function
- | 0 -> []
- | n -> e :: times e (n-1)
-
-let skip_n_char_ beg end_ s =
- let full_len = String.length s in
- if beg < full_len && full_len - beg - end_ > 0
- then String.sub s beg (full_len - beg - end_)
- else ""
-let skip_n_char n s = skip_n_char_ n 0 s
-
-let rec non_index_from s beg c =
- if s.[beg] = c then non_index_from s (beg+1) c else beg
-let non_index s c = non_index_from s 0 c
-
-let rec non_rindex_from s beg c =
- if s.[beg] = c then non_rindex_from s (beg-1) c else beg
-let non_rindex s c = non_rindex_from s (String.length s - 1) c
-
-let rec explode_string = function
- | "" -> []
- | s -> (String.get s 0) :: explode_string (String.sub s 1 (String.length s - 1))
-
-let count_matching_char s c =
- let rec count_matching_char_ nb i =
- try
- let i' = String.index_from s i c in
- count_matching_char_ (nb+1) (i'+1)
- with Not_found -> nb
- in
- count_matching_char_ 0 0
-
-let is_uppercase c = Char.lowercase c <> c
-let is_lowercase c = Char.uppercase c <> c
-
-let char_is_alphanumerical c =
- let i = Char.code c in
- Char.code 'a' <= i && i <= Char.code 'z' ||
- Char.code 'A' <= i && i <= Char.code 'Z' ||
- Char.code '0' <= i && i <= Char.code '9'
-
-let char_is_alphanumerical_ c =
- let i = Char.code c in
- Char.code 'a' <= i && i <= Char.code 'z' ||
- Char.code 'A' <= i && i <= Char.code 'Z' ||
- Char.code '0' <= i && i <= Char.code '9' || c = '_'
-
-let char_is_alpha c =
- let i = Char.code c in
- Char.code 'a' <= i && i <= Char.code 'z' ||
- Char.code 'A' <= i && i <= Char.code 'Z'
-
-let char_is_number c =
- let i = Char.code c in
- Char.code '0' <= i && i <= Char.code '9'
-
-let count_chars_in_string s c =
- let rec rec_count_chars_in_string from =
- try
- let from' = String.index_from s from c in
- 1 + rec_count_chars_in_string (from' + 1)
- with
- Not_found -> 0
- in rec_count_chars_in_string 0
-
-let rec string_fold_left f val_ s =
- let val_ = ref val_ in
- for i = 0 to String.length s - 1 do
- val_ := f !val_ s.[i]
- done ;
- !val_
-
-(*
-let rec string_forall_with f i s =
- try
- f s.[i] && string_forall_with f (i+1) s
- with Invalid_argument _ -> true
-*)
-let string_forall_with f i s =
- let len = String.length s in
- let rec string_forall_with_ i =
- i >= len || f s.[i] && string_forall_with_ (i+1)
- in string_forall_with_ i
-
-let starts_with_non_lowercase s = s <> "" && s.[0] <> '_' && not (is_lowercase s.[0])
-
-let rec fold_lines f init chan =
- try
- let line = input_line chan in
- fold_lines f (f init line) chan
- with End_of_file -> init
-let readlines chan = List.rev (fold_lines (fun l e -> e::l) [] chan)
-
-let split_at c s =
- let rec split_at_ accu i =
- try
- let i' = String.index_from s i c in
- split_at_ (String.sub s i (i' - i) :: accu) (i'+1)
- with Not_found -> rev (skip_n_char i s :: accu)
- in
- split_at_ [] 0
-
-let split_at2 c1 c2 s =
- let rec split_at2_ accu i i2 =
- try
- let i3 = String.index_from s i2 c1 in
- if s.[i3+1] = c2 then split_at2_ (String.sub s i (i3 - i) :: accu) (i3+2) (i3+2) else
- split_at2_ accu i i3
- with Not_found | Invalid_argument _ -> rev (skip_n_char i s :: accu)
- in
- split_at2_ [] 0 0
-
-let words s =
- let rec words_ accu i s =
- try
- let i2 = non_index_from s i ' ' in
- try
- let i3 = String.index_from s i2 ' ' in
- words_ (String.sub s i2 (i3 - i2) :: accu) (i3+1) s
- with Not_found -> rev (skip_n_char i2 s :: accu)
- with Invalid_argument _ -> rev accu
- in
- collect (words_ [] 0) (split_at '\n' s)
-
-let to_CamelCase s_ =
- let l = ref [] in
- let s = String.copy s_ in
- for i = 1 to String.length s - 1 do
- if is_uppercase (String.unsafe_get s i) && is_lowercase (String.unsafe_get s (i-1)) then (
- String.set s i (Char.lowercase (String.get s i)) ;
- l := i :: !l
- )
- done ;
- if !l = [] then None else
- let offset, s' = fold_left (fun (offset, s') i ->
- i, s' ^ String.sub s offset (i-offset) ^ "_"
- ) (0, "") (rev !l) in
- Some (s' ^ String.sub s offset (String.length s - offset))
-
-let concat_symlink file link =
- if str_begins_with "..//" link then (* ..//foo => /foo *)
- skip_n_char 3 link
- else
- let file = if str_ends_with file "/" then chop file else file in (* s|/$|| *)
- let rec reduce file link =
- if str_begins_with "../" link then
- let file = String.sub file 0 (String.rindex file '/') in (* s|/[^/]+$|| *)
- reduce file (skip_n_char 3 link)
- else
- file ^ "/" ^ link
- in
- reduce file link
-
-let expand_symlinks file =
- match split_at '/' file with
- | "" :: l ->
- let rec remove_dotdot accu nb = function
- | [] -> if nb = 0 then accu else failwith "remove_dotdot"
- | ".." :: l -> remove_dotdot accu (nb + 1) l
- | e :: l -> if nb > 0 then remove_dotdot accu (nb - 1) l else remove_dotdot (e :: accu) nb l
- in
- let l = remove_dotdot [] 0 (List.rev l) in
- List.fold_left (fun file piece ->
- fix_point (fun file ->
- try concat_symlink file ("../" ^ Unix.readlink file)
- with _ -> file
- ) (file ^ "/" ^ piece)) "" l
- | _ -> internal_error (Printf.sprintf "expand_symlinks: %s is relative\n" file)
-
-let mtime f = (Unix.stat f).Unix.st_mtime
-
-let rec updir dir nb =
- if nb = 0 then dir else
- match dir with
- | "." -> String.concat "/" (times ".." nb)
- | _ ->
- if Filename.basename dir = ".." then
- dir ^ "/" ^ String.concat "/" (times ".." nb)
- else
- updir (Filename.dirname dir) (nb-1)
-
-let (string_of_ref : 'a ref -> string) = fun r ->
- Printf.sprintf "0x%x" (Obj.magic r : int)
-
-let print_endline_flush s = print_endline s ; flush stdout
-
-let is_int n = n = floor n
-
-(* total order *)
-let rec compare_lists cmp l1 l2 =
- match l1, l2 with
- | [], [] -> 0
- | [], _ -> -1
- | _, [] -> 1
- | e1::l1, e2::l2 ->
- match cmp e1 e2 with
- | 0 -> compare_lists cmp l1 l2
- | v -> v
-
-let compare_best a b =
- match a, b with
- | 0, 0 -> 0
- | 1, 1 | 1, 0 | 0, 1 -> 1
- | -1, -1 | -1, 0 | 0, -1 -> -1
- | 1, -1 | -1, 1 -> raise Not_comparable
- | _ -> failwith "uh?"
-
-(* partial order *)
-let combine_comparison_list l =
- fold_left compare_best 0 l
-
-let min_with_cmp less_than a b =
- if less_than a b then a
- else if less_than b a then b
- else raise Not_comparable
-
-let max_with_cmp less_than a b =
- if less_than a b then b
- else if less_than b a then a
- else raise Not_comparable
-
-let rec fold_left2_compare f e l1 l2 =
- match l1, l2 with
- | [], [] -> e
- | e1::l1, e2::l2 -> fold_left2_compare f (f e e1 e2) l1 l2
- | _ -> raise Not_comparable
-
-let rec exists_compare cmp = function
- | [] -> raise Not_comparable
- | e :: l -> try cmp e with Not_comparable -> exists_compare cmp l
-
-let forall_compare cmp = fold_left (fun n e -> compare_best n (cmp e)) 0
-let forall2_compare cmp = fold_left2_compare (fun n e1 e2 -> compare_best n (cmp e1 e2)) 0
-
-let exists2_compare left_dropping cmp l1 l2 =
- let rec forall_compare_ n = function
- | [], [] -> n
- | _, [] -> compare_best n left_dropping
- | [], _ -> compare_best n (-left_dropping)
- | e1::l1, e2::l2 ->
- match try Some (cmp e1 e2) with Not_comparable -> None with
- | Some n' -> forall_compare_ (compare_best n n') (l1, l2)
- | None ->
- if n = left_dropping then
- forall_compare_ left_dropping (l1, e2::l2)
- else if n = -left_dropping then
- forall_compare_ (-left_dropping) (e1::l1, l2)
- else
- (* need to try both *)
- try forall_compare_ left_dropping (l1, e2::l2)
- with Not_comparable -> forall_compare_ (-left_dropping) (e1::l1, l2)
- in forall_compare_ 0 (l1, l2)
-
-
-let rec compare_sorted_sets is_same l1 l2 =
- match l1, l2 with
- | [], [] -> 0
- | [], _ -> -1
- | _, [] -> 1
- | e1::l1, e2::l2 -> if is_same e1 e2 then compare_sorted_sets is_same l1 l2 else raise Not_found
-
-let scan_list_while_modifying f l =
- let rec scan_list_while_modifying_ prev = function
- | [] -> prev
- | e :: next ->
- let prev', next' = some_or (f prev next e) (prev @ [e], next) in
- scan_list_while_modifying_ prev' next'
- in scan_list_while_modifying_ [] l
-
-let bools2compare = function
- | true, true -> 0
- | true, false -> -1
- | false, true -> 1
- | _ -> raise Not_comparable
-
-let lpush l e = l := e :: !l
-
-(*
-let is_greater2compare is_greater a b =
- match is_greater a b, is_greater b a with
-
- *)
-
-module OrderedString =
- struct
- type t = string
- let compare = compare
- end;;
-
-module StringSet = Set.Make(OrderedString);;
-
-let stringSet_to_list = StringSet.elements
-let stringSet_add set e = StringSet.add e set
-let stringSet_difference = StringSet.diff
-let list_to_StringSet l = fold_left stringSet_add StringSet.empty l
-
-(* this character messes emacs caml mode *)
-let char_quote = '"'
diff --git a/perl_checker.src/common.mli b/perl_checker.src/common.mli
deleted file mode 100644
index 86a13cd..0000000
--- a/perl_checker.src/common.mli
+++ /dev/null
@@ -1,276 +0,0 @@
-exception Found
-exception Not_comparable
-exception GraphSort_circular_deps
-type ('a, 'b) either = Left of 'a | Right of 'b
-type ('a, 'b) or_option = Or_some of 'a | Or_error of 'b
-val internal_error : string -> 'a
-val id : 'a -> 'a
-val double : 'a -> 'a * 'a
-val swap : 'a * 'b -> 'b * 'a
-val safe_tl : 'a list -> 'a list
-val fstfst : ('a * 'b) * 'c -> 'a
-val sndfst : ('a * 'b) * 'c -> 'b
-val fstsnd : 'a * ('b * 'c) -> 'b
-val sndsnd : 'a * ('b * 'c) -> 'c
-val fst3 : 'a * 'b * 'c -> 'a
-val snd3 : 'a * 'b * 'c -> 'b
-val ter3 : 'a * 'b * 'c -> 'c
-val sndter3 : 'a * 'b * 'c -> 'b * 'c
-val o : ('a -> 'b) -> ('c -> 'a) -> 'c -> 'b
-val curry : ('a * 'b -> 'c) -> 'a -> 'b -> 'c
-val uncurry : ('a -> 'b -> 'c) -> 'a * 'b -> 'c
-val uncons : 'a list -> 'a * 'a list
-val has_env : string -> bool
-val some : 'a option -> 'a
-val some_or : 'a option -> 'a -> 'a
-val option2l : 'a option -> 'a list
-val prefer_some : ('a -> 'a -> 'a) -> 'a option -> 'a option -> 'a option
-val collect_accu : ('a -> 'b list) -> 'b list -> 'a list -> 'b list
-val collect : ('a -> 'b list) -> 'a list -> 'b list
-val merge_some : ('a -> 'a -> 'a) -> 'a option -> 'a option -> 'a option
-val uniq : 'a list -> 'a list
-val uniq_ : ('a -> 'a -> bool) -> 'a list -> 'a list
-val non_uniq : 'a list -> 'a list
-val member_ : ('a -> 'b -> bool) -> 'a -> 'b list -> bool
-val find_some : ('a -> 'b option) -> 'a list -> 'b
-val fold_left1 : ('a -> 'a -> 'a) -> 'a list -> 'a
-val find_index : 'a -> 'a list -> int
-val find_some_ : ('a -> 'b option) -> 'a list -> 'b option
-val fpartition : ('a -> 'b option) -> 'a list -> 'b list * 'a list
-val partition_either :
- ('a -> ('b, 'c) either) -> 'a list -> 'b list * 'c list
-val keep_best : ('a * 'a -> 'a option) -> 'a list -> 'a list
-val keep_bests : ('a * 'a -> 'a option) -> 'a list -> 'a list
-val fold_right1 : ('a -> 'a -> 'a) -> 'a list -> 'a
-val for_all2_ : ('a -> 'b -> bool) -> 'a list -> 'b list -> bool
-val for_all2_true : ('a -> 'b -> bool) -> 'a list -> 'b list -> bool
-val maxl : 'a list -> 'a
-val stack2list : 'a Stack.t -> 'a list
-val stack_exists : ('a -> bool) -> 'a Stack.t -> bool
-val queue2list : 'a Queue.t -> 'a list
-val fix_point : ('a -> 'a) -> 'a -> 'a
-val fix_point_withenv : ('a -> 'b -> 'b * 'a) -> 'a -> 'b -> 'b * 'a
-val fix_point_ : int -> ('a -> 'a) -> 'a -> 'a * int
-val group_by_2 : 'a list -> ('a * 'a) list
-val fluid_let : 'a ref -> 'a -> (unit -> 'b) -> 'b
-val do0_withenv :
- (('a -> unit) -> 'b -> 'c) -> ('d -> 'a -> 'd) -> 'd -> 'b -> 'd
-val do0_withenv2 :
- (('a -> 'b -> unit) -> 'c -> 'd) ->
- ('e -> 'a -> 'b -> 'e) -> 'e -> 'c -> 'e
-val do_withenv :
- (('a -> 'b) -> 'c -> 'd) -> ('e -> 'a -> 'b * 'e) -> 'e -> 'c -> 'd * 'e
-val do2_withenv :
- (('a -> 'b -> 'c) -> 'd -> 'e -> 'f) ->
- ('g -> 'a -> 'b -> 'c * 'g) -> 'g -> 'd -> 'e -> 'f * 'g
-val do_collect :
- (('a -> 'b -> unit) -> 'c -> 'd) -> ('a -> 'b -> 'e list) -> 'c -> 'e list
-val map_withitself : ('a list -> 'a -> 'a) -> 'a list -> 'a list
-val map_t2 : ('a -> 'b) -> 'a * 'a -> 'b * 'b
-val map_t3 : ('a -> 'b) -> 'a * 'a * 'a -> 'b * 'b * 'b
-val map_option : ('a -> 'b) -> 'a option -> 'b option
-val map_optionoption : ('a -> 'b option) -> 'a option -> 'b option
-val t2_option2option_t2 : 'a option * 'b option -> ('a * 'b) option
-val l_option2option_l : 'a option list -> 'a list option
-val map_option_env : ('a -> 'b) -> 'a option * 'c -> 'b option * 'c
-val t2_to_list : 'a * 'a -> 'a list
-val t3_to_list : 'a * 'a * 'a -> 'a list
-val if_some : bool -> 'a -> 'a option
-val fold_left_option : ('a -> 'b -> 'a option) -> 'a -> 'b list -> 'a option
-val collect_some_withenv :
- ('a -> 'b -> 'c option * 'a) -> 'a -> 'b list -> 'c list * 'a
-val for_all_option_withenv :
- ('a list -> 'b) ->
- ('c -> 'd -> 'a option * 'c) -> 'c -> 'd list -> 'b option * 'c
-val for_all2_option_withenv :
- ('a list -> 'b) ->
- ('c -> 'd -> 'e -> 'a option * 'c) ->
- 'c -> 'd list -> 'e list -> 'b option * 'c
-val map_or_option : ('a -> 'b) -> ('a, 'c) or_option -> ('b, 'c) or_option
-val map_index : ('a -> int -> 'b) -> 'a list -> 'b list
-val filter_index : ('a -> int -> bool) -> 'a list -> 'a list
-val iter_index : ('a -> int -> 'b) -> 'a list -> unit
-val map_fst : ('a -> 'b) -> 'a * 'c -> 'b * 'c
-val map_snd : ('a -> 'b) -> 'c * 'a -> 'c * 'b
-val map_withenv : ('a -> 'b -> 'c * 'a) -> 'a -> 'b list -> 'c list * 'a
-val find_withenv : ('a -> 'b -> bool * 'a) -> 'a -> 'b list -> 'b * 'a
-val filter_withenv : ('a -> 'b -> bool * 'a) -> 'a -> 'b list -> 'b list * 'a
-val exists_withenv : ('a -> 'b -> bool * 'a) -> 'a -> 'b list -> bool * 'a
-val map_t2_withenv : ('a -> 'b -> 'c * 'a) -> 'a -> 'b * 'b -> ('c * 'c) * 'a
-val for_all_withenv : ('a -> 'b -> bool * 'a) -> 'a -> 'b list -> bool * 'a
-val collect_withenv :
- ('a -> 'b -> 'c list * 'a) -> 'a -> 'b list -> 'c list * 'a
-val partition_either_withenv :
- ('a -> 'b -> ('c, 'd) either * 'a) ->
- 'a -> 'b list -> ('c list * 'd list) * 'a
-val map2_withenv :
- ('a -> 'b -> 'c -> 'd * 'a) -> 'a -> 'b list -> 'c list -> 'd list * 'a
-val for_all2_withenv :
- ('a -> 'b -> 'c -> bool * 'a) -> 'a -> 'b list -> 'c list -> bool * 'a
-val take : int -> 'a list -> 'a list
-val last_n : int -> 'a list -> 'a list
-val last : 'a list -> 'a
-val skipfirst : 'a -> 'a list -> 'a list
-val removelast : 'a list -> 'a list
-val split_last : 'a list -> 'a list * 'a
-val iter_assoc_val : ('a -> unit) -> ('b * 'a) list -> unit
-val map_assoc_val : ('a -> 'b) -> ('c * 'a) list -> ('c * 'b) list
-val assoc_or_fail : 'a -> ('a * 'b) list -> 'b
-val assoc_by : ('a -> 'b -> bool) -> 'a -> ('b * 'c) list -> 'c
-val update_assoc_by :
- ('a -> 'b -> bool) -> ('c -> 'c) -> 'a -> ('b * 'c) list -> ('b * 'c) list
-val update_assoc : ('a -> 'a) -> 'b -> ('b * 'a) list -> ('b * 'a) list
-val update_assoc_by_with_default :
- 'a ->
- ('b -> 'b -> bool) -> ('a -> 'a) -> 'b -> ('b * 'a) list -> ('b * 'a) list
-val update_all_assoc_by :
- ('a -> 'b -> bool) -> ('c -> 'c) -> 'a -> ('b * 'c) list -> ('b * 'c) list
-val rassoc : 'a -> ('b * 'a) list -> 'b
-val all_assoc : 'a -> ('a * 'b) list -> 'b list
-val all_assoc_by : ('a -> 'b -> bool) -> 'a -> ('b * 'c) list -> 'c list
-val prepare_want_all_assoc : ('a * 'b) list -> ('a * 'b list) list
-val prepare_want_all_assoc_by :
- ('a -> 'a -> bool) -> ('a * 'a) list -> ('a * 'a list) list
-val prepare_want_all_assoc_by_ :
- ('a -> 'a -> bool) ->
- ('b -> 'b -> bool) -> ('a * 'b) list -> ('a * 'b list) list
-val count_uniq : 'a list -> ('a * int) list
-val repeat : 'a -> int -> 'a list
-val inits : 'a list -> 'a list list
-val tails : 'a list -> 'a list list
-val apply : ('a -> 'b) -> 'a -> 'b
-val map3 : ('a -> 'b -> 'c -> 'd) -> 'a list -> 'b list -> 'c list -> 'd list
-val filter2 : ('a * 'b -> bool) -> 'a list -> 'b list -> 'a list * 'b list
-val break_at : ('a -> bool) -> 'a list -> 'a list * 'a list
-val break : 'a -> 'a list -> 'a list * 'a list
-val drop_while : ('a -> bool) -> 'a list -> 'a list
-val break_at_indice : int -> 'a list -> 'a list * 'a list
-val rev_nth : 'a -> 'a list -> int
-val getset_nth : 'a list -> int -> ('a -> 'a) -> 'a list
-val set_nth : 'a list -> int -> 'a -> 'a list
-val adjustModDown : int -> int -> int
-val adjustModUp : int -> int -> int
-val hashtbl_find : ('a -> 'b -> bool) -> ('a, 'b) Hashtbl.t -> 'a
-val hashtbl_map : ('a -> 'b -> 'b) -> ('a, 'b) Hashtbl.t -> unit
-val hashtbl_values : ('a, 'b) Hashtbl.t -> 'b list
-val hashtbl_keys : ('a, 'b) Hashtbl.t -> 'a list
-val hashtbl_to_list : ('a, 'b) Hashtbl.t -> ('a * 'b) list
-val hashtbl_collect : ('a -> 'b -> 'c list) -> ('a, 'b) Hashtbl.t -> 'c list
-val hashtbl_exists : ('a -> 'b -> bool) -> ('a, 'b) Hashtbl.t -> bool
-val memoize : ('a -> 'b) -> 'a -> 'b
-val array_shift : 'a array -> 'a array
-val array_last_n : int -> 'a array -> 'a array
-val array_collect : ('a -> 'b list) -> 'a array -> 'b list
-val lvector_product : 'a list list -> 'a list list
-val vector_product2 : 'a list -> 'a list -> ('a * 'a) list
-val transpose : 'a list list -> 'a list list
-val range : int -> int -> int list
-val sum : int list -> int
-val filter_some_with : ('a -> 'b option) -> 'a list -> 'b list
-val filter_some : 'a option list -> 'a list
-val difference : 'a list -> 'a list -> 'a list
-val difference_ : ('a -> 'b -> bool) -> 'b list -> 'a list -> 'b list
-val intersection_by : ('a -> 'b -> bool) -> 'a list -> 'b list -> 'a list
-val intersection_and_differences :
- ('a -> 'b -> bool) -> 'b list -> 'a list -> 'a list * 'b list * 'a list
-val triangularize : 'a list -> ('a * 'a list) list
-val diagonalize : 'a list -> ('a * 'a list) list
-val list_of_nonempty_sublists : 'a list -> 'a list list
-val graph_is_sorted_by : ('a -> 'b -> bool) -> ('b * 'a list) list -> bool
-val graph_closure_by :
- ('a -> 'a -> bool) ->
- ('a * 'a list) list -> (('a * 'a list) list, 'a * 'a) or_option
-val graph_sort_by :
- ('a -> 'a -> bool) ->
- ('a * 'a list) list -> (('a * 'a list) list, 'a * 'a) or_option
-val int_sort : int list -> int list
-val str_begins_with : string -> string -> bool
-val strstr : string -> string -> int
-val str_contains : string -> string -> bool
-val str_ends_with : string -> string -> bool
-val chop : string -> string
-val chomps : string -> string
-val times : 'a -> int -> 'a list
-val skip_n_char_ : int -> int -> string -> string
-val skip_n_char : int -> string -> string
-val non_index_from : string -> int -> char -> int
-val non_index : string -> char -> int
-val non_rindex_from : string -> int -> char -> int
-val non_rindex : string -> char -> int
-val explode_string : string -> char list
-val count_matching_char : string -> char -> int
-val is_uppercase : char -> bool
-val is_lowercase : char -> bool
-val char_is_alphanumerical : char -> bool
-val char_is_alphanumerical_ : char -> bool
-val char_is_alpha : char -> bool
-val char_is_number : char -> bool
-val count_chars_in_string : string -> char -> int
-val string_fold_left : ('a -> char -> 'a) -> 'a -> string -> 'a
-val string_forall_with : (char -> bool) -> int -> string -> bool
-val starts_with_non_lowercase : string -> bool
-val fold_lines : ('a -> string -> 'a) -> 'a -> in_channel -> 'a
-val readlines : in_channel -> string list
-val split_at : char -> string -> string list
-val split_at2 : char -> char -> string -> string list
-val words : string -> string list
-val to_CamelCase : string -> string option
-val concat_symlink : string -> string -> string
-val expand_symlinks : string -> string
-val mtime : string -> float
-val updir : string -> int -> string
-val string_of_ref : 'a ref -> string
-val print_endline_flush : string -> unit
-val is_int : float -> bool
-val compare_lists : ('a -> 'b -> int) -> 'a list -> 'b list -> int
-val compare_best : int -> int -> int
-val combine_comparison_list : int list -> int
-val min_with_cmp : ('a -> 'a -> bool) -> 'a -> 'a -> 'a
-val max_with_cmp : ('a -> 'a -> bool) -> 'a -> 'a -> 'a
-val fold_left2_compare :
- ('a -> 'b -> 'c -> 'a) -> 'a -> 'b list -> 'c list -> 'a
-val exists_compare : ('a -> 'b) -> 'a list -> 'b
-val forall_compare : ('a -> int) -> 'a list -> int
-val forall2_compare : ('a -> 'b -> int) -> 'a list -> 'b list -> int
-val exists2_compare : int -> ('a -> 'b -> int) -> 'a list -> 'b list -> int
-val compare_sorted_sets : ('a -> 'b -> bool) -> 'a list -> 'b list -> int
-val scan_list_while_modifying :
- ('a list -> 'a list -> 'a -> ('a list * 'a list) option) ->
- 'a list -> 'a list
-val bools2compare : bool * bool -> int
-val lpush : 'a list ref -> 'a -> unit
-module OrderedString : sig type t = string val compare : 'a -> 'a -> int end
-module StringSet :
- sig
- type elt = OrderedString.t
- type t = Set.Make(OrderedString).t
- val empty : t
- val is_empty : t -> bool
- val mem : elt -> t -> bool
- val add : elt -> t -> t
- val singleton : elt -> t
- val remove : elt -> t -> t
- val union : t -> t -> t
- val inter : t -> t -> t
- val diff : t -> t -> t
- val compare : t -> t -> int
- val equal : t -> t -> bool
- val subset : t -> t -> bool
- val iter : (elt -> unit) -> t -> unit
- val fold : (elt -> 'a -> 'a) -> t -> 'a -> 'a
- val for_all : (elt -> bool) -> t -> bool
- val exists : (elt -> bool) -> t -> bool
- val filter : (elt -> bool) -> t -> t
- val partition : (elt -> bool) -> t -> t * t
- val cardinal : t -> int
- val elements : t -> elt list
- val min_elt : t -> elt
- val max_elt : t -> elt
- val choose : t -> elt
- val split : elt -> t -> t * bool * t
- end
-val stringSet_to_list : StringSet.t -> StringSet.elt list
-val stringSet_add : StringSet.t -> StringSet.elt -> StringSet.t
-val stringSet_difference : StringSet.t -> StringSet.t -> StringSet.t
-val list_to_StringSet : StringSet.elt list -> StringSet.t
-val char_quote : char
diff --git a/perl_checker.src/config_file.ml b/perl_checker.src/config_file.ml
deleted file mode 100644
index a5ee94f..0000000
--- a/perl_checker.src/config_file.ml
+++ /dev/null
@@ -1,40 +0,0 @@
-open Common
-
-type config_file = {
- basedir : int option ;
- }
-
-let ignored_packages = ref []
-
-let default = { basedir = None }
-
-
-let config_cache = Hashtbl.create 16
-
-let read dir =
- try Hashtbl.find config_cache dir with Not_found ->
- try
- let file_name = dir ^ "/.perl_checker" in
- let fh = open_in file_name in
- let config =
- fold_lines (fun config line ->
- match words line with
- | [ "Basedir"; ".." ] -> { config with basedir = Some 1 }
- | [ "Basedir"; "../.." ] -> { config with basedir = Some 2 }
- | [] -> config (* blank line *)
- | [ "Ignore"; pkg ]
- | [ pkg ] (* the deprecated form *)
- -> lpush ignored_packages pkg; config
- | _ -> prerr_endline (Printf.sprintf "bad line \"%s\" in %s" line file_name); config
- ) default fh
- in
- Hashtbl.add config_cache dir config ;
- if !Flags.verbose then print_endline_flush ("reading config file " ^ file_name);
- config
- with Sys_error _ -> default
-
-
-let rec read_any dir depth =
- if depth = 0 then () else
- let _ = read dir in
- read_any (updir dir 1) (depth - 1)
diff --git a/perl_checker.src/config_file.mli b/perl_checker.src/config_file.mli
deleted file mode 100644
index d5ad2f2..0000000
--- a/perl_checker.src/config_file.mli
+++ /dev/null
@@ -1,6 +0,0 @@
-type config_file = { basedir : int option; }
-val ignored_packages : string list ref
-val default : config_file
-val config_cache : (string, config_file) Hashtbl.t
-val read : string -> config_file
-val read_any : string -> int -> unit
diff --git a/perl_checker.src/flags.ml b/perl_checker.src/flags.ml
deleted file mode 100644
index 187c140..0000000
--- a/perl_checker.src/flags.ml
+++ /dev/null
@@ -1,43 +0,0 @@
-open Common
-open Types
-
-let verbose = ref false
-let quiet = ref false
-let generate_pot = ref false
-let expand_tabs = ref (Some 8)
-let no_cache = ref false
-
-let check_unused_global_vars = ref false
-let check_white_space = ref true
-let check_suggest_simpler = ref true
-let check_void = ref true
-let check_context = ref true
-let check_strange = ref true
-let check_traps = ref true
-let check_complex_expressions = ref true
-let normalized_expressions = ref true
-let check_help_perl_checker = ref true
-let suggest_functional = ref true
-let check_prototypes = ref true
-let check_names = ref true
-let check_import_export = ref true
-let allow_MDK_Common = ref true
-
-let is_warning_type_set = function
- | Warn_white_space -> !check_white_space
- | Warn_suggest_simpler -> !check_suggest_simpler
- | Warn_unused_global_vars -> !check_unused_global_vars
- | Warn_void -> !check_void
- | Warn_context -> !check_context
- | Warn_strange -> !check_strange
- | Warn_traps -> !check_traps
- | Warn_complex_expressions -> !check_complex_expressions
- | Warn_normalized_expressions -> !normalized_expressions
- | Warn_suggest_functional -> !suggest_functional
- | Warn_prototypes -> !check_prototypes
- | Warn_names -> !check_names
- | Warn_import_export -> !check_import_export
- | Warn_MDK_Common -> !allow_MDK_Common
- | Warn_help_perl_checker -> !check_help_perl_checker
-
-let are_warning_types_set l = not !quiet && List.for_all is_warning_type_set l
diff --git a/perl_checker.src/flags.mli b/perl_checker.src/flags.mli
deleted file mode 100644
index 2dc3b26..0000000
--- a/perl_checker.src/flags.mli
+++ /dev/null
@@ -1,22 +0,0 @@
-val verbose : bool ref
-val quiet : bool ref
-val generate_pot : bool ref
-val expand_tabs : int option ref
-val no_cache : bool ref
-val check_unused_global_vars : bool ref
-val check_white_space : bool ref
-val check_suggest_simpler : bool ref
-val check_void : bool ref
-val check_context : bool ref
-val check_strange : bool ref
-val check_traps : bool ref
-val check_complex_expressions : bool ref
-val normalized_expressions : bool ref
-val check_help_perl_checker : bool ref
-val suggest_functional : bool ref
-val check_prototypes : bool ref
-val check_names : bool ref
-val check_import_export : bool ref
-val allow_MDK_Common : bool ref
-val is_warning_type_set : Types.warning -> bool
-val are_warning_types_set : Types.warning list -> bool
diff --git a/perl_checker.src/global_checks.ml b/perl_checker.src/global_checks.ml
deleted file mode 100644
index a63e652..0000000
--- a/perl_checker.src/global_checks.ml
+++ /dev/null
@@ -1,639 +0,0 @@
-open Types
-open Common
-open Printf
-open Config_file
-open Parser_helper
-open Tree
-
-type state = {
- per_files : (string, per_file) Hashtbl.t ;
- per_packages : (string, per_package) Hashtbl.t ;
- methods : (string, (string * variable_used ref * prototype option) list) Hashtbl.t ;
- global_vars_used : ((context * string * string) * pos) list ref ;
- packages_being_classes : (string, unit) Hashtbl.t ;
- packages_dependencies : (string * string, unit) Hashtbl.t ;
- packages_dependencies_maybe : (string * string * string, unit) Hashtbl.t ;
- }
-
-type vars = {
- my_vars : ((context * string) * (pos * variable_used ref * prototype option)) list list ;
- our_vars : ((context * string) * (pos * variable_used ref * prototype option)) list list ;
- locally_imported : ((context * string) * (string * variable_used ref * prototype option)) list ;
- required_vars : (context * string * string) list ;
- current_package : per_package ;
- is_toplevel : bool ;
- write_only : bool ;
- state : state ;
- }
-
-
-let rec get_imported state current_package (package_name, (imports, pos)) =
- try
- let package_used = Hashtbl.find state.per_packages package_name in
- let exports = package_used.exports in
- let get_var_by_name var =
- let (b, prototype) =
- try sndter3 (Hashtbl.find package_used.vars_declared var)
- with Not_found ->
- try
- sndter3 (List.assoc var (get_imports state package_used))
- with Not_found ->
- warn_with_pos [Warn_import_export] pos (sprintf "name %s is not defined in package %s" (variable2s var) package_name) ;
- ref Access_various, None
- in
- var, (package_name, b, prototype)
- in
- match imports with
- | None ->
- let re = match exports.special_export with
- | Some Re_export_all -> get_imports state package_used
- | Some Fake_export_all ->
- (* HACK: if package exporting-all is ignored, ignore package importing *)
- if List.mem package_name !ignored_packages then Tree.ignore_package current_package.package_name;
-
- Hashtbl.fold (fun var (_pos, b, proto) l -> (var, (package_name, b, proto)) :: l) package_used.vars_declared []
- | _ -> [] in
- let l = List.map get_var_by_name exports.export_auto in
- re @ l
- | Some l ->
- let imports_vars =
- collect (function
- | I_raw, tag ->
- (try
- List.assoc tag exports.export_tags
- with Not_found -> warn_with_pos [Warn_import_export] pos (sprintf "package %s doesn't export tag %s" package_name tag) ; [])
- | variable ->
- if List.mem variable exports.export_ok || List.mem variable exports.export_auto then
- [ variable ]
- else
- (warn_with_pos [Warn_import_export] pos (sprintf "package %s doesn't export %s" package_name (variable2s variable)) ; [])
- ) l
- in
- List.map get_var_by_name imports_vars
- with Not_found -> []
-
-and get_imports state package =
- match !(package.imported) with
- | Some l -> l
- | None ->
- let l = collect (get_imported state package) package.uses in
- package.imported := Some l ;
- l
-
-let do_para_comply_with_prototype para proto =
- match proto with
- | Some proto ->
- (match para with
- | [] as paras
- | [List [List paras]]
- | [List paras] ->
- if List.exists is_not_a_scalar paras then 0 else
- let len = List.length paras in
- if len < proto.proto_nb_min then -1
- else (match proto.proto_nb_max with
- | Some max -> if len > max then 1 else 0
- | None -> 0)
- | _ -> 0)
- | _ -> 0
-
-let check_para_comply_with_prototype para proto =
- match para with
- | None -> ()
- | Some(pos, para) ->
- match do_para_comply_with_prototype para proto with
- | -1 -> warn_with_pos [Warn_prototypes] pos "not enough parameters"
- | 1 -> warn_with_pos [Warn_prototypes] pos "too many parameters"
- | _ -> ()
-
-let is_anonymous_variable_name s = String.length s > 1 && s.[0] = '_'
-
-let add_to_packages_really_used state current_package used_name =
- Hashtbl.replace state.packages_dependencies (current_package.package_name, used_name) () ;
- (*List.iter (fun (p1, p2) -> prerr_endline (Printf.sprintf "%s -> %s" p1 p2)) (hashtbl_keys state.packages_dependencies);*)
- ()
-
-let add_to_packages_maybe_used state current_package used_name method_name =
- Hashtbl.replace state.packages_dependencies_maybe (current_package.package_name, used_name, method_name) () ;
- (*List.iter (fun (p1, p2) -> prerr_endline (Printf.sprintf "%s -> %s" p1 p2)) (hashtbl_keys state.packages_dependencies_maybe);*)
- ()
-
-let variable_used write_only used =
- if !used != Access_various then
- used := if write_only then Access_write_only else Access_various
-
-let is_my_declared vars t =
- List.exists (fun l ->
- List.mem_assoc t l && (variable_used vars.write_only (snd3 (List.assoc t l)) ; true)
- ) vars.my_vars
-let is_our_declared vars t =
- List.exists (fun l ->
- List.mem_assoc t l && (variable_used vars.write_only (snd3 (List.assoc t l)) ; true)
- ) vars.our_vars
-
-let is_var_declared_raw write_only state package var para =
- match
- try
- let _, used, proto = Hashtbl.find package.vars_declared var in
- Some(used, proto)
- with Not_found -> try
- let package_name, used, proto = List.assoc var (get_imports state package) in
- add_to_packages_really_used state package package_name ;
- Some(used, proto)
- with Not_found ->
- None
- with
- | Some (used, proto) ->
- check_para_comply_with_prototype para proto ;
- variable_used write_only used ;
- true
- | None ->
- false
-
-let is_var_declared vars var para =
- List.mem_assoc var vars.locally_imported ||
- is_var_declared_raw vars.write_only vars.state vars.current_package var para
-
-let is_global_var_declared vars (context, fq, name) para =
- try
- let package = Hashtbl.find vars.state.per_packages fq in
- add_to_packages_really_used vars.state vars.current_package package.package_name ;
- is_var_declared_raw vars.write_only vars.state package (context, name) para
- with Not_found -> false
-
-
-let is_global_var context ident =
- match context with
- | I_scalar ->
- (match ident with
- | "@" | "!" | ">" | "\\" | "$" | "^A" | "'" | "/" | "?" | "<" | "^W" | "|" | "^I" | "&" | "."
- | "0" | "1" | "2" | "3" | "4" | "5" | "6" | "7" | "8" | "9" -> true
- | _ -> false)
- | I_array ->
- (match ident with
- | "ARGV" | "INC" -> true
- | _ -> false)
- | I_hash ->
- (match ident with
- | "ENV" | "SIG" -> true
- | _ -> false)
- | I_star ->
- (match ident with
- | "STDIN" | "STDOUT" | "STDERR" | "DATA"
- | "__FILE__" | "__LINE__" | "undef" -> true
- | _ -> false)
- | I_func ->
- (match ident with
- | "-b" | "-c" | "-d" | "-e" | "-f" | "-l" | "-r" | "-s" | "-w" | "-x"
- | "abs" | "alarm" | "atan2" | "bless"
- | "caller" | "chdir" | "chmod" | "chomp" | "chop" | "chown" | "chr" | "chroot" | "close" | "closedir" | "cos" | "crypt"
- | "defined" | "delete" | "die"
- | "each" | "endpwent" | "eof" | "eval" | "exec" | "exists" | "exit"
- | "fcntl" | "fileno" | "flock" | "formline" | "fork"
- | "gethostbyaddr" | "gethostbyname" | "getgrent" | "getgrnam" | "getgrgid" | "getppid" | "getpwent" | "getpwnam" | "getpwuid" | "getservbyname" | "glob" | "gmtime" | "goto" | "grep" | "hex"
- | "index" | "int" | "ioctl" | "join" | "keys" | "kill"
- | "last" | "lc" | "lcfirst" | "length" | "link" | "localtime" | "log" | "lstat"
- | "map" | "mkdir" | "next" | "no" | "oct" | "open" | "opendir" | "ord"
- | "pack" | "pipe" | "pop" | "print" | "printf" | "push" | "quotemeta"
- | "rand" | "read" | "readdir" | "readlink" | "redo" | "ref" | "rename" | "require" | "return" | "reverse" | "rindex" | "rmdir"
- | "scalar" | "seek" | "select" | "setpwent" | "shift" | "sin" | "sleep" | "sort" | "splice" | "split" | "sprintf" | "sqrt" | "stat" | "substr"
- | "symlink" | "syscall" | "sysopen" | "sysread" | "sysseek" | "system" | "syswrite" | "tie" | "time"
- | "uc" | "ucfirst" | "umask" | "undef" | "unlink" | "unpack" | "unshift" | "utime" | "values" | "vec" | "wait" | "waitpid" | "wantarray" | "warn" | "write"
- -> true
-
- | _ -> false)
- | _ -> false
-
-let check_variable (context, var) vars para =
- match var with
- | Ident(_, s, pos) when context <> I_func && is_anonymous_variable_name s && s <> "__FILE__" && s <> "__LINE__" ->
- warn_with_pos [Warn_normalized_expressions] pos (sprintf "variable %s must not be used\n (variable with name _XXX are reserved for unused variables)" (variable2s(context, string_of_fromparser var)))
- | Ident(Some pkg, _, _) when uses_external_package pkg || List.mem pkg !ignored_packages -> ()
- | Ident(None, ident, pos) ->
- if is_my_declared vars (context, ident) || is_our_declared vars (context, ident) || is_var_declared vars (context, ident) para || is_global_var context ident
- then ()
- else warn_with_pos [Warn_names] pos (if context = I_func then "unknown function " ^ ident else "undeclared variable " ^ variable2s(context, ident))
- | Ident(Some fq, name, pos) ->
- if (fq = "CORE") && is_global_var context name || is_global_var_declared vars (context, fq, name) para
- then ()
- else
- if context = I_func then
- warn_with_pos [Warn_names] pos ("unknown function " ^ string_of_fromparser var)
- else
- lpush vars.state.global_vars_used ((context, fq, name), pos)
- | _ -> ()
-
-let declare_My vars (mys, pos) =
- let l_new = List.filter (fun (context, ident) ->
- if context = I_raw then
- if ident = "undef" then false else die_with_pos pos (sprintf "bad ident %s in my" ident)
- else true
- ) mys in
- let l_pre = List.hd vars.my_vars in
- List.iter (fun v ->
- if List.mem_assoc v l_pre then warn_with_pos [Warn_names] pos (sprintf "redeclared variable %s" (variable2s v))
- ) l_new ;
- { vars with my_vars = (List.map (fun v -> v, (pos, ref Access_none, None)) l_new @ l_pre) :: List.tl vars.my_vars }
-
-let declare_Our vars (ours, pos) =
- match vars.our_vars with
- | [] -> vars (* we're at the toplevel, already declared in vars_declared *)
- | l_pre :: other ->
- List.iter (fun v ->
- if List.mem_assoc v l_pre && v <> (I_scalar, "_") then warn_with_pos [Warn_names] pos (sprintf "redeclared variable %s" (variable2s v))
- ) ours ;
- { vars with our_vars = (List.map (fun v -> v, (pos, ref Access_none, None)) ours @ l_pre) :: other }
-
-let declare_My_our vars (my_or_our, l, pos) =
- match my_or_our with
- | "my" -> declare_My vars (l, pos)
- | "local"
- | "our" -> declare_Our vars (l, pos)
- | _ -> internal_error "declare_My_our"
-
-let un_parenthesize_one_elt_List = function
- | [List l] -> l
- | l -> l
-
-let check_unused_local_variables vars =
- List.iter (fun ((context, s as v), (pos, used, _proto)) ->
- if !used != Access_various then
- match s with
- | "BEGIN" | "END" | "DESTROY" -> ()
- | "_" when context = I_array ->
- warn_with_pos [Warn_normalized_expressions] pos "if the function doesn't take any parameters, please use the empty prototype.\nexample \"sub foo() { ... }\""
- | _ ->
- if s.[0] != '_' || s = "_" then
- let msg = if !used = Access_write_only then sprintf "variable %s assigned, but not read" else sprintf "unused variable %s" in
- warn_with_pos [Warn_names] pos (msg (variable2s v))
- ) (List.hd vars.my_vars)
-
-let check_variables vars t =
- let rec check_variables_ vars t = fold_tree check vars t
- and check vars = function
- | Block l ->
- let vars' = { vars with my_vars = [] :: vars.my_vars ; our_vars = [] :: vars.our_vars } in
- let vars' = List.fold_left check_variables_ vars' l in
- check_unused_local_variables vars' ;
- Some vars
- | Call(Deref(I_func, Ident(None, "sort", _)), (Anonymous_sub(_, Block f, pos) :: l)) ->
- let vars = List.fold_left check_variables_ vars l in
- let vars' = { vars with my_vars = [ (I_scalar, "a"), (pos, ref Access_various, None) ; (I_scalar, "b"), (pos, ref Access_various, None) ] :: vars.my_vars ; our_vars = [] :: vars.our_vars } in
- let vars' = List.fold_left check_variables_ vars' f in
- check_unused_local_variables vars' ;
- Some vars
-
- | Call(Deref(I_func, Ident(None, func, func_pos)), Anonymous_sub(_, Block f, pos) :: l)
- when List.mem func [ "grep" ; "map" ; "substInFile" ; "map_index" ; "each_index" ; "partition" ; "find_index" ; "grep_index" ; "find" ; "any" ; "every" ; "uniq_" ] ->
- let vars = List.fold_left check_variables_ vars l in
- let vars' = { vars with my_vars = [] :: vars.my_vars ; our_vars = [(I_scalar, "_"), (pos, ref Access_various, None)] :: vars.our_vars } in
- let vars' = List.fold_left check_variables_ vars' f in
- check_unused_local_variables vars' ;
- check_variable (I_func, Ident(None, func, func_pos)) vars None ;
- Some vars
-
- | Call(Deref(I_func, (Ident _ as ident)), [ Deref(I_star, (Ident(None, "_", _))) ]) ->
- (* the &f case: allow access to @_ *)
- check_variable (I_func, ident) vars None ;
- let _ = is_my_declared vars (I_array, "_") in
- Some vars
-
- | Call(Deref(I_func, (Ident _ as ident)), [ List [ Deref(I_array, (Ident(None, "_", pos))) ] ]) ->
- (* special warning if @_ is unbound *)
- check_variable (I_func, ident) vars None ;
- if not (is_my_declared vars (I_array, "_")) then
- warn_with_pos [Warn_suggest_simpler] pos (sprintf "replace %s(@_) with &%s" (string_of_fromparser ident) (string_of_fromparser ident)) ;
- Some vars
-
- | Call(Deref(I_func, Ident(None, "require", _)), [Ident _]) -> Some vars
-
- | Call(Deref(I_func, Ident(None, "shift", pos)) as var, [])
- | Call(Deref(I_func, Ident(None, "pop", pos)) as var, []) ->
- check vars (Call(var, [ Deref(I_array, Ident(None, (if vars.is_toplevel then "ARGV" else "_"), pos)) ]))
-
- | Call(Deref(context, (Ident(_, _, pos) as var)), para) ->
- check_variable (context, var) vars (Some(pos, para)) ;
- let vars = List.fold_left check_variables_ vars para in
- Some vars
-
-(* | Call_op("=", -> List.fold_left (fold_tree f) env l*)
-
- | Call_op("while infix", [ expr ; (List [ Call_op("<>", _, _) ] as l) ], pos)
- | Call_op("for infix", [ expr ; l ], pos) ->
- let vars = check_variables_ vars l in
- let vars' = { vars with my_vars = [] :: vars.my_vars ; our_vars = [(I_scalar, "_"), (pos, ref Access_various, None)] :: vars.our_vars } in
- let vars' = check_variables_ vars' expr in
- if List.hd(vars'.my_vars) <> [] then warn_with_pos [Warn_traps] pos "you can't declare variables in foreach postfix";
- Some vars
-
- | Call_op("foreach my", [my; expr; Block block], _) ->
- let vars = check_variables_ vars expr in
- let vars = check_variables_ vars (Block (my :: block)) in
- Some vars
- | Call_op(op, l, _) when op = "if" || op = "while" || op = "unless" || op = "until" ->
- let vars' = { vars with my_vars = [] :: vars.my_vars ; our_vars = [] :: vars.our_vars } in
- let vars' = List.fold_left check_variables_ vars' l in
- check_unused_local_variables vars' ;
- Some vars
-
- | Sub_declaration(Ident(fq, name, pos) as ident, perl_proto, Block body, kind) ->
- let vars = declare_Our vars ([ I_func, string_of_fromparser ident ], pos) in
-
- let my_vars, l =
- match has_proto perl_proto (Block body) with
- | Some(mys, mys_pos, body) ->
- [], My_our ("my", mys, mys_pos) :: body
- | _ ->
- let dont_check_use =
- kind = Glob_assign ||
- fq = None && List.mem name ["DESTROY"] ||
- Hashtbl.mem vars.state.packages_being_classes (some_or fq vars.current_package.package_name)
- in
- [(I_array, "_"), (pos, ref (if dont_check_use then Access_various else Access_none), None)], body
- in
- let local_vars =
- if fq = None && name = "AUTOLOAD"
- then [ (I_scalar, "AUTOLOAD"), (pos, ref Access_various, None) ]
- else [] in
-
- let vars' = { vars with my_vars = my_vars :: vars.my_vars ; our_vars = local_vars :: vars.our_vars ; is_toplevel = false } in
- let vars' = List.fold_left check_variables_ vars' l in
- check_unused_local_variables vars' ;
- Some vars
-
- | Anonymous_sub(_, Block l, pos) ->
- let vars' = { vars with my_vars = [(I_array, "_"), (pos, ref Access_various, None)] :: vars.my_vars ; is_toplevel = false } in
- let vars' = List.fold_left check_variables_ vars' l in
- check_unused_local_variables vars' ;
- Some vars
-
- | Call_op("foreach", [ expr ; Block l ], pos) ->
- let vars = check_variables_ vars expr in
- let vars' = { vars with my_vars = [] :: vars.my_vars ; our_vars = [(I_scalar, "_"), (pos, ref Access_various, None)] :: vars.our_vars } in
- let vars' = List.fold_left check_variables_ vars' l in
- check_unused_local_variables vars' ;
- Some vars
-
- | Anonymous_sub _
- | Sub_declaration _ -> internal_error "check_variables"
-
- | Ident _ as var ->
- check_variable (I_star, var) vars None ;
- Some vars
-
- | My_our(my_or_our, mys, pos) -> Some(declare_My_our vars (my_or_our, mys, pos))
- | Deref(context, (Ident _ as var)) ->
- check_variable (context, var) vars None ;
- Some vars
- | Deref_with(context, _, (Ident _ as var), para) ->
- let vars = check_variables_ vars para in
- check_variable (context, var) vars None ;
- Some vars
-
- | Call_op("=", [My_our(my_or_our, mys, pos); e], _) ->
- (* check e first *)
- let vars = check_variables_ vars e in
- List.iter (fun (context, var) ->
- if non_scalar_context context then warn_with_pos [Warn_prototypes] pos (sprintf "%s takes all the arguments, %s is undef in any case" (variable2s (context, var)) (variable2s (last mys)))
- ) (removelast mys) ; (* mys is never empty *)
- Some(declare_My_our vars (my_or_our, mys, pos))
-
- | Call_op("if infix", [List [My_our _]; List [Num("0", _)]], _) -> None (* special allowed case *)
- | Call_op(op, List (My_our _ :: _) :: _, pos)
- | Call_op(op, My_our _ :: _, pos)
- | Call_op(op, Call_op("local", _, _) :: _, pos) ->
- if op <> "=" then warn_with_pos [Warn_traps] pos (sprintf "applying %s on a new initialized variable is wrong" op);
- None
-
- | Call_op("=", [ Deref(context, (Ident _ as var)) ; para], _) ->
- check_variable (context, var) { vars with write_only = true } None ;
- Some (check_variables_ vars para)
-
- | Call_op("=", [ List [ List l ] ; para], _) ->
- let vars = List.fold_left (fun vars -> function
- | Deref(context, (Ident _ as var)) ->
- check_variable (context, var) { vars with write_only = true } None ;
- vars
- | e -> check_variables_ vars e
- ) vars l in
- let vars = check_variables_ vars para in
- Some vars
-
- | Method_call(Raw_string(package_name, pos), Raw_string ("import", _), para) ->
- let args =
- match para with
- | [] -> None
- | [ List [v] ] -> Some(from_qw v)
- | _ -> die_with_pos pos "bad import statement" in
- let l = get_imported vars.state vars.current_package (package_name, (args, pos)) in
- let vars =
- if vars.is_toplevel then (
- vars.current_package.imported := Some (get_imports vars.state vars.current_package @ l) ;
- vars
- ) else
- { vars with locally_imported = l @ vars.locally_imported } in
- Some vars
-
- | Method_call(Raw_string(pkg, _) as class_, Raw_string(method_, pos), para) ->
- let vars = List.fold_left check_variables_ vars para in
- let rec search pkg =
- if is_global_var_declared vars (I_func, pkg, method_) (Some(pos, [ List (class_ :: un_parenthesize_one_elt_List para) ])) then true
- else
- let package = Hashtbl.find vars.state.per_packages pkg in
- List.exists search (List.map fst (some_or package.isa []))
- in
- (try
- if not (uses_external_package pkg || List.mem pkg !ignored_packages || search pkg || method_ = "bootstrap") then
- warn_with_pos [Warn_import_export] pos (sprintf "unknown method %s starting in package %s" method_ pkg);
- with Not_found -> warn_with_pos [Warn_import_export] pos (sprintf "unknown package %s" pkg));
- Some vars
-
- | Method_call(o, Raw_string(method_, pos), para) ->
- let vars = check_variables_ vars o in
- let vars = List.fold_left check_variables_ vars para in
- (try
- let l = Hashtbl.find vars.state.methods method_ in
- let l_and = List.map (fun (pkg_name, used, proto) -> pkg_name, used, do_para_comply_with_prototype [ List (o :: un_parenthesize_one_elt_List para) ] proto) l in
- let l_and' =
- match List.filter (fun (_, _, n) -> n = 0) l_and with
- | [] ->
- (match uniq (List.map ter3 l_and) with
- | [-1] -> warn_with_pos [Warn_prototypes] pos "not enough parameters"
- | [ 1] -> warn_with_pos [Warn_prototypes] pos "too many parameters"
- | _ -> warn_with_pos [Warn_prototypes] pos "not enough or too many parameters") ;
- l_and
- | l -> l
- in
- List.iter (fun (pkg_name, _, _) -> add_to_packages_maybe_used vars.state vars.current_package pkg_name method_) l_and' ;
- List.iter (fun (_, used, _) -> used := Access_various) l_and'
- with Not_found ->
- if not (List.mem method_ [ "isa"; "can" ]) then
- warn_with_pos [Warn_names] pos ("unknown method " ^ method_)) ;
- Some vars
-
- | _ -> None
- in
- let vars = List.fold_left check_variables_ { vars with my_vars = [[]] } t in
- vars
-
-let check_tree state package =
- let vars = { my_vars = [[]]; our_vars = []; locally_imported = []; required_vars = []; current_package = package; state = state; is_toplevel = true; write_only = false } in
- if !Flags.verbose then print_endline_flush ("checking package " ^ package.package_name) ;
- let vars = check_variables vars package.body in
- check_unused_local_variables vars ;
- ()
-
-let imported_add i1 i2 = if i1 = None && i2 = None then None else Some (some_or i1 [] @ some_or i2 [])
-
-let add_package_to_state state package =
- let package =
- try
- let existing_package = Hashtbl.find state.per_packages package.package_name in
- (*print_endline_flush (existing_package.file_name ^ " vs " ^ package.file_name); *)
- let vars_declared = existing_package.vars_declared in
- Hashtbl.iter (fun var pos -> Hashtbl.replace vars_declared var pos) package.vars_declared ;
- let p = {
- package_name = package.package_name ; has_package_name = package.has_package_name ;
- isa = if existing_package.isa = None then package.isa else existing_package.isa ;
- body = existing_package.body @ package.body ;
- uses = existing_package.uses @ package.uses ;
- required_packages = existing_package.required_packages @ package.required_packages ;
- vars_declared = vars_declared ;
- imported = ref (imported_add !(existing_package.imported) !(package.imported)) ;
- exports = { export_ok = existing_package.exports.export_ok @ package.exports.export_ok ;
- export_auto = existing_package.exports.export_auto @ package.exports.export_auto ;
- export_tags = existing_package.exports.export_tags @ package.exports.export_tags ;
- special_export = None }
- } in
- Hashtbl.replace state.per_packages package.package_name p ;
- p
- with Not_found -> package
- in
- Hashtbl.replace state.per_packages package.package_name package
-
-let add_file_to_files per_files file =
- Hashtbl.replace per_files file.file_name file
-
-let check_unused_vars package =
- Hashtbl.iter (fun (context, name) (pos, is_used, _proto) ->
- if !is_used != Access_various && not (List.mem name ["BEGIN"; "END"; "DESTROY"; "ISA"; "AUTOLOAD"; "EXPORT"; "EXPORT_OK"; "EXPORT_TAGS"]) then
- warn_with_pos [Warn_unused_global_vars] pos (sprintf "unused %s%s::%s" (if context = I_func then "function " else "variable " ^ context2s context) package.package_name name)
- ) package.vars_declared
-
-let arrange_global_vars_declared global_vars_declared state =
- Hashtbl.iter (fun (context, fq, name) (pos, proto) ->
- let package =
- try
- Hashtbl.find state.per_packages fq
- with Not_found ->
- (* creating a new shadow package *)
- let package =
- {
- package_name = fq;
- has_package_name = true ;
- exports = empty_exports ;
- imported = ref None ;
- vars_declared = Hashtbl.create 16 ;
- uses = [] ;
- required_packages = [] ;
- body = [] ;
- isa = None ;
- } in
- Hashtbl.add state.per_packages fq package ;
- package
- in
- if not (Hashtbl.mem package.vars_declared (context, name)) then
- Hashtbl.add package.vars_declared (context, name) (pos, ref Access_none, proto)
- (* otherwise dropping this second declaration *)
- ) global_vars_declared ;
- state
-
-let get_methods_available state =
- let classes = uniq (
- hashtbl_collect (fun _ package ->
- match package.isa with
- | None ->
- if Hashtbl.mem package.vars_declared (I_func, "new") then [package] else []
- | Some l ->
- package :: List.map (fun (pkg, pos) ->
- try
- Hashtbl.find state.per_packages pkg
- with Not_found -> die_with_pos pos ("bad package " ^ pkg)
- ) l
- ) state.per_packages
- ) in
- List.iter (fun pkg ->
- Hashtbl.replace state.packages_being_classes pkg.package_name () ;
- Hashtbl.iter (fun (context, v) (_pos, is_used, proto) ->
- if context = I_func then
- let l = try Hashtbl.find state.methods v with Not_found -> [] in
- Hashtbl.replace state.methods v ((pkg.package_name, is_used, proto) :: l)
- ) pkg.vars_declared
- ) classes ;
- state
-
-
-let default_per_files() = Hashtbl.create 16
-let default_state per_files = {
- per_files = per_files;
- per_packages = Hashtbl.create 16;
- methods = Hashtbl.create 256;
- global_vars_used = ref [];
- packages_being_classes = Hashtbl.create 16;
- packages_dependencies = Hashtbl.create 16;
- packages_dependencies_maybe = Hashtbl.create 16
-}
-
-let cache_cache = Hashtbl.create 16
-
-let pkgs2s prefix l =
- let l = List.sort compare (List.map (fun pkg -> pkg.file_name) l) in
- String.concat "" (List.map (fun s -> prefix ^ s ^ "\n") l)
-
-let read_packages_from_cache per_files dir =
- if !Flags.no_cache || Hashtbl.mem cache_cache dir then () else
- try
- Hashtbl.add cache_cache dir ();
- let file = dir ^ "/.perl_checker.cache" in
- let fh = open_in file in
- let magic = input_line fh in
- if magic <> "perl_checker cache " ^ Build.date then () else
- let l = Marshal.from_channel fh in
- close_in fh ;
-
- let l = List.filter (fun file ->
- not (Hashtbl.mem per_files file.file_name) &&
- (try file.build_time > mtime file.file_name with _ -> false)
- ) l in
-
- if !Flags.verbose then print_endline_flush (sprintf "using cached files\n%sfrom %s" (pkgs2s " " l) file) ;
-
- List.iter (fun file ->
- Info.add_a_file file.file_name file.lines_starts ;
- add_file_to_files per_files file
- ) l
- with Sys_error _ | End_of_file -> ()
-
-let write_packages_cache per_files dir =
- try
- let l = List.filter (fun per_file -> per_file.require_name <> None) (hashtbl_values per_files) in
- let file = dir ^ "/.perl_checker.cache" in
- let fh = open_out file in
- output_string fh ("perl_checker cache " ^ Build.date ^ "\n") ;
- Marshal.to_channel fh l [] ;
- close_out fh ;
- if !Flags.verbose then print_endline_flush (sprintf "saving cached files\n%sin %s" (pkgs2s " " l) file)
- with Sys_error _ -> ()
-
-let generate_package_dependencies_graph state file =
- let fh = open_out file in
-
- List.iter (fun (p1, p2) ->
- output_string fh (p1 ^ " -> " ^ p2 ^ "\n")
- ) (List.sort compare (hashtbl_keys state.packages_dependencies)) ;
-
- let l = Hashtbl.fold (fun (p1, p2, method_) _ l -> ((p1, method_), p2) :: l) state.packages_dependencies_maybe [] in
- List.iter (fun ((p1, method_), l) ->
- output_string fh (p1 ^ " ?-> " ^ String.concat " " l ^ " (" ^ method_ ^ ")\n")
- ) (List.sort compare (prepare_want_all_assoc l));
-
- close_out fh
diff --git a/perl_checker.src/global_checks.mli b/perl_checker.src/global_checks.mli
deleted file mode 100644
index 9edacbf..0000000
--- a/perl_checker.src/global_checks.mli
+++ /dev/null
@@ -1,26 +0,0 @@
-open Types
-open Tree
-
-type state = {
- per_files : (string, per_file) Hashtbl.t ;
- per_packages : (string, per_package) Hashtbl.t ;
- methods : (string, (string * variable_used ref * prototype option) list) Hashtbl.t ;
- global_vars_used : ((context * string * string) * pos) list ref ;
- packages_being_classes : (string, unit) Hashtbl.t ;
- packages_dependencies : (string * string, unit) Hashtbl.t ;
- packages_dependencies_maybe : (string * string * string, unit) Hashtbl.t ;
- }
-
-val default_per_files : unit -> (string, per_file) Hashtbl.t
-val default_state : (string, per_file) Hashtbl.t -> state
-val check_tree : state -> per_package -> unit
-val add_file_to_files : (string, per_file) Hashtbl.t -> per_file -> unit
-val add_package_to_state : state -> per_package -> unit
-val check_unused_vars : per_package -> unit
-val arrange_global_vars_declared : (context * string * string, pos * Tree.prototype option) Hashtbl.t -> state -> state
-val get_methods_available : state -> state
-
-val read_packages_from_cache : (string, per_file) Hashtbl.t -> string -> unit
-val write_packages_cache : (string, per_file) Hashtbl.t -> string -> unit
-
-val generate_package_dependencies_graph : state -> string -> unit
diff --git a/perl_checker.src/info.ml b/perl_checker.src/info.ml
deleted file mode 100644
index ab76b9f..0000000
--- a/perl_checker.src/info.ml
+++ /dev/null
@@ -1,76 +0,0 @@
-open List
-open Printf
-open Common
-
-let (lines_starts : (string, int list) Hashtbl.t) = Hashtbl.create 4
-let current_file_lines_starts = ref []
-let current_file_current_line = ref 0
-let current_file = ref ""
-
-let start_a_new_file file =
- if !current_file <> "" then Hashtbl.add lines_starts !current_file !current_file_lines_starts ;
- current_file := file ;
- current_file_lines_starts := [0]
-
-let add_a_file file file_lines_starts = Hashtbl.replace lines_starts file file_lines_starts
-
-let get_lines_starts_for_file file =
- if file = !current_file then !current_file_lines_starts else Hashtbl.find lines_starts file
-
-let cwd = expand_symlinks (Unix.getcwd())
-
-let file_to_absolute_file file =
- let abs_file =
- if file.[0] = '/' then file else
- if file = "." then cwd else cwd ^ "/" ^ file
- in
- expand_symlinks abs_file
-
-let absolute_file_to_file =
- let s1 = Filename.dirname cwd in
- if String.length s1 < 4 then (fun x -> x) else
- let short_cwd =
- let s2 = Filename.dirname s1 in
- if String.length s2 < 4 then s1 else
- let s3 = Filename.dirname s2 in (* allow up to ../../../xxx *)
- if String.length s3 < 4 then s2 else s3 in
- memoize (fun abs_file ->
- if str_begins_with (short_cwd ^ "/") abs_file then
- let rec to_file rel cwd =
- if str_begins_with (cwd ^ "/") abs_file then
- rel ^ skip_n_char_ (String.length cwd + 1) 0 abs_file
- else
- to_file ("../" ^ rel) (Filename.dirname cwd)
- in
- to_file "" cwd
- else
- abs_file)
-
-let raw_pos2raw_line file a =
- let starts = map_index (fun a b -> a,b) (rev (get_lines_starts_for_file file)) in
- let ((offset, line), _) = find (fun (_,(e,_)) -> e > a) (combine starts (tl starts @ [999999999, 999999999])) in
- line, offset
-
-let pos2line (file, a, b) =
- let line, offset = raw_pos2raw_line file a in
- file, line, a - offset + 1, b - offset + 1
-
-let pos2s (file, a, b) = sprintf "(%s, %d, %d)" file a b
-
-let pos2sfull pos =
- try
- let file, line, n1, n2 = pos2line pos in
- sprintf "File \"%s\", line %d, character %d-%d\n" (absolute_file_to_file file) (line + 1) n1 n2
- with Not_found -> failwith ("bad position " ^ pos2s pos)
-
-let pos2s_for_po pos =
- let file, line, _, _ = pos2line pos in
- absolute_file_to_file file ^ ":" ^ string_of_int (line + 1)
-
-let is_on_same_line file (a,b) =
- let line_a, _ = raw_pos2raw_line file a in
- let line_b, _ = raw_pos2raw_line file b in
- line_a = line_b
-
-let is_on_same_line_current (a,b) = is_on_same_line !current_file (a,b)
-let pos2sfull_current a b = pos2sfull (!current_file, a, b)
diff --git a/perl_checker.src/info.mli b/perl_checker.src/info.mli
deleted file mode 100644
index d337316..0000000
--- a/perl_checker.src/info.mli
+++ /dev/null
@@ -1,17 +0,0 @@
-val lines_starts : (string, int list) Hashtbl.t
-val current_file_lines_starts : int list ref
-val current_file_current_line : int ref
-val current_file : string ref
-val start_a_new_file : string -> unit
-val add_a_file : string -> int list -> unit
-val get_lines_starts_for_file : string -> int list
-val file_to_absolute_file : string -> string
-val absolute_file_to_file : string -> string
-val raw_pos2raw_line : string -> int -> int * int
-val pos2line : string * int * int -> string * int * int * int
-val pos2s : string * int * int -> string
-val pos2sfull : string * int * int -> string
-val pos2s_for_po : string * int * int -> string
-val is_on_same_line : string -> int * int -> bool
-val is_on_same_line_current : int * int -> bool
-val pos2sfull_current : int -> int -> string
diff --git a/perl_checker.src/lexer.mll b/perl_checker.src/lexer.mll
deleted file mode 100644
index f416499..0000000
--- a/perl_checker.src/lexer.mll
+++ /dev/null
@@ -1,1057 +0,0 @@
-{ (* -*- caml -*- *)
-open Common
-open Types
-open Lexing
-open Info
-
-let bpos = -1,-1
-
-type raw_token =
- | EOF of raw_pos
- | SPACE of int
- | CR
- | INT of (string * raw_pos)
- | FLOAT of (string * raw_pos)
- | RAW_STRING of (string * raw_pos)
- | STRING of (raw_interpolated_string * raw_pos)
- | PATTERN of (raw_interpolated_string * string * raw_pos)
- | QR_PATTERN of (raw_interpolated_string * string * raw_pos)
- | PATTERN_SUBST of (raw_interpolated_string * raw_interpolated_string * string * raw_pos)
- | BAREWORD of (string * raw_pos)
- | BAREWORD_PAREN of (string * raw_pos)
- | REVISION of (string * raw_pos)
- | PERL_CHECKER_COMMENT of (string * raw_pos)
- | PO_COMMENT of (string * raw_pos)
- | POD of (string * raw_pos)
- | LABEL of (string * raw_pos)
- | COMMAND_STRING of (raw_interpolated_string * raw_pos)
- | PRINT_TO_STAR of ((string * string) * raw_pos)
- | PRINT_TO_SCALAR of ((string * string) * raw_pos)
- | QUOTEWORDS of (string * raw_pos)
- | COMPACT_HASH_SUBSCRIPT of (string * raw_pos)
- | RAW_HERE_DOC of ((string * raw_pos) ref * raw_pos)
- | HERE_DOC of (raw_interpolated_string * raw_pos) ref * raw_pos
- | FORMAT of (raw_interpolated_string * raw_pos) ref * raw_pos
- | SCALAR_IDENT of (string option * string * raw_pos)
- | ARRAY_IDENT of (string option * string * raw_pos)
- | HASH_IDENT of (string option * string * raw_pos)
- | FUNC_IDENT of (string option * string * raw_pos)
- | STAR_IDENT of (string option * string * raw_pos)
- | RAW_IDENT of (string option * string * raw_pos)
- | RAW_IDENT_PAREN of (string option * string * raw_pos)
- | ARRAYLEN_IDENT of (string option * string * raw_pos)
- | SUB_WITH_PROTO of (string * raw_pos)
- | FUNC_DECL_WITH_PROTO of (string option * string * string * raw_pos)
-
- | IF of raw_pos | ELSIF of raw_pos | ELSE of raw_pos | UNLESS of raw_pos | DO of raw_pos | WHILE of raw_pos | UNTIL of raw_pos | MY_OUR of (string * raw_pos) | CONTINUE of raw_pos | SUB of raw_pos
- | LOCAL of raw_pos | FOR of (string * raw_pos) | USE of raw_pos | PACKAGE of raw_pos | BEGIN of raw_pos | END of raw_pos | PRINT of (string * raw_pos)
- | NEW of (raw_pos) | AT of raw_pos | DOLLAR of raw_pos | PERCENT of raw_pos | AMPERSAND of raw_pos
- | STAR of raw_pos | ARRAYLEN of raw_pos | SEMI_COLON of raw_pos | PKG_SCOPE of raw_pos | PAREN of raw_pos | PAREN_END of raw_pos | BRACKET of raw_pos
- | BRACKET_END of raw_pos | BRACKET_HASHREF of raw_pos | ARRAYREF of raw_pos | ARRAYREF_END of raw_pos | ARROW of raw_pos | INCR of raw_pos | DECR of raw_pos
- | CONCAT of raw_pos | POWER of raw_pos | TIGHT_NOT of raw_pos | BIT_NEG of raw_pos | REF of raw_pos | ONE_SCALAR_PARA of (string * raw_pos) | PATTERN_MATCH of raw_pos | PATTERN_MATCH_NOT of raw_pos | MULT of (string * raw_pos) | MULT_L_STR of raw_pos
- | PLUS of (string * raw_pos) | BIT_SHIFT of (string * raw_pos)
- | LT of raw_pos | GT of raw_pos | COMPARE_OP of (string * raw_pos) | COMPARE_OP_STR of (string * raw_pos) | EQ_OP of (string * raw_pos) | EQ_OP_STR of (string * raw_pos)
- | BIT_AND of raw_pos | BIT_OR of raw_pos | BIT_XOR of raw_pos | AND_TIGHT of raw_pos | OR_TIGHT of raw_pos | DOTDOT of (string * raw_pos)
- | QUESTION_MARK of raw_pos | COLON of raw_pos | ASSIGN of (string * raw_pos) | COMMA of raw_pos | RIGHT_ARROW of raw_pos | NOT of raw_pos | AND of raw_pos | OR of raw_pos | XOR of raw_pos
-
-and raw_interpolated_string = (string * raw_token list) list
-
-let new_any mcontext any spaces pos = { mcontext = mcontext ; any = any ; spaces = spaces ; pos = pos }
-
-let pos lexbuf = lexeme_start lexbuf, lexeme_end lexbuf
-let pos2sfull_with start end_ = Info.pos2sfull (!current_file, start, end_)
-let pos2sfull lexbuf = pos2sfull_with (lexeme_start lexbuf) (lexeme_end lexbuf)
-
-let warn_with_pos warn_types (start, end_) err = if Flags.are_warning_types_set warn_types then print_endline_flush (pos2sfull_with start end_ ^ err)
-let warn warn_types lexbuf err = warn_with_pos warn_types (pos lexbuf) err
-let die lexbuf err = failwith (pos2sfull_with (lexeme_start lexbuf) (lexeme_end lexbuf) ^ err)
-
-let rec concat_bareword_paren accu = function
- | PRINT(s, pos1) :: PAREN(pos2) :: l
- | BAREWORD(s, pos1) :: PAREN(pos2) :: l ->
- concat_bareword_paren (PAREN(pos2) :: BAREWORD_PAREN(s, pos1) :: accu) l
- | RAW_IDENT(kind, ident, pos1) :: PAREN(pos2) :: l ->
- concat_bareword_paren (PAREN(pos2) :: RAW_IDENT_PAREN(kind, ident, pos1) :: accu) l
- | PO_COMMENT(_, pos) as e :: l ->
- let l = drop_while (function CR | SPACE _ -> true | _ -> false) l in
- (match l with
- | PO_COMMENT _ :: _
- (* the check will be done on this PO_COMMENT *)
- | BAREWORD("N", _) :: PAREN(_) :: _
- | BAREWORD("N_", _) :: PAREN(_) :: _ ->
- concat_bareword_paren (e :: accu) l
- | _ ->
- warn_with_pos [Warn_MDK_Common] pos "N(...) must follow the #-PO: comment, with nothing in between" ;
- concat_bareword_paren accu l)
- | [] -> List.rev accu
- | e :: l ->
- concat_bareword_paren (e :: accu) l
-
-let rec bracket_bareword_is_hashref accu = function
- | (pos, Parser.BRACKET bracket) :: (_, Parser.BAREWORD _ as bareword) :: (_, Parser.RIGHT_ARROW _ as right_arrow) :: l ->
- bracket_bareword_is_hashref (right_arrow :: bareword :: (pos, Parser.BRACKET_HASHREF bracket) :: accu) l
- | [] -> List.rev accu
- | e :: l ->
- bracket_bareword_is_hashref (e :: accu) l
-
-
-let rec raw_token_to_pos_and_token spaces = function
- | INT(s, pos) -> pos, Parser.NUM(new_any M_int s spaces pos)
- | FLOAT(s, pos) -> pos, Parser.NUM(new_any M_float s spaces pos)
- | RAW_STRING(s, pos) -> pos, Parser.RAW_STRING(new_any M_string s spaces pos)
- | RAW_HERE_DOC(r, pos) -> pos, Parser.RAW_HERE_DOC(new_any M_string !r spaces pos)
- | STRING(l, pos) -> pos, Parser.STRING(new_any M_string (raw_interpolated_string_to_tokens l) spaces pos)
- | COMMAND_STRING(l, pos) -> pos, Parser.COMMAND_STRING(new_any (M_mixed [M_string; M_array]) (raw_interpolated_string_to_tokens l) spaces pos)
- | QR_PATTERN(s, opts, pos) -> pos, Parser.QR_PATTERN(new_any M_special (raw_interpolated_string_to_tokens s, opts) spaces pos)
- | PATTERN(s, opts, pos) -> pos, Parser.PATTERN(new_any M_special (raw_interpolated_string_to_tokens s, opts) spaces pos)
- | PATTERN_SUBST(from, to_, opts, pos) -> pos, Parser.PATTERN_SUBST(new_any M_special (raw_interpolated_string_to_tokens from, raw_interpolated_string_to_tokens to_, opts) spaces pos)
- | HERE_DOC(l, pos) -> pos, Parser.HERE_DOC(new_any M_string (raw_interpolated_string_to_tokens (fst !l), snd !l) spaces pos)
- | FORMAT(l, pos) -> pos, Parser.FORMAT(new_any M_string (raw_interpolated_string_to_tokens (fst !l), snd !l) spaces pos)
- | BAREWORD(s, pos) -> pos, Parser.BAREWORD(new_any M_special s spaces pos)
- | BAREWORD_PAREN(s, pos) -> pos, Parser.BAREWORD_PAREN(new_any M_special s spaces pos)
- | REVISION(s, pos) -> pos, Parser.REVISION(new_any M_revision s spaces pos)
- | PERL_CHECKER_COMMENT(s, pos) -> pos, Parser.PERL_CHECKER_COMMENT(new_any M_none s spaces pos)
- | PO_COMMENT(s, pos) -> pos, Parser.PO_COMMENT(new_any M_special s spaces pos)
- | POD(s, pos) -> pos, Parser.POD(new_any M_special s spaces pos)
- | LABEL(s, pos) -> pos, Parser.LABEL(new_any M_none s spaces pos)
- | PRINT(s, pos) -> pos, Parser.PRINT(new_any M_special s spaces pos)
- | PRINT_TO_STAR(s, pos) -> pos, Parser.PRINT_TO_STAR(new_any M_special s spaces pos)
- | PRINT_TO_SCALAR(s, pos) -> pos, Parser.PRINT_TO_SCALAR(new_any M_special s spaces pos)
- | QUOTEWORDS(s, pos) -> pos, Parser.QUOTEWORDS(new_any M_array s spaces pos)
- | COMPACT_HASH_SUBSCRIPT(s, pos) -> pos, Parser.COMPACT_HASH_SUBSCRIPT(new_any M_special s spaces pos)
- | SCALAR_IDENT(kind, name, pos) -> pos, Parser.SCALAR_IDENT(new_any M_special (kind, name) spaces pos)
- | ARRAY_IDENT(kind, name, pos) -> pos, Parser.ARRAY_IDENT(new_any M_special (kind, name) spaces pos)
- | HASH_IDENT(kind, name, pos) -> pos, Parser.HASH_IDENT(new_any M_special (kind, name) spaces pos)
- | FUNC_IDENT(kind, name, pos) -> pos, Parser.FUNC_IDENT(new_any M_special (kind, name) spaces pos)
- | STAR_IDENT(kind, name, pos) -> pos, Parser.STAR_IDENT(new_any M_special (kind, name) spaces pos)
- | RAW_IDENT(kind, name, pos) -> pos, Parser.RAW_IDENT(new_any M_special (kind, name) spaces pos)
- | RAW_IDENT_PAREN(kind, name, pos) -> pos, Parser.RAW_IDENT_PAREN(new_any M_special (kind, name) spaces pos)
- | ARRAYLEN_IDENT(kind, name, pos) -> pos, Parser.ARRAYLEN_IDENT(new_any M_special (kind, name) spaces pos)
- | SUB_WITH_PROTO(proto, pos) -> pos, Parser.SUB_WITH_PROTO(new_any M_special proto spaces pos)
- | FUNC_DECL_WITH_PROTO(fq, name, proto, pos) -> pos, Parser.FUNC_DECL_WITH_PROTO(new_any M_special (fq, name, proto) spaces pos)
-
- | NEW(pos) -> pos, Parser.NEW(new_any M_special () spaces pos)
- | COMPARE_OP(s, pos) -> pos, Parser.COMPARE_OP(new_any M_special s spaces pos)
- | COMPARE_OP_STR(s, pos) -> pos, Parser.COMPARE_OP_STR(new_any M_special s spaces pos)
- | EQ_OP(s, pos) -> pos, Parser.EQ_OP(new_any M_special s spaces pos)
- | EQ_OP_STR(s, pos) -> pos, Parser.EQ_OP_STR(new_any M_special s spaces pos)
- | ASSIGN(s, pos) -> pos, Parser.ASSIGN(new_any M_special s spaces pos)
- | FOR(s, pos) -> pos, Parser.FOR(new_any M_special s spaces pos)
-
- | DOTDOT(s, pos) -> pos, Parser.DOTDOT(new_any M_special s spaces pos)
- | MULT(s, pos) -> pos, Parser.MULT(new_any M_special s spaces pos)
- | BIT_SHIFT(s, pos) -> pos, Parser.BIT_SHIFT(new_any M_special s spaces pos)
- | PLUS(s, pos) -> pos, Parser.PLUS(new_any M_special s spaces pos)
- | ONE_SCALAR_PARA(s, pos) -> pos, Parser.ONE_SCALAR_PARA(new_any M_special s spaces pos)
- | MY_OUR(s, pos) -> pos, Parser.MY_OUR(new_any M_special s spaces pos)
-
- | EOF (pos) -> pos, Parser.EOF (new_any M_special () spaces pos)
- | IF (pos) -> pos, Parser.IF (new_any M_special () spaces pos)
- | ELSIF (pos) -> pos, Parser.ELSIF (new_any M_special () spaces pos)
- | ELSE (pos) -> pos, Parser.ELSE (new_any M_special () spaces pos)
- | UNLESS (pos) -> pos, Parser.UNLESS (new_any M_special () spaces pos)
- | DO (pos) -> pos, Parser.DO (new_any M_special () spaces pos)
- | WHILE (pos) -> pos, Parser.WHILE (new_any M_special () spaces pos)
- | UNTIL (pos) -> pos, Parser.UNTIL (new_any M_special () spaces pos)
- | CONTINUE (pos) -> pos, Parser.CONTINUE (new_any M_special () spaces pos)
- | SUB (pos) -> pos, Parser.SUB (new_any M_special () spaces pos)
- | LOCAL (pos) -> pos, Parser.LOCAL (new_any M_special () spaces pos)
- | USE (pos) -> pos, Parser.USE (new_any M_special () spaces pos)
- | PACKAGE (pos) -> pos, Parser.PACKAGE (new_any M_special () spaces pos)
- | BEGIN (pos) -> pos, Parser.BEGIN (new_any M_special () spaces pos)
- | END (pos) -> pos, Parser.END (new_any M_special () spaces pos)
- | AT (pos) -> pos, Parser.AT (new_any M_special () spaces pos)
- | DOLLAR (pos) -> pos, Parser.DOLLAR (new_any M_special () spaces pos)
- | PERCENT (pos) -> pos, Parser.PERCENT (new_any M_special () spaces pos)
- | AMPERSAND (pos) -> pos, Parser.AMPERSAND (new_any M_special () spaces pos)
- | STAR (pos) -> pos, Parser.STAR (new_any M_special () spaces pos)
- | ARRAYLEN (pos) -> pos, Parser.ARRAYLEN (new_any M_special () spaces pos)
- | SEMI_COLON (pos) -> pos, Parser.SEMI_COLON (new_any M_none () spaces pos)
- | PKG_SCOPE (pos) -> pos, Parser.PKG_SCOPE (new_any M_special () spaces pos)
- | PAREN (pos) -> pos, Parser.PAREN (new_any M_special () spaces pos)
- | PAREN_END (pos) -> pos, Parser.PAREN_END (new_any M_special () spaces pos)
- | BRACKET (pos) -> pos, Parser.BRACKET (new_any M_special () spaces pos)
- | BRACKET_END (pos) -> pos, Parser.BRACKET_END (new_any M_special () spaces pos)
- | BRACKET_HASHREF (pos) -> pos, Parser.BRACKET_HASHREF (new_any M_special () spaces pos)
- | ARRAYREF (pos) -> pos, Parser.ARRAYREF (new_any M_special () spaces pos)
- | ARRAYREF_END (pos) -> pos, Parser.ARRAYREF_END (new_any M_special () spaces pos)
- | ARROW (pos) -> pos, Parser.ARROW (new_any M_special () spaces pos)
- | INCR (pos) -> pos, Parser.INCR (new_any M_special () spaces pos)
- | DECR (pos) -> pos, Parser.DECR (new_any M_special () spaces pos)
- | POWER (pos) -> pos, Parser.POWER (new_any M_special () spaces pos)
- | TIGHT_NOT (pos) -> pos, Parser.TIGHT_NOT (new_any M_special () spaces pos)
- | BIT_NEG (pos) -> pos, Parser.BIT_NEG (new_any M_special () spaces pos)
- | REF (pos) -> pos, Parser.REF (new_any M_special () spaces pos)
- | PATTERN_MATCH (pos) -> pos, Parser.PATTERN_MATCH (new_any M_special () spaces pos)
- | PATTERN_MATCH_NOT(pos) -> pos, Parser.PATTERN_MATCH_NOT(new_any M_special () spaces pos)
- | LT (pos) -> pos, Parser.LT (new_any M_special () spaces pos)
- | GT (pos) -> pos, Parser.GT (new_any M_special () spaces pos)
- | BIT_AND (pos) -> pos, Parser.BIT_AND (new_any M_special () spaces pos)
- | BIT_OR (pos) -> pos, Parser.BIT_OR (new_any M_special () spaces pos)
- | BIT_XOR (pos) -> pos, Parser.BIT_XOR (new_any M_special () spaces pos)
- | AND_TIGHT (pos) -> pos, Parser.AND_TIGHT (new_any M_special () spaces pos)
- | OR_TIGHT (pos) -> pos, Parser.OR_TIGHT (new_any M_special () spaces pos)
- | QUESTION_MARK (pos) -> pos, Parser.QUESTION_MARK (new_any M_special () spaces pos)
- | COLON (pos) -> pos, Parser.COLON (new_any M_special () spaces pos)
- | COMMA (pos) -> pos, Parser.COMMA (new_any M_special () spaces pos)
- | CONCAT (pos) -> pos, Parser.CONCAT (new_any M_special () spaces pos)
- | MULT_L_STR (pos) -> pos, Parser.MULT_L_STR (new_any M_special () spaces pos)
- | RIGHT_ARROW (pos) -> pos, Parser.RIGHT_ARROW (new_any M_special () spaces pos)
- | NOT (pos) -> pos, Parser.NOT (new_any M_special () spaces pos)
- | AND (pos) -> pos, Parser.AND (new_any M_special () spaces pos)
- | OR (pos) -> pos, Parser.OR (new_any M_special () spaces pos)
- | XOR (pos) -> pos, Parser.XOR (new_any M_special () spaces pos)
-
- | SPACE _ | CR -> internal_error "raw_token_to_token"
-
-and raw_token_to_token spaces raw_token =
- let _, token = raw_token_to_pos_and_token spaces raw_token in
- token
-
-and raw_interpolated_string_to_tokens l =
- List.map (fun (s, rtok) -> s, concat_spaces [] Space_0 rtok) l
-
-and concat_spaces ret spaces = function
- | CR :: l -> concat_spaces ret Space_cr l
- | SPACE n :: l ->
- let spaces' =
- match spaces with
- | Space_cr -> Space_cr
- | Space_0 -> if n = 1 then Space_1 else Space_n
- | _ -> Space_n
- in
- concat_spaces ret spaces' l
- | [] -> List.rev ret
- | token :: l -> concat_spaces (raw_token_to_pos_and_token spaces token :: ret) Space_0 l
-
-let rec lexbuf2list accu t lexbuf =
- match t lexbuf with
- | EOF pos -> List.rev (EOF pos :: accu)
- | e -> lexbuf2list (e :: accu) t lexbuf
-
-let get_token token lexbuf =
- let tokens = lexbuf2list [] token lexbuf in
- let tokens = concat_bareword_paren [] tokens in
- let tokens = concat_spaces [] Space_0 tokens in
- let tokens = bracket_bareword_is_hashref [] tokens in
- tokens
-
-let next_rule = Stack.create()
-
-
-let putback lexbuf nb = lexbuf.Lexing.lex_curr_pos <- lexbuf.Lexing.lex_curr_pos - nb
-
-let add_a_new_line raw_pos =
- incr current_file_current_line ;
- lpush current_file_lines_starts raw_pos
-
-let here_docs = Queue.create()
-let raw_here_docs = Queue.create()
-let current_here_doc_mark = ref ""
-
-let here_doc_next_line mark =
- let here_doc_ref = ref([], bpos) in
- Queue.push (mark, here_doc_ref) here_docs ;
- here_doc_ref
-let raw_here_doc_next_line mark =
- let here_doc_ref = ref("", bpos) in
- Queue.push (mark, here_doc_ref) raw_here_docs ;
- here_doc_ref
-
-let delimit_char = ref '/'
-let delimit_char_open = ref '('
-let delimit_char_close = ref ')'
-type string_escape_kinds = Double_quote | Qq | Delimited | Here_doc
-let string_escape_kind = ref Double_quote
-let string_quote_escape = ref false
-let string_escape_useful = ref (Left false)
-let not_ok_for_match = ref (-1)
-let string_nestness = ref 0
-let string_is_i18n = ref false
-
-let building_current_interpolated_string = Stack.create()
-let building_current_string = Stack.create()
-let current_string_start_pos = ref 0
-let current_string_start_line = ref 0
-
-let die_in_string lexbuf err = failwith (pos2sfull_with !current_string_start_pos (lexeme_end lexbuf) ^ err)
-let warn_escape_unneeded lexbuf c =
- let s = String.make 1 c in warn [Warn_suggest_simpler] lexbuf ("you can replace \\" ^ s ^ " with " ^ s)
-let next_interpolated toks =
- let r = Stack.top building_current_string in
- Queue.push (!r, toks) (Stack.top building_current_interpolated_string) ;
- r := ""
-
-let raw_ins t lexbuf =
- Stack.push (ref "") building_current_string;
- current_string_start_pos := lexeme_start lexbuf;
- t lexbuf ;
- !(Stack.pop building_current_string), (!current_string_start_pos, lexeme_end lexbuf)
-
-let ins t lexbuf =
- Stack.push (Queue.create()) building_current_interpolated_string ;
- Stack.push (ref "") building_current_string;
- current_string_start_pos := lexeme_start lexbuf;
- t lexbuf ;
- next_interpolated [] ;
- let _ = Stack.pop building_current_string in
- queue2list (Stack.pop building_current_interpolated_string), (!current_string_start_pos, lexeme_end lexbuf)
-
-let raw_ins_to_string t lexbuf =
- let s, pos = raw_ins t lexbuf in
- not_ok_for_match := lexeme_end lexbuf;
- RAW_STRING(s, pos)
-let ins_to_string t lexbuf =
- string_escape_useful := Left false ;
- string_quote_escape := false ;
- let s, pos = ins t lexbuf in
-
- if not !string_is_i18n then
- (match !string_escape_useful, s with
- | Right c, [ _, [] ] ->
- let s = String.make 1 c in
- warn_with_pos [Warn_suggest_simpler] pos ("you can replace \"xxx\\" ^ s ^ "xxx\" with 'xxx" ^ s ^ "xxx', that way you don't need to escape <" ^ s ^ ">")
- | _ ->
- if !string_quote_escape then
- let full_s = String.concat "" (List.map fst s) in
- let nb = string_fold_left (fun nb c ->
- if nb < 0 then nb else
- if c = '(' then nb + 1 else
- if c = ')' then nb - 1 else nb
- ) 0 full_s in
- if nb = 0 then
- warn_with_pos [Warn_suggest_simpler] pos "you can replace \"xxx\\\"xxx\" with qq(xxx\"xxx), that way you don't need to escape <\">"
- );
-
- not_ok_for_match := lexeme_end lexbuf;
- string_is_i18n := false ;
- STRING(s, pos)
-
-let next_s s t lexbuf =
- let r = Stack.top building_current_string in r := !r ^ s ;
- t lexbuf
-let next t lexbuf = next_s (lexeme lexbuf) t lexbuf
-
-let ins_re re_delimited_string lexbuf =
- let s, pos = ins re_delimited_string lexbuf in
- List.iter (fun (s, _) ->
- if str_contains s "[^\\s]" then warn [Warn_suggest_simpler] lexbuf "you can replace [^\\s] with \\S";
- if str_contains s "[^\\w]" then warn [Warn_suggest_simpler] lexbuf "you can replace [^\\w] with \\W"
- ) s ;
- s, pos
-
-let string_interpolate token pre lexbuf =
- let s = lexeme lexbuf in
- let local_lexbuf = Lexing.from_string (pre ^ s ^ " ") in (* add a space to help tokenizing "xxx$$" *)
- local_lexbuf.lex_start_p <- lexbuf.lex_start_p ;
- local_lexbuf.lex_curr_p <- lexbuf.lex_start_p ;
- local_lexbuf.lex_abs_pos <- lexeme_start lexbuf ;
- let l = lexbuf2list [] token local_lexbuf in
- let l = concat_bareword_paren [] l in
- next_interpolated l;
- (Stack.pop next_rule) lexbuf
-
-let ident_type_from_char fq name lexbuf c =
- not_ok_for_match := lexeme_end lexbuf;
- match c with
- | '$' -> SCALAR_IDENT(fq, name, pos lexbuf)
- | '@' -> ARRAY_IDENT (fq, name, pos lexbuf)
- | '%' -> HASH_IDENT (fq, name, pos lexbuf)
- | '&' -> FUNC_IDENT (fq, name, pos lexbuf)
- | '*' -> STAR_IDENT (fq, name, pos lexbuf)
- | _ -> internal_error "ident_type_from_char"
-
-let split_at_two_colons s =
- let i_fq = String.rindex s ':' in
- String.sub s 0 (i_fq - 1), skip_n_char (i_fq + 1) s
-
-let ident_from_lexbuf lexbuf =
- let fq, name = split_at_two_colons (lexeme lexbuf) in
- RAW_IDENT(Some fq, name, pos lexbuf)
-
-let typed_ident_from_lexbuf lexbuf =
- let s = lexeme lexbuf in
- ident_type_from_char None (skip_n_char 1 s) lexbuf s.[0]
-
-let typed_fqident_from_lexbuf lexbuf =
- let s = lexeme lexbuf in
- let fq, name = split_at_two_colons (skip_n_char 1 s) in
- ident_type_from_char (Some fq) name lexbuf s.[0]
-
-let arraylen_ident_from_lexbuf lexbuf =
- not_ok_for_match := lexeme_end lexbuf;
- let s = lexeme lexbuf in
- ARRAYLEN_IDENT(None, skip_n_char 2 s, pos lexbuf)
-
-let arraylen_fqident_from_lexbuf lexbuf =
- let s = lexeme lexbuf in
- let fq, name = split_at_two_colons (skip_n_char 2 s) in
- ARRAYLEN_IDENT(Some fq, name, pos lexbuf)
-
-let check_multi_line_delimited_string opts (start, end_) =
- let check =
- match opts with
- | None -> true
- | Some s -> not (String.contains s 'x') in
- if check then
- if !current_file_current_line <> !current_string_start_line then
- failwith (pos2sfull_with start end_ ^ "multi-line patterns are not allowed (or use /x modifier)")
-
-let hex_in_string lexbuf next_rule s =
- let i =
- try int_of_string ("0x" ^ s)
- with Failure("int_of_string") -> die_in_string lexbuf ("Bad_hex_in_string \"" ^ lexeme lexbuf ^ "\"")
- in
- let s =
- if i < 256 then
- String.make 1 (Char.chr i)
- else
- "\\x{" ^ s ^ "}" in
- next_s s (Stack.pop next_rule) lexbuf
-
-let set_delimit_char lexbuf op =
- let c = lexeme_char lexbuf (String.length op) in
- delimit_char := c;
- match c with
- | '@' -> warn [Warn_complex_expressions] lexbuf ("don't use " ^ op ^ "@...@, replace @ with / ! , or |")
- | ':' -> warn [Warn_complex_expressions] lexbuf ("don't use " ^ op ^ ":...:, replace : with / ! , or |")
- | _ -> ()
-
-let set_delimit_char_open lexbuf op =
- let char_open = lexeme_char lexbuf (String.length op) in
- let char_close =
- match char_open with
- | '(' -> ')'
- | '{' -> '}'
- | _ -> internal_error "set_delimit_char_open"
- in
- if op = "qx" then
- warn [Warn_complex_expressions] lexbuf (Printf.sprintf "don't use qx%c...%c, use `...` instead" char_open char_close)
- else if char_open = '{' then
- warn [Warn_complex_expressions] lexbuf ("don't use " ^ op ^ "{...}, use " ^ op ^ "(...) instead");
- delimit_char_open := char_open;
- delimit_char_close := char_close
-}
-
-let stash = [ '$' '@' '%' '&' '*' ]
-let ident_start = ['a'-'z' 'A'-'Z' '_']
-let ident = ident_start ['0'-'9' 'A'-'Z' 'a'-'z' '_'] *
-let pattern_separator = [ '/' '!' ',' '|' '@' ':' ]
-let pattern_open = [ '(' '{' ]
-let pattern_close = [ ')' '}' ]
-
-let in_string_expr = (ident | (ident? ("::" ident)+)) "->"? (('{' [^ '{' '}' '\n']* '}') | ('[' [^ '[' ']' '\n']* ']'))*
-
-rule token = parse
-| [' ' '\t']+ {
- (* propagate not_ok_for_match when it was set by the previous token *)
- if lexeme_start lexbuf = !not_ok_for_match then not_ok_for_match := lexeme_end lexbuf;
- SPACE(lexeme_end lexbuf - lexeme_start lexbuf)
- }
-| "# perl_checker: " [^ '\n']* { PERL_CHECKER_COMMENT(skip_n_char 16 (lexeme lexbuf), pos lexbuf) }
-| "#-PO: " [^ '\n']* { PO_COMMENT(skip_n_char 1 (lexeme lexbuf), pos lexbuf) }
-| '#' [^ '\n']* { SPACE(1) }
-
-| "\n=" {
- add_a_new_line(lexeme_end lexbuf - 1);
- let _ = ins pod_command lexbuf in token lexbuf
- }
-
-| '\n' {
- add_a_new_line(lexeme_end lexbuf);
- (try
- let (mark, r) = Queue.pop here_docs in
- current_here_doc_mark := mark ;
- r := ins here_doc lexbuf
- with Queue.Empty ->
- try
- let (mark, r) = Queue.pop raw_here_docs in
- current_here_doc_mark := mark ;
- r := raw_ins raw_here_doc lexbuf
- with Queue.Empty -> ());
- CR
- }
-| "->" { ARROW(pos lexbuf) }
-| "++" { INCR(pos lexbuf) }
-| "--" { DECR(pos lexbuf) }
-| "**" { POWER(pos lexbuf) }
-| "!" { TIGHT_NOT(pos lexbuf) }
-| "~" { BIT_NEG(pos lexbuf) }
-| "=~" { PATTERN_MATCH(pos lexbuf) }
-| "!~" { PATTERN_MATCH_NOT(pos lexbuf) }
-| "*" { MULT(lexeme lexbuf, pos lexbuf) }
-| "%" { MULT(lexeme lexbuf, pos lexbuf) }
-| "x" { MULT_L_STR(pos lexbuf) }
-| "+" { PLUS(lexeme lexbuf, pos lexbuf) }
-| "-" { PLUS(lexeme lexbuf, pos lexbuf) }
-| "." { CONCAT(pos lexbuf) }
-| "<<" { BIT_SHIFT(lexeme lexbuf, pos lexbuf) }
-| ">>" { BIT_SHIFT(lexeme lexbuf, pos lexbuf) }
-| "<" { LT(pos lexbuf) }
-| ">" { GT(pos lexbuf) }
-| "<=" | ">=" { COMPARE_OP(lexeme lexbuf, pos lexbuf) }
-| "lt" | "gt" | "le" | "ge" { COMPARE_OP_STR(lexeme lexbuf, pos lexbuf) }
-| "==" | "!=" | "<=>" { EQ_OP(lexeme lexbuf, pos lexbuf) }
-| "eq" | "ne" | "cmp" { EQ_OP_STR(lexeme lexbuf, pos lexbuf) }
-| "&" { BIT_AND(pos lexbuf) }
-| "|" { BIT_OR(pos lexbuf) }
-| "^" { BIT_XOR(pos lexbuf) }
-| "&&" { AND_TIGHT(pos lexbuf) }
-| "||" { OR_TIGHT(pos lexbuf) }
-| ".." { DOTDOT(lexeme lexbuf, pos lexbuf) }
-| "..." { DOTDOT(lexeme lexbuf, pos lexbuf) }
-| "?" { QUESTION_MARK(pos lexbuf) }
-| ":" { COLON(pos lexbuf) }
-| "::" { PKG_SCOPE(pos lexbuf) }
-
-| "=" | "+=" | "-=" | "*=" | ".=" | "|=" | "&=" | "^=" | "||=" | "&&=" { ASSIGN(lexeme lexbuf, pos lexbuf) }
-
-| "<<=" | ">>=" | "**=" {
- warn [Warn_complex_expressions] lexbuf (Printf.sprintf "don't use \"%s\", use the expanded version instead" (lexeme lexbuf)) ;
- ASSIGN(lexeme lexbuf, pos lexbuf)
- }
-
-| "," { COMMA(pos lexbuf) }
-| "=>" { RIGHT_ARROW(pos lexbuf) }
-| "not" { NOT(pos lexbuf) }
-| "and" { AND(pos lexbuf) }
-| "or" { OR(pos lexbuf) }
-| "xor" { XOR(pos lexbuf) }
-
-| "if" { IF(pos lexbuf) }
-| "else" { ELSE(pos lexbuf) }
-| "elsif" { ELSIF(pos lexbuf) }
-| "unless" { UNLESS(pos lexbuf) }
-| "do" { DO(pos lexbuf) }
-| "while" { WHILE(pos lexbuf) }
-| "until" { UNTIL(pos lexbuf) }
-| "foreach" { FOR(lexeme lexbuf, pos lexbuf) }
-| "for" { FOR(lexeme lexbuf, pos lexbuf) }
-| "my" { MY_OUR(lexeme lexbuf, pos lexbuf) }
-| "our" { MY_OUR(lexeme lexbuf, pos lexbuf) }
-| "local" { LOCAL(pos lexbuf) }
-| "continue" { CONTINUE(pos lexbuf) }
-| "sub" { SUB(pos lexbuf) }
-| "package" { PACKAGE(pos lexbuf) }
-| "use" { USE(pos lexbuf) }
-| "BEGIN" { BEGIN(pos lexbuf) }
-| "END" { END(pos lexbuf) }
-| "print" { PRINT(lexeme lexbuf, pos lexbuf) }
-| "printf" { PRINT(lexeme lexbuf, pos lexbuf) }
-| "new" { NEW(pos lexbuf) }
-| "format" { let pos = pos lexbuf in FORMAT(here_doc_next_line ".", pos) }
-| "delete"
-| "defined"
-| "length"
-| "keys"
-| "exists"
-| "shift"
-| "pop"
-| "eval"
-| "ref" { ONE_SCALAR_PARA(lexeme lexbuf, pos lexbuf) }
-
-| "split"
-| "grep" { (* ok_for_match! *) BAREWORD(lexeme lexbuf, pos lexbuf) }
-
-| "print " ['A'-'Z'] ['A'-'Z' '0'-'9']* ['\n' ' '] {
- putback lexbuf 1;
- PRINT_TO_STAR(("print", skip_n_char 6 (lexeme lexbuf)), pos lexbuf)
- }
-| "print $" ident ['\n' ' '] {
- putback lexbuf 1;
- PRINT_TO_SCALAR(("print", skip_n_char 7 (lexeme lexbuf)), pos lexbuf);
- }
-| "printf " ['A'-'Z'] ['A'-'Z' '0'-'9']* ['\n' ' '] {
- putback lexbuf 1;
- PRINT_TO_STAR(("printf", skip_n_char 7 (lexeme lexbuf)), pos lexbuf)
- }
-| "printf $" ident ['\n' ' '] {
- putback lexbuf 1;
- PRINT_TO_SCALAR(("printf", skip_n_char 8 (lexeme lexbuf)), pos lexbuf);
- }
-
-| ident ' '* "=>" { (* needed so that (if => 1) works *)
- let s = lexeme lexbuf in
- let end_ = String.length s - 1 in
- let ident_end = non_rindex_from s (end_ - 2) ' ' in
- putback lexbuf (end_ - ident_end);
- BAREWORD(String.sub s 0 (ident_end+1), pos lexbuf)
- }
-
-| "{" ident "}" { (* needed so that $h{if} works *)
- not_ok_for_match := lexeme_end lexbuf;
- COMPACT_HASH_SUBSCRIPT(skip_n_char_ 1 1 (lexeme lexbuf), pos lexbuf)
- }
-
-| '@' { AT(pos lexbuf) }
-| '$' { DOLLAR(pos lexbuf) }
-| '$' '#' { ARRAYLEN(pos lexbuf) }
-| '%' ['$' '{'] { putback lexbuf 1; PERCENT(pos lexbuf) }
-| '&' ['$' '{'] { putback lexbuf 1; AMPERSAND(pos lexbuf) }
-| '*' ['$' '{'] { putback lexbuf 1; if lexeme_start lexbuf = !not_ok_for_match then MULT("*", pos lexbuf) else STAR(pos lexbuf) }
-
-
-| ';' { SEMI_COLON(pos lexbuf) }
-| '(' { PAREN(pos lexbuf) }
-| '{' { BRACKET(pos lexbuf) }
-| "+{"{ BRACKET_HASHREF(pos lexbuf) }
-| '[' { ARRAYREF(pos lexbuf) }
-| ')' { not_ok_for_match := lexeme_end lexbuf; PAREN_END(pos lexbuf) }
-| '}' { not_ok_for_match := lexeme_end lexbuf; BRACKET_END(pos lexbuf) }
-| ']' { not_ok_for_match := lexeme_end lexbuf; ARRAYREF_END(pos lexbuf) }
-
-| "/" {
- if lexeme_start lexbuf = !not_ok_for_match then MULT("/", pos lexbuf)
- else (
- delimit_char := '/' ;
- current_string_start_line := !current_file_current_line;
- let s, pos = ins_re re_delimited_string lexbuf in
- let opts, _ = raw_ins pattern_options lexbuf in
- check_multi_line_delimited_string (Some opts) pos ;
- PATTERN(s, opts, pos)
- )
- }
-
-| "/=" {
- if lexeme_start lexbuf = !not_ok_for_match then ASSIGN(lexeme lexbuf, pos lexbuf)
- else (
- putback lexbuf 1 ;
- delimit_char := '/' ;
- let s, pos = ins_re re_delimited_string lexbuf in
- let opts, _ = raw_ins pattern_options lexbuf in
- PATTERN(s, opts, pos)
- )
- }
-
-| "m" pattern_separator {
- set_delimit_char lexbuf "m" ;
- current_string_start_line := !current_file_current_line;
- let s, pos = ins_re re_delimited_string lexbuf in
- let opts, _ = raw_ins pattern_options lexbuf in
- check_multi_line_delimited_string (Some opts) pos ;
- PATTERN(s, opts, pos)
-}
-
-| "qr" pattern_separator {
- set_delimit_char lexbuf "qr" ;
- current_string_start_line := !current_file_current_line;
- let s, pos = ins_re re_delimited_string lexbuf in
- let opts, _ = raw_ins pattern_options lexbuf in
- check_multi_line_delimited_string (Some opts) pos ;
- QR_PATTERN(s, opts, pos)
-}
-
-| "qw" pattern_separator {
- set_delimit_char lexbuf "qw" ;
- current_string_start_line := !current_file_current_line;
- let s, pos = raw_ins delimited_string lexbuf in
- warn_with_pos [Warn_complex_expressions] pos (Printf.sprintf "don't use qw%c...%c, use qw(...) instead" !delimit_char !delimit_char) ;
- QUOTEWORDS(s, pos)
-}
-
-| "s" pattern_separator {
- set_delimit_char lexbuf "s" ;
- current_string_start_line := !current_file_current_line;
- let s1, (start, _) = ins_re re_delimited_string lexbuf in
- let s2, (_, end_) = ins delimited_string lexbuf in
- let opts, _ = raw_ins pattern_options lexbuf in
- let pos = start, end_ in
- if String.contains opts 'e' && sum (List.map (fun (s, _) -> count_chars_in_string s '"') s2) > 2 then
- die lexbuf ("do not write so complicated things in the eval part of s///,\n" ^
- "i generate wrong warnings for things like s/xxx/die \"yyy \\\"zzz\\\" \"/") ;
- check_multi_line_delimited_string (Some opts) pos ;
- PATTERN_SUBST(s1, s2, opts, pos)
-}
-
-| "tr" pattern_separator {
- set_delimit_char lexbuf "tr" ;
- current_string_start_line := !current_file_current_line;
- let s1, (start, _) = ins delimited_string lexbuf in
- let s2, (_, end_) = ins delimited_string lexbuf in
- let opts, _ = raw_ins pattern_options lexbuf in
- let pos = start, end_ in
- check_multi_line_delimited_string None pos ;
- PATTERN_SUBST(s1, s2, opts, pos)
-}
-
-| "<<" ident {
- not_ok_for_match := lexeme_end lexbuf;
- HERE_DOC(here_doc_next_line (skip_n_char 2 (lexeme lexbuf)), pos lexbuf)
- }
-| "<<\"" ident "\"" {
- warn_with_pos [Warn_suggest_simpler] (lexeme_start lexbuf + 2, lexeme_end lexbuf) "Don't use <<\"MARK\", use <<MARK instead" ;
- not_ok_for_match := lexeme_end lexbuf;
- HERE_DOC(here_doc_next_line (skip_n_char_ 3 1 (lexeme lexbuf)), pos lexbuf)
- }
-| "<<'" ident "'" {
- not_ok_for_match := lexeme_end lexbuf;
- RAW_HERE_DOC(raw_here_doc_next_line (skip_n_char_ 3 1 (lexeme lexbuf)), pos lexbuf)
- }
-| "<<" ' '+ "'"
-| "<<" ' '+ ident
-| "<<" ' '* '"' {
- failwith (pos2sfull_with (lexeme_start lexbuf + 2) (lexeme_end lexbuf) ^ "No space allowed between \"<<\" and the marker")
- }
-
-| "\\"+ stash
-| "\\" ['0'-'9' 'A'-'Z' 'a'-'z']
-| "\\" ' '* '('
- { lexbuf.lex_curr_pos <- lexbuf.lex_start_pos + 1; REF(pos lexbuf) }
-
-| "sub(" [ '$' '@' '\\' '&' ';' '%' ]* ')' {
- SUB_WITH_PROTO(skip_n_char_ 4 1 (lexeme lexbuf), pos lexbuf)
- }
-
-| "sub" ' '+ ident ' '* '(' [ '$' '@' '\\' '&' ';' '%' ]* ')' {
- (* bloody prototypes, must be caught especially otherwise "($)" is badly tokenized *)
- (* and alas "($@)" is both valid as an expression and a prototype *)
- let s = lexeme lexbuf in
- let ident_start = non_index_from s 3 ' ' in
-
- let proto_start = String.index_from s ident_start '(' in
- let ident_end = non_rindex_from s (proto_start-1) ' ' in
- let ident = String.sub s ident_start (ident_end - ident_start + 1) in
- let prototype = skip_n_char_ (proto_start + 1) 1 s in
-
- FUNC_DECL_WITH_PROTO(None, ident, prototype, pos lexbuf)
- }
-
-| "sub" ' '+ ident ("::" ident)+ ' '* '(' [ '$' '@' '\\' '&' ';' '%' ]* ')' {
- (* bloody prototypes, must be caught especially otherwise "($)" is badly tokenized *)
- (* and alas "($@)" is both valid as an expression and a prototype *)
- let s = lexeme lexbuf in
- let ident_start = non_index_from s 3 ' ' in
-
- let proto_start = String.index_from s ident_start '(' in
- let ident_end = non_rindex_from s (proto_start-1) ' ' in
- let ident = String.sub s ident_start (ident_end - ident_start + 1) in
- let prototype = skip_n_char_ (proto_start + 1) 1 s in
-
- let fq, name = split_at_two_colons ident in
- FUNC_DECL_WITH_PROTO(Some fq, name, prototype, pos lexbuf)
- }
-
-| "$#" ident? ("::" ident)+ { arraylen_fqident_from_lexbuf lexbuf }
-| "$#" ident { arraylen_ident_from_lexbuf lexbuf }
-
-| stash ident? ("::" ident)+ { typed_fqident_from_lexbuf lexbuf }
-| stash ident
-| '$' [^ '{' ' ' '\n' '$']
-| "$^" [^ '{' ' ' '\n'] { typed_ident_from_lexbuf lexbuf }
-
-| "$$" [^ 'a'-'z' 'A'-'Z' '_' '{'] { putback lexbuf 1; SCALAR_IDENT(None, "$", pos lexbuf) }
-
-| stash "::" { putback lexbuf 2; ident_type_from_char None "main" lexbuf (lexeme_char lexbuf 0) }
-
-| ident? ("::" ident)+ { ident_from_lexbuf lexbuf }
-| ident { not_ok_for_match := lexeme_end lexbuf;
- let word = lexeme lexbuf in
- if word = "qx" then die lexbuf "don't use qx{...}, use `...` instead" else
- BAREWORD(word, pos lexbuf) }
-
-| ident ":" { LABEL(lexeme lexbuf, pos lexbuf) }
-
-| '-' [ 'a'-'z' 'A'-'Z' ] [ ' ' '(' ';' ] { putback lexbuf 1; ONE_SCALAR_PARA(lexeme lexbuf, pos lexbuf) }
-
-| ['0'-'9'] ['0'-'9' '_']* '.' ['0'-'9'] ['0'-'9' '_']* ('.' ['0'-'9'] ['0'-'9' '_']*)+
-| 'v' ['0'-'9'] ['0'-'9' '_']* ('.' ['0'-'9'] ['0'-'9' '_']*)*
- {
- not_ok_for_match := lexeme_end lexbuf;
- REVISION(lexeme lexbuf, pos lexbuf)
- }
-
-| ['0'-'9']* '.' ['0'-'9']+ (['e' 'E']['-' '+']?['0'-'9']+)? {
- not_ok_for_match := lexeme_end lexbuf;
- FLOAT(lexeme lexbuf, pos lexbuf)
- }
-| ['0'-'9'] ['0'-'9' '_']* (['e' 'E']['-' '+']?['0'-'9']+)?
-| "0x" ['0'-'9' 'a'-'f' 'A'-'F']+ {
- not_ok_for_match := lexeme_end lexbuf;
- INT(lexeme lexbuf, pos lexbuf)
- }
-
-| 'N' '_'? "(\"" { string_is_i18n := true ; putback lexbuf 2 ; BAREWORD(lexeme lexbuf, pos lexbuf) }
-
-| '"' { ins_to_string string lexbuf }
-| "'" { raw_ins_to_string rawstring lexbuf }
-| '`' { delimit_char := '`';
- current_string_start_line := !current_file_current_line;
- not_ok_for_match := lexeme_end lexbuf;
- let s, pos = ins delimited_string lexbuf in
- check_multi_line_delimited_string None pos ;
- COMMAND_STRING(s, pos) }
-| "q" pattern_open { set_delimit_char_open lexbuf "q"; raw_ins_to_string qstring lexbuf }
-| "qq" pattern_open { set_delimit_char_open lexbuf "qq"; ins_to_string qqstring lexbuf }
-| "qx" pattern_open { set_delimit_char_open lexbuf "qx"; ins_to_string qqstring lexbuf }
-| "qw" pattern_open { set_delimit_char_open lexbuf "qw"; let s, pos = raw_ins qstring lexbuf in QUOTEWORDS(s, pos) }
-
-| "\n__END__" [^ '0'-'9' 'A'-'Z' 'a'-'z' '_']
-| "\n__DATA__" [^ '0'-'9' 'A'-'Z' 'a'-'z' '_']
-| eof { EOF(pos lexbuf) }
-| _ { failwith (Printf.sprintf "%serror tokenizing <<%s>>" (pos2sfull lexbuf) (lexeme lexbuf)) }
-
-and string = parse
-| '"' { () }
-| '\\' { Stack.push string next_rule ; string_escape_kind := Double_quote; string_escape lexbuf }
-| '$' { Stack.push string next_rule ; string_interpolate_scalar lexbuf }
-| '@' { Stack.push string next_rule ; string_interpolate_array lexbuf }
-| '\n' {
- add_a_new_line(lexeme_end lexbuf);
- next string lexbuf
- }
-| "'" { string_escape_useful := Left true ; next string lexbuf }
-| [^ '\n' '\\' '"' '$' '@']+ { next string lexbuf }
-| eof { die_in_string lexbuf "Unterminated_string" }
-
-and delimited_string = parse
-| '\\' { Stack.push delimited_string next_rule ; string_escape_kind := Delimited; string_escape lexbuf }
-| '$' { Stack.push delimited_string next_rule ; delimited_string_interpolate_scalar lexbuf }
-| '@' { Stack.push delimited_string next_rule ; delimited_string_interpolate_array lexbuf }
-| '\n' {
- add_a_new_line(lexeme_end lexbuf);
- next delimited_string lexbuf
- }
-| eof { die_in_string lexbuf "Unterminated_delimited_string" }
-| [ ^ '\n' '\\' '$' '@'] { if lexeme_char lexbuf 0 <> !delimit_char then next delimited_string lexbuf }
-
-and re_delimited_string = parse
-| '\\' { Stack.push re_delimited_string next_rule ; re_string_escape lexbuf }
-| '$' { Stack.push re_delimited_string next_rule ; delimited_string_interpolate_scalar lexbuf }
-| '@' { if lexeme_char lexbuf 0 <> !delimit_char then
- (Stack.push re_delimited_string next_rule ; delimited_string_interpolate_array lexbuf) }
-| '\n' {
- add_a_new_line(lexeme_end lexbuf);
- next re_delimited_string lexbuf
- }
-| eof { die_in_string lexbuf "Unterminated_delimited_string" }
-| [ ^ '\n' '\\' '$' '@'] { if lexeme_char lexbuf 0 <> !delimit_char then next re_delimited_string lexbuf }
-
-and rawstring = parse
-| ''' { () }
-| '\n' {
- add_a_new_line(lexeme_end lexbuf);
- next rawstring lexbuf
- }
-| '\\' { next rawstring lexbuf }
-| "\\'" { next_s "'" rawstring lexbuf }
-| [^ '\n' ''' '\\']+ { next rawstring lexbuf }
-| eof { die_in_string lexbuf "Unterminated_rawstring" }
-
-and qqstring = parse
-| pattern_close {
- if lexeme_char lexbuf 0 = !delimit_char_close then
- if !string_nestness <> 0 then (decr string_nestness; next qqstring lexbuf)
- else ()
- else next qstring lexbuf
- }
-| pattern_open {
- if lexeme_char lexbuf 0 = !delimit_char_open then incr string_nestness;
- next qqstring lexbuf
- }
-| '\\' { Stack.push qqstring next_rule ; string_escape_kind := Qq; string_escape lexbuf }
-| '$' { Stack.push qqstring next_rule ; string_interpolate_scalar lexbuf }
-| '@' { Stack.push qqstring next_rule ; string_interpolate_array lexbuf }
-| '\n' {
- add_a_new_line(lexeme_end lexbuf);
- next qqstring lexbuf
- }
-| [^ '\n' '(' ')' '{' '}' '\\' '$' '@']+ { next qqstring lexbuf }
-| eof { die_in_string lexbuf "Unterminated_qqstring" }
-
-and qstring = parse
-| pattern_close {
- if lexeme_char lexbuf 0 = !delimit_char_close then
- if !string_nestness <> 0 then (decr string_nestness ; next qstring lexbuf)
- else ()
- else next qstring lexbuf
- }
-| pattern_open {
- if lexeme_char lexbuf 0 = !delimit_char_open then incr string_nestness;
- next qstring lexbuf
- }
-| '\n' {
- add_a_new_line(lexeme_end lexbuf);
- next qstring lexbuf
- }
-| [^ '\n' '(' ')' '{' '}']+ { next qstring lexbuf }
-| eof { die_in_string lexbuf "Unterminated_qstring" }
-
-and here_doc = parse
-| '\\' { Stack.push here_doc next_rule ; string_escape_kind := Here_doc; string_escape lexbuf }
-| '$' { Stack.push here_doc next_rule ; string_interpolate_scalar lexbuf }
-| '@' { Stack.push here_doc next_rule ; string_interpolate_array lexbuf }
-| [ ^ '\n' '\\' '$' '@' ]* {
- let s = lexeme lexbuf in
- if chomps s <> !current_here_doc_mark
- then next_s s here_doc lexbuf
- else if s <> !current_here_doc_mark then warn_with_pos [Warn_traps] (pos lexbuf) "Trailing spaces after HERE-document mark"
- }
-| '\n' {
- add_a_new_line(lexeme_end lexbuf);
- next here_doc lexbuf
- }
-| eof { die_in_string lexbuf "Unterminated_here_doc" }
-
-and raw_here_doc = parse
-| [ ^ '\n' ]* {
- let s = lexeme lexbuf in
- if chomps s <> !current_here_doc_mark
- then next_s s raw_here_doc lexbuf
- else if s <> !current_here_doc_mark then warn_with_pos [Warn_traps] (pos lexbuf) "Trailing spaces after HERE-document mark"
- }
-| '\n' {
- add_a_new_line(lexeme_end lexbuf);
- next raw_here_doc lexbuf
- }
-| eof { die_in_string lexbuf "Unterminated_raw_here_doc" }
-
-
-and string_escape = parse
-| ['0'-'9'] { string_escape_useful := Left true; next_s (String.make 1 (Char.chr (int_of_string (lexeme lexbuf)))) (Stack.pop next_rule) lexbuf }
-| 'n' { string_escape_useful := Left true; next_s "\n" (Stack.pop next_rule) lexbuf }
-| 't' { string_escape_useful := Left true; next_s "\t" (Stack.pop next_rule) lexbuf }
-| "x{" [^ '}']* '}' { string_escape_useful := Left true; hex_in_string lexbuf next_rule (skip_n_char_ 2 1 (lexeme lexbuf)) }
-| 'x' [^ '{'] _ { string_escape_useful := Left true; hex_in_string lexbuf next_rule (skip_n_char 1 (lexeme lexbuf)) }
-| '\n' { die lexbuf "do not use \"\\\" before end-of-line, it's useless and generally bad" }
-| '\\'{ next_s "\\" (Stack.pop next_rule) lexbuf }
-| 'Q' {
- warn [Warn_complex_expressions] lexbuf ("don't use \\Q, use quotemeta instead");
- string_escape_useful := Left true; next_s ("\\" ^ lexeme lexbuf) (Stack.pop next_rule) lexbuf }
-| ['b' 'f' 'a' 'r'] { string_escape_useful := Left true; next_s ("\\" ^ lexeme lexbuf) (Stack.pop next_rule) lexbuf }
-| ['$' '@' '%' '{' '[' ':'] {
- if !string_escape_useful = Left false then string_escape_useful := Right (lexeme_char lexbuf 0) ;
- next_s (lexeme lexbuf) (Stack.pop next_rule) lexbuf
- }
-| _ {
- let c = lexeme_char lexbuf 0 in
- (match !string_escape_kind with
- | Double_quote ->
- if c <> '"' then
- warn_escape_unneeded lexbuf c
- else (
- if !string_escape_useful = Left false then string_escape_useful := Right c ;
- string_quote_escape := true
- )
- | Qq -> if c <> !delimit_char_open && c <> !delimit_char_close then warn_escape_unneeded lexbuf c
- | Here_doc -> warn_escape_unneeded lexbuf c
- | Delimited -> if c = !delimit_char then
- warn [Warn_suggest_simpler] lexbuf ("change the delimit character " ^ String.make 1 !delimit_char ^ " to get rid of this escape")
- else warn_escape_unneeded lexbuf c);
- let s = if c = '"' then String.make 1 c else "\\" ^ String.make 1 c in
- next_s s (Stack.pop next_rule) lexbuf
- }
-
-and re_string_escape = parse
-| ['0'-'9'] { next_s (String.make 1 (Char.chr (int_of_string (lexeme lexbuf)))) (Stack.pop next_rule) lexbuf }
-| '\\'{ next_s "\\" (Stack.pop next_rule) lexbuf }
-| 'n' { next_s "\n" (Stack.pop next_rule) lexbuf }
-| 't' { next_s "\t" (Stack.pop next_rule) lexbuf }
-| "x{" [^ '}']* '}' { hex_in_string lexbuf next_rule (skip_n_char_ 2 1 (lexeme lexbuf)) }
-| 'x' [^ '{'] _ { hex_in_string lexbuf next_rule (skip_n_char 1 (lexeme lexbuf)) }
-| '\n' { die lexbuf "do not use \"\\\" before end-of-line, it's useless and generally bad" }
-| ['r' 'b' 'f' '$' '@' '%' 's' 'S' 'd' 'D' 'w' 'W' 'Q' 'E' 'b' 'Z' 'z' '^' '.' '*' '+' '?' '[' ']' '(' ')' '|' '{' '}' '-' ':'] {
- next_s ("\\" ^ lexeme lexbuf) (Stack.pop next_rule) lexbuf
- }
-| _ {
- let c = lexeme_char lexbuf 0 in
- if c = !delimit_char then
- warn [Warn_suggest_simpler] lexbuf ("change the delimit character " ^ String.make 1 !delimit_char ^ " to get rid of this escape")
- else warn_escape_unneeded lexbuf c ;
- next_s ("\\" ^ lexeme lexbuf) (Stack.pop next_rule) lexbuf
- }
-
-and string_interpolate_scalar = parse
-| '$' ident
-| ['0'-'9']
-| '{' [^ '{' '}']* '}'
-| in_string_expr
-| [^ '{' '}' ' ' '\n' '"'] { (* eg: $! $$ *)
- string_interpolate token "$" lexbuf
- }
-
-| "{"
-| ident "->"? '{'
-| '"' { putback lexbuf 1; next_s "$" (Stack.pop next_rule) lexbuf }
-| eof { next_s "$" (Stack.pop next_rule) lexbuf }
-| _ { warn [Warn_strange] lexbuf (Printf.sprintf "weird \"%s\" in string" (lexeme lexbuf)); next_s ("$" ^ lexeme lexbuf) (Stack.pop next_rule) lexbuf }
-
-and delimited_string_interpolate_scalar = parse (* needed for delimited string like m!foo$! where $! should not be taken as is... *)
-| '$' ident
-| ['0'-'9']
-| '{' [^ '{' '}']* '}'
-| (ident | (ident? ("::" ident)+)) "->"? ('{' [^ '{' '}' '\n']* '}')*
-| (ident | (ident? ("::" ident)+)) "->"? (('{' [^ '{' '}' '\n']* '}') | ('[' ('$' ident | ['0'-'9']+) ']'))*
- {
- string_interpolate token "$" lexbuf
- }
-
-| (ident | (ident? ("::" ident)+)) "->"? (('{' [^ '{' '}' '\n']* '}') | ('[' ['$' '0'-'9'] [^ '[' ']' '\n']* ']'))*
- {
- die lexbuf (Printf.sprintf "I really can't handle this, [xxx] can be indexing or not based on stellar position :-(")
- }
-
-| "{"
-| ident "->"? '{'
-| eof { next_s "$" (Stack.pop next_rule) lexbuf }
-| _ {
- let c = lexeme_char lexbuf 0 in
- if c <> !delimit_char && c <> '|' && c<>')' && c<>'/' && c<>' ' then warn [Warn_strange] lexbuf (Printf.sprintf "weird \"%s\" in string" (lexeme lexbuf));
- putback lexbuf 1;
- next_s "$" (Stack.pop next_rule) lexbuf
- }
-
-and string_interpolate_array = parse
-| '$' ident
-| '{' [^ '{' '}']* '}'
-| in_string_expr { string_interpolate token "@" lexbuf }
-
-| [ '@' '*' '<' '>' ']' '.' '(' ' ' ] { next_s ("@" ^ lexeme lexbuf) (Stack.pop next_rule) lexbuf }
-| '"' { putback lexbuf 1; next_s "@" (Stack.pop next_rule) lexbuf }
-| eof { next_s "@" (Stack.pop next_rule) lexbuf }
-| _ { warn [Warn_strange] lexbuf (Printf.sprintf "weird \"%s\" in string" (lexeme lexbuf)); next_s ("@" ^ lexeme lexbuf) (Stack.pop next_rule) lexbuf }
-
-and delimited_string_interpolate_array = parse
-| '$' ident
-| '{' [^ '{' '}']* '}'
-| in_string_expr
- { string_interpolate token "@" lexbuf }
-
-| [ '@' '*' '<' '>' ']' '.' '(' ' ' ] { next_s ("@" ^ lexeme lexbuf) (Stack.pop next_rule) lexbuf }
-| eof { next_s "@" (Stack.pop next_rule) lexbuf }
-| _ {
- let c = lexeme_char lexbuf 0 in
- if c <> !delimit_char then warn [Warn_strange] lexbuf (Printf.sprintf "weird \"%s\" in string" (lexeme lexbuf));
- putback lexbuf 1;
- next_s "@" (Stack.pop next_rule) lexbuf
- }
-
-and pattern_options = parse
-| [ 'g' 'i' 'm' 'o' 's' 'x' 'e' 'd' ] { next pattern_options lexbuf }
-| _ { putback lexbuf 1; () }
-
-and pod_command = parse
-| [^ '\n' ]+ {
- let s = lexeme lexbuf in
- let command = String.sub s 0 (try String.index s ' ' with Not_found -> String.length s) in
- match command with
- | "cut" ->
- if !(Stack.top building_current_string) = "" then
- failwith(pos2sfull lexbuf ^ "found POD command \"=cut\" but it is not a POD block")
- | "head1" | "head2" | "head3" | "head4" | "over" | "item" | "back" | "pod" | "begin" | "end" | "for" ->
- next pod lexbuf
- | s -> failwith(pos2sfull lexbuf ^ "unknown POD command \"" ^ s ^ "\"")
- }
-| _ { failwith(pos2sfull lexbuf ^ "POD command expected") }
-
-and pod = parse
-| "\n=" {
- add_a_new_line(lexeme_end lexbuf - 1);
- next pod_command lexbuf
- }
-| "\n" [^ '=' '\n'] [^ '\n']*
-| "\n" {
- add_a_new_line(lexeme_end lexbuf);
- next pod lexbuf
- }
-| eof
-| _ { failwith(pos2sfull_with !current_string_start_pos (lexeme_end lexbuf) ^ "POD block still open") }
diff --git a/perl_checker.src/parser.mly b/perl_checker.src/parser.mly
deleted file mode 100644
index a9bf396..0000000
--- a/perl_checker.src/parser.mly
+++ /dev/null
@@ -1,500 +0,0 @@
-%{ (* -*- caml -*- *)
- open Types
- open Common
- open Parser_helper
-
- let parse_error msg = die_rule msg
- let prog_ref = ref None
- let to_String e = Parser_helper.to_String (some !prog_ref) e
- let from_PATTERN e = Parser_helper.from_PATTERN (some !prog_ref) e
- let from_PATTERN_SUBST e = Parser_helper.from_PATTERN_SUBST (some !prog_ref) e
-%}
-
-
-%token <unit Types.any_spaces_pos> EOF
-%token <string Types.any_spaces_pos> NUM RAW_STRING BAREWORD BAREWORD_PAREN REVISION COMMENT POD LABEL PO_COMMENT PERL_CHECKER_COMMENT ONE_SCALAR_PARA
-%token <(string * string) Types.any_spaces_pos> PRINT_TO_STAR PRINT_TO_SCALAR
-%token <string Types.any_spaces_pos> QUOTEWORDS COMPACT_HASH_SUBSCRIPT
-%token <(string * Types.raw_pos) Types.any_spaces_pos> RAW_HERE_DOC
-%token <(string * ((int * int) * token) list) list Types.any_spaces_pos> STRING COMMAND_STRING
-%token <((string * ((int * int) * token) list) list * Types.raw_pos) Types.any_spaces_pos> HERE_DOC FORMAT
-
-%token <((string * ((int * int) * token) list) list * string) Types.any_spaces_pos> PATTERN QR_PATTERN
-%token <((string * ((int * int) * token) list) list * (string * ((int * int) * token) list) list * string) Types.any_spaces_pos> PATTERN_SUBST
-
-%token <(string option * string) Types.any_spaces_pos> SCALAR_IDENT ARRAY_IDENT HASH_IDENT FUNC_IDENT STAR_IDENT RAW_IDENT RAW_IDENT_PAREN ARRAYLEN_IDENT
-%token <string Types.any_spaces_pos> SUB_WITH_PROTO
-%token <(string option * string * string) Types.any_spaces_pos> FUNC_DECL_WITH_PROTO
-
-%token <string Types.any_spaces_pos> FOR PRINT
-%token <unit Types.any_spaces_pos> NEW
-%token <string Types.any_spaces_pos> COMPARE_OP COMPARE_OP_STR EQ_OP EQ_OP_STR
-%token <string Types.any_spaces_pos> ASSIGN MY_OUR
-
-%token <unit Types.any_spaces_pos> IF ELSIF ELSE UNLESS DO WHILE UNTIL CONTINUE SUB LOCAL
-%token <unit Types.any_spaces_pos> USE PACKAGE BEGIN END
-%token <unit Types.any_spaces_pos> AT DOLLAR PERCENT AMPERSAND STAR ARRAYLEN
-%token <unit Types.any_spaces_pos> SEMI_COLON PKG_SCOPE
-%token <unit Types.any_spaces_pos> PAREN PAREN_END
-%token <unit Types.any_spaces_pos> BRACKET BRACKET_END BRACKET_HASHREF
-%token <unit Types.any_spaces_pos> ARRAYREF ARRAYREF_END
-
-%token <unit Types.any_spaces_pos> ARROW
-%token <unit Types.any_spaces_pos> INCR DECR
-%token <unit Types.any_spaces_pos> POWER
-%token <unit Types.any_spaces_pos> TIGHT_NOT BIT_NEG REF
-%token <unit Types.any_spaces_pos> PATTERN_MATCH PATTERN_MATCH_NOT
-%token <string Types.any_spaces_pos> MULT
-%token <string Types.any_spaces_pos> PLUS
-%token <string Types.any_spaces_pos> BIT_SHIFT
-%token <unit Types.any_spaces_pos> LT GT CONCAT MULT_L_STR
-%token <unit Types.any_spaces_pos> BIT_AND
-%token <unit Types.any_spaces_pos> BIT_OR BIT_XOR
-%token <unit Types.any_spaces_pos> AND_TIGHT
-%token <unit Types.any_spaces_pos> OR_TIGHT
-%token <string Types.any_spaces_pos> DOTDOT
-%token <unit Types.any_spaces_pos> QUESTION_MARK COLON
-%token <unit Types.any_spaces_pos> COMMA RIGHT_ARROW
-%token <unit Types.any_spaces_pos> NOT
-%token <unit Types.any_spaces_pos> AND
-%token <unit Types.any_spaces_pos> OR XOR
-
-%nonassoc PREC_LOW
-%nonassoc LOOPEX
-
-%right OR XOR
-%right AND
-%right NOT
-%nonassoc LSTOP
-%left COMMA RIGHT_ARROW
-
-%right ASSIGN
-%right QUESTION_MARK COLON
-%nonassoc DOTDOT
-%left OR_TIGHT
-%left AND_TIGHT
-%left BIT_OR BIT_XOR
-%left BIT_AND
-%nonassoc EQ_OP EQ_OP_STR
-%nonassoc LT GT COMPARE_OP COMPARE_OP_STR
-%nonassoc UNIOP ONE_SCALAR_PARA
-%left BIT_SHIFT
-%left PLUS CONCAT
-%left MULT MULT_L_STR
-%left PATTERN_MATCH PATTERN_MATCH_NOT
-%right TIGHT_NOT BIT_NEG REF UNARY_MINUS
-%right POWER
-%nonassoc INCR DECR
-%left ARROW
-
-%nonassoc PAREN_END
-%left PAREN PREC_HIGH
-%left ARRAYREF BRACKET
-
-%type <Types.fromparser list> prog
-%type <prio_expr_spaces_pos> expr term
-%type <fromparser any_spaces_pos> scalar bracket_subscript variable restricted_subscripted
-
-%start prog
-
-
-%%
-prog: lines EOF {fst $1.any}
-
-lines: /* A collection of "lines" in the program */
-| { default_esp ([], true) }
-| sideff { new_1esp ([$1.any], false) $1 }
-| line lines { if fst $2.any <> [] then mcontext_check_none "value is dropped" $1.any $1; new_esp $2.mcontext ($1.any @ fst $2.any, snd $2.any) $1 $2 }
-
-line:
-| decl { new_1esp [$1.any] $1 }
-| if_then_else { new_1esp [$1.any] $1 }
-| loop { new_1esp [$1.any] $1 }
-| LABEL { sp_cr($1); new_1esp [Label $1.any] $1 }
-| PERL_CHECKER_COMMENT {sp_p($1); new_1esp [Perl_checker_comment($1.any, get_pos $1)] $1 }
-| semi_colon {warn_rule [Warn_white_space] "unneeded \";\""; new_1esp [Semi_colon] $1 }
-| sideff semi_colon {new_1esp [$1.any ; Semi_colon] $1 }
-| BRACKET lines BRACKET_END {new_esp $2.mcontext [lines_to_Block $2 $3] $1 $3}
-
-if_then_else: /* Real conditional expressions */
-| IF PAREN expr PAREN_END BRACKET lines BRACKET_END elsif else_ {sp_p($1); sp_n($2); sp_0($3); sp_0_or_cr($4); sp_p($5); mcontext_check M_bool $3; to_Call_op (if $9.any = [] then M_none else mcontext_lmerge ($6.mcontext :: mcontext_lmaybe $8 @ [$9.mcontext])) "if" (prio_lo P_loose $3 :: lines_to_Block $6 $7 :: $8.any @ $9.any) $1 $9}
-| UNLESS PAREN expr PAREN_END BRACKET lines BRACKET_END elsif else_ {sp_p($1); sp_n($2); sp_0($3); sp_0_or_cr($4); sp_p($5); mcontext_check M_bool $3; check_unless_else $8 $9; to_Call_op M_none "unless" (prio_lo P_loose $3 :: lines_to_Block $6 $7 :: $8.any @ $9.any) $1 $9}
-
-elsif:
-| {default_esp []}
-| ELSIF PAREN expr PAREN_END BRACKET lines BRACKET_END elsif {sp_p($1); sp_n($2); sp_0($3); sp_0_or_cr($4); sp_p($5); mcontext_check M_bool $3; new_esp (mcontext_lmerge ($6.mcontext :: mcontext_lmaybe $8)) (prio_lo P_loose $3 :: lines_to_Block $6 $7 :: $8.any) $1 $8}
-
-else_:
-| { default_esp [] }
-| ELSE BRACKET lines BRACKET_END {sp_p($1); sp_n($2); new_esp $3.mcontext [lines_to_Block $3 $4] $1 $4}
-
-loop:
-| WHILE PAREN expr PAREN_END BRACKET lines BRACKET_END cont {sp_p($1); sp_n($2); sp_0($3); sp_0_or_cr($4); sp_p($5); mcontext_check M_bool $3; to_Call_op M_none "while" [ prio_lo P_loose $3; lines_to_Block $6 $7 ] $1 $8}
-| UNTIL PAREN expr PAREN_END BRACKET lines BRACKET_END cont {sp_p($1); sp_n($2); sp_0($3); sp_0_or_cr($4); sp_p($5); mcontext_check M_bool $3; to_Call_op M_none "until" [ prio_lo P_loose $3; lines_to_Block $6 $7 ] $1 $8}
-| FOR PAREN expr_or_empty semi_colon expr_or_empty semi_colon expr_or_empty PAREN_END BRACKET lines BRACKET_END {sp_p($1); check_for($1); sp_n($2); sp_0($3); sp_p($5); sp_p($7); sp_0($8); sp_n($9); to_Call_op M_none "for" [ $3.any; $5.any; $7.any; lines_to_Block $10 $11 ] $1 $11}
-| FOR SCALAR_IDENT PAREN expr PAREN_END BRACKET lines BRACKET_END cont { warn_rule [Warn_normalized_expressions] "don't use for without \"my\"ing the iteration variable"; sp_p($1); sp_0($4); sp_0_or_cr($5); sp_p($6); mcontext_check M_list $4; to_Call_op M_none "foreach" [ prio_lo P_loose $4; lines_to_Block $7 $8 ] $1 $9}
-| FOR PAREN expr PAREN_END BRACKET lines BRACKET_END cont {sp_p($1); sp_n($2); sp_0($3); sp_0_or_cr($4); sp_p($5); mcontext_check M_list $3; check_for_foreach $1 $3; to_Call_op M_none "foreach" [ prio_lo P_loose $3; lines_to_Block $6 $7 ] $1 $8}
-| for_my lines BRACKET_END cont { to_Call_op M_none "foreach my" ($1.any @ [ lines_to_Block $2 $3 ]) $1 $4}
-
-for_my:
-| FOR MY_OUR SCALAR_IDENT PAREN expr PAREN_END BRACKET {sp_p($1); check_my($2); check_foreach($1); sp_n($4); sp_0($5); sp_0_or_cr($6); sp_p($7); new_esp M_none [ My_our($2.any, [I_scalar, snd $3.any], get_pos $3); prio_lo P_loose $5 ] $1 $7}
-
-
-cont: /* Continue blocks */
-| {default_esp ()}
-| CONTINUE BRACKET lines BRACKET_END {sp_p($1); sp_n($2); check_block_lines $3 $4; new_esp $3.mcontext () $1 $4}
-
-sideff: /* An expression which may have a side-effect */
-| expr { new_1esp $1.any.expr $1 }
-| expr IF expr {sp_p($2); sp_p($3); mcontext_check M_bool $3; call_op_if_infix (prio_lo P_loose $1) (prio_lo P_loose $3) $1 $3}
-| expr UNLESS expr {sp_p($2); sp_p($3); mcontext_check M_bool $3; call_op_unless_infix (prio_lo P_loose $1) (prio_lo P_loose $3) $1 $3}
-| expr WHILE expr {sp_p($2); sp_p($3); mcontext_check M_bool $3; to_Call_op M_none "while infix" [ prio_lo P_loose $1 ; prio_lo P_loose $3 ] $1 $3}
-| expr UNTIL expr {sp_p($2); sp_p($3); mcontext_check M_bool $3; to_Call_op M_none "until infix" [ prio_lo P_loose $1 ; prio_lo P_loose $3 ] $1 $3}
-| expr FOR expr {sp_p($2); sp_p($3); mcontext_check M_list $3; check_foreach($2); to_Call_op M_none "for infix" [ prio_lo P_loose $1 ; prio_lo P_loose $3 ] $1 $3}
-
-decl:
-| FORMAT BAREWORD ASSIGN {to_Call_op M_none "format" [Raw_string($2.any, get_pos $2) ; to_String false (new_1esp (fst $1.any) $1)] $1 $3}
-| FORMAT ASSIGN {new_esp M_none Too_complex $1 $2}
-| func_decl semi_colon {if snd $1.any = None then die_rule "there is no need to pre-declare in Perl!" else (warn_rule [Warn_normalized_expressions] "please don't use prototype pre-declaration" ; new_esp M_special Too_complex $1 $2) }
-| func_decl BRACKET BRACKET_END {sp_n($2); sp_0_or_cr($3); let name, proto = $1.any in new_esp M_none (sub_declaration (name, proto) [] Real_sub_declaration) $1 $3}
-| func_decl BRACKET lines BRACKET_END {sp_n($2); check_block_lines $3 $4; new_esp M_none (sub_declaration $1.any (fst $3.any) Real_sub_declaration) $1 $4}
-| func_decl BRACKET BRACKET expr BRACKET_END BRACKET_END {sp_n($2); sp_p($3); sp_p($4); sp_p($5); check_block_expr false Undef $5 $6; new_esp M_none (sub_declaration $1.any [hash_ref $4] Real_sub_declaration) $1 $6}
-| func_decl BRACKET BRACKET expr BRACKET_END semi_colon BRACKET_END {sp_n($2); sp_p($3); sp_p($4); sp_p($5); check_block_expr true Semi_colon $6 $7; new_esp M_none (sub_declaration $1.any [hash_ref $4; Semi_colon] Real_sub_declaration) $1 $7}
-| PACKAGE word semi_colon {sp_0_or_cr($1); sp_1($2); new_esp M_none (Package $2.any) $1 $3}
-| BEGIN BRACKET lines BRACKET_END {sp_0_or_cr($1); sp_1($2); new_esp M_none (Sub_declaration(Ident(None, "BEGIN", get_pos $1), None, lines_to_Block $3 $4, Glob_assign)) $1 $4}
-| END BRACKET lines BRACKET_END {sp_0_or_cr($1); sp_1($2); new_esp M_none (Sub_declaration(Ident(None, "END", get_pos $1), None, lines_to_Block $3 $4, Glob_assign)) $1 $4}
-| use {$1}
-
-use:
-| use_word listexpr semi_colon {sp_n($2); new_esp M_none (Use($1.any, $2.any.expr)) $1 $3}
-| use_revision word_paren PAREN listexpr PAREN_END {sp_0($4); sp_0_or_cr($5); new_esp M_none (Use($2.any, $4.any.expr)) $1 $5}
-
-use_word:
-| use_revision word comma {new_esp M_none $2.any $1 $3}
-| use_revision word {new_esp M_none $2.any $1 $2}
-| use_revision {new_1esp Undef $1 }
-
-use_revision:
-| USE REVISION comma {$1}
-| USE REVISION {$1}
-| USE {$1}
-
-func_decl:
-| SUB word { new_esp M_none ($2.any, None) $1 $2}
-| SUB ONE_SCALAR_PARA { new_esp M_none (Ident(None, $2.any, get_pos $2), None) $1 $2}
-| SUB BAREWORD_PAREN PAREN PAREN_END { warn_rule [Warn_white_space] "remove carriage return between \"sub\" and the function name"; new_esp M_none (Ident(None, $2.any, get_pos $2), Some "") $1 $4 }
-| FUNC_DECL_WITH_PROTO {new_1esp (Ident(fst3 $1.any, snd3 $1.any, get_pos $1), Some (ter3 $1.any)) $1 }
-
-listexpr: /* Basic list expressions */
-| %prec PREC_LOW { default_pesp P_tok []}
-| argexpr %prec PREC_LOW {$1}
-
-expr: /* Ordinary expressions; logical combinations */
-| expr AND expr {sp_p($2); sp_p($3); mcontext_check M_bool $1; mcontext_check_none "value should be dropped" [$3.any.expr] $3; to_Call_op_ M_none P_and "and" [ prio_lo P_and $1; prio_lo_after P_and $3 ] $1 $3}
-| expr OR expr {sp_p($2); sp_p($3); mcontext_check M_bool $1; mcontext_check_none "value should be dropped" [$3.any.expr] $3; to_Call_op_ M_none P_or "or" [ prio_lo P_or $1; prio_lo_after P_or $3 ] $1 $3}
-| argexpr %prec PREC_LOW { new_1pesp $1.any.priority (List $1.any.expr) $1 }
-
-argexpr: /* Expressions are a list of terms joined by commas */
-| argexpr comma { new_pesp $1.mcontext P_comma $1.any.expr $1 $2}
-| bareword RIGHT_ARROW term {if not_simple ($3.any.expr) then sp_p($3); new_pesp (mtuple_context_concat M_string $3.mcontext) P_comma (followed_by_comma [$1.any] false @ [$3.any.expr]) $1 $3}
-| bareword RIGHT_ARROW BRACKET expr BRACKET_END {sp_p($3); sp_p($5); new_pesp (mtuple_context_concat M_string (M_ref M_hash)) P_comma (followed_by_comma [$1.any] false @ [ hash_ref $4 ]) $1 $5}
-| argexpr comma term {prio_lo_check P_comma $1.any.priority $1.pos (last $1.any.expr); if not_simple ($3.any.expr) then sp_p($3); new_pesp (mtuple_context_concat $1.mcontext $3.mcontext) P_comma (followed_by_comma $1.any.expr $2.any @ [$3.any.expr]) $1 $3}
-| argexpr comma BRACKET expr BRACKET_END {sp_p($3); sp_p($5); new_pesp (mtuple_context_concat $1.mcontext (M_ref M_hash)) P_comma (followed_by_comma $1.any.expr $2.any @ [ hash_ref $4 ]) $1 $5}
-| term %prec PREC_LOW { new_1pesp $1.any.priority [$1.any.expr] $1 }
-
-/********************************************************************************/
-term:
-| term
- COMPARE_OP_STR term {sp_p $2; symops P_cmp M_string M_bool $2.any $1 $2 $3}
-| term COMPARE_OP term {sp_p $2; symops P_cmp M_float M_bool $2.any $1 $2 $3}
-| term LT term {sp_p $2; symops P_cmp M_float M_bool "<" $1 $2 $3}
-| term GT term {sp_p $2; symops P_cmp M_float M_bool ">" $1 $2 $3}
-| term EQ_OP term {sp_p $2; symops P_eq M_float M_bool $2.any $1 $2 $3}
-| term EQ_OP_STR term {sp_p $2; symops P_eq M_string M_bool $2.any $1 $2 $3}
-
-| term BIT_AND term {sp_p $2; symops P_bit M_int M_int "&" $1 $2 $3}
-| term BIT_OR term { symops P_bit M_int M_int "|" $1 $2 $3}
-| term BIT_XOR term {sp_p $2; symops P_bit M_int M_int "^" $1 $2 $3}
-
-| term POWER term { symops P_tight M_float (mcontext_float_or_int [$1.mcontext; $3.mcontext]) "**" $1 $2 $3}
-| term PLUS term { symops P_add M_float (mcontext_float_or_int [$1.mcontext; $3.mcontext]) $2.any $1 $2 $3}
-| term CONCAT term {sp_p $2; symops P_add M_string M_string "." $1 $2 $3}
-| term BIT_SHIFT term { symops (P_paren_wanted P_tight) M_int M_int $2.any $1 $2 $3}
-| term XOR term {sp_p $2; symops (P_paren_wanted P_expr) M_bool M_bool "xor" $1 $2 $3}
-| term DOTDOT term { symops (P_paren_wanted P_expr) M_unknown_scalar M_string $2.any $1 $2 $3}
-
-| term AND_TIGHT term {sp_p $2; sp_same $2 $3; mcontext_check M_bool $1; let pri = P_tight_and in to_Call_op_ (mcontext_to_scalar $3.mcontext) pri "&&" [prio_lo pri $1; prio_lo_after pri $3] $1 $3}
-| term OR_TIGHT term {sp_p $2; sp_same $2 $3; mcontext_check M_bool $1; let pri = P_tight_or in to_Call_op_ (mcontext_to_scalar (mcontext_merge $1.mcontext $3.mcontext)) pri "||" [prio_lo pri $1; prio_lo_after pri $3] $1 $3}
-
-| term MULT term {sp_same $2 $3; let pri = P_mul in to_Call_op_ (mcontext_float_or_int [$1.mcontext; $3.mcontext]) pri $2.any [prio_lo_concat $1; prio_lo_after pri $3] $1 $3}
-| term MULT_L_STR term {sp_same $2 $3; mcontext_check M_int $3; let pri = P_mul in to_Call_op_ (if mcontext_lower $1.mcontext M_string then M_string else M_list) pri "x"
- [prio_lo_concat $1; prio_lo_after pri $3] $1 $3}
-
-| term ASSIGN term {sp_same $2 $3; let pri = P_assign in to_Call_assign_op_ (mcontext_op_assign $1 $3) pri $2.any ($1.any.expr) (prio_lo_after pri $3) $1 $3}
-
-| term ASSIGN BRACKET expr_bracket_end {sp_p($2); sp_p($3); sp_p($4); to_Call_assign_op_ (M_mixed [M_ref M_hash; M_none]) P_assign $2.any (prio_lo P_assign $1) $4.any $1 $4}
-| term AND_TIGHT BRACKET expr_bracket_end {sp_p($2); sp_p($3); sp_p($4); to_Call_op_ M_bool P_tight_and "&&" [prio_lo P_assign $1; $4.any] $1 $4}
-| term OR_TIGHT BRACKET expr_bracket_end {sp_p($2); sp_p($3); sp_p($4); to_Call_op_ M_bool P_tight_or "||" [prio_lo P_assign $1; $4.any] $1 $4}
-
-
-| term PATTERN_MATCH PATTERN {sp_n($2); sp_p($3); check_unneeded_var_dollar_ ($1); mcontext_check M_string $1; to_Call_op_ M_array P_expr "m//" ($1.any.expr :: from_PATTERN $3) $1 $3}
-| term PATTERN_MATCH_NOT PATTERN {sp_n($2); sp_p($3); check_unneeded_var_dollar_not($1); mcontext_check M_string $1; to_Call_op_ M_int P_expr "!m//" ($1.any.expr :: from_PATTERN $3) $1 $3}
-| term PATTERN_MATCH PATTERN_SUBST {sp_n($2); sp_p($3); check_unneeded_var_dollar_s ($1); to_Call_op_ (M_mixed[M_none; M_int]) P_expr "s///" ($1.any.expr :: from_PATTERN_SUBST $3) $1 $3}
-| term PATTERN_MATCH_NOT PATTERN_SUBST {die_with_rawpos $2.pos "use =~ instead of !~ and negate the return value"}
-
-| term PATTERN_MATCH QR_PATTERN {warn [Warn_traps] $3.pos "use m/.../ or /.../ instead of qr/.../ when you do a pattern matching"; to_Call_op_ M_array P_expr "m//" ($1.any.expr :: from_PATTERN $3) $1 $3}
-| term PATTERN_MATCH_NOT QR_PATTERN {warn [Warn_traps] $3.pos "use m/.../ or /.../ instead of qr/.../ when you do a pattern matching"; to_Call_op_ M_int P_expr "!m//" ($1.any.expr :: from_PATTERN $3) $1 $3}
-| term PATTERN_MATCH scalar { new_pesp M_array P_expr (Call(Too_complex, [$1.any.expr ; $3.any ])) $1 $3}
-| term PATTERN_MATCH_NOT scalar { new_pesp M_int P_expr (Call(Too_complex, [$1.any.expr ; $3.any ])) $1 $3}
-
-| term PATTERN_MATCH RAW_STRING {warn [Warn_complex_expressions] $3.pos "use a regexp, not a string"; to_Call_op_ M_array P_expr "m//" [ $1.any.expr; to_Raw_string $3 ] $1 $3}
-| term PATTERN_MATCH_NOT RAW_STRING {warn [Warn_complex_expressions] $3.pos "use a regexp, not a string"; to_Call_op_ M_int P_expr "!m//" [ $1.any.expr; to_Raw_string $3 ] $1 $3}
-| term PATTERN_MATCH STRING {warn [Warn_complex_expressions] $3.pos "use a regexp, not a string"; to_Call_op_ M_array P_expr "m//" [ $1.any.expr; to_String false $3 ] $1 $3}
-| term PATTERN_MATCH_NOT STRING {warn [Warn_complex_expressions] $3.pos "use a regexp, not a string"; to_Call_op_ M_int P_expr "!m//" [ $1.any.expr; to_String false $3 ] $1 $3}
-
-
-| term QUESTION_MARK term COLON term {sp_p($2); sp_p($3); sp_p($4); sp_p($5); mcontext_check M_bool $1; to_Call_op_ (mcontext_merge $3.mcontext $5.mcontext) P_ternary "?:" (check_ternary_paras(prio_lo P_ternary $1, prio_lo_after P_ternary $3, prio_lo_after P_ternary $5)) $1 $5}
-| term QUESTION_MARK term COLON BRACKET expr BRACKET_END {sp_p($2); sp_p($3); sp_p($4); sp_p($5); sp_p($6); sp_p($7); mcontext_check M_bool $1; to_Call_op_ (mcontext_merge $3.mcontext (M_ref M_hash)) P_ternary "?:" (check_ternary_paras(prio_lo P_ternary $1, prio_lo_after P_ternary $3, hash_ref $6)) $1 $7}
-| term QUESTION_MARK BRACKET expr BRACKET_END COLON term {sp_p($2); sp_p($3); sp_p($4); sp_p($5); sp_p($6); sp_p($7); mcontext_check M_bool $1; to_Call_op_ (mcontext_merge $7.mcontext (M_ref M_hash)) P_ternary "?:" (check_ternary_paras(prio_lo P_ternary $1, hash_ref $4, prio_lo_after P_ternary $7)) $1 $7}
-| term QUESTION_MARK BRACKET expr BRACKET_END COLON BRACKET expr BRACKET_END {sp_p($2); sp_p($3); sp_p($4); sp_p($5); sp_p($6); sp_p($7); sp_p($8); sp_p($9); mcontext_check M_bool $1; to_Call_op_ (M_ref M_hash) P_ternary "?:" (check_ternary_paras(prio_lo P_ternary $1, hash_ref $4, hash_ref $8)) $1 $9}
-
-/* Unary operators and terms */
-| PLUS term %prec UNARY_MINUS {
- sp_0($2);
- match $1.any with
- | "+" ->
- warn_rule [Warn_normalized_expressions] "don't use unary +" ;
- to_Call_op_ (mcontext_float_or_int [$2.mcontext]) P_tight "+ unary" [$2.any.expr] $1 $2
- | "-" ->
- (match $2.any.expr with
- | Ident(_, _, pos) when $2.spaces = Space_0 ->
- let s = "-" ^ string_of_fromparser $2.any.expr in
- warn_rule [Warn_complex_expressions] (Printf.sprintf "don't use %s, use '%s' instead" s s);
- new_pesp M_string P_tok (Raw_string(s, pos)) $1 $2
- | _ -> to_Call_op_ (mcontext_float_or_int [$2.mcontext]) P_tight "- unary" [$2.any.expr] $1 $2)
- | _ -> die_rule "syntax error"
-}
-| TIGHT_NOT term {check_negatable_expr $2; mcontext_check M_bool $2; to_Call_op_ M_bool P_tight "not" [$2.any.expr] $1 $2}
-| BIT_NEG term { mcontext_check M_int $2; to_Call_op_ M_int P_expr "~" [$2.any.expr] $1 $2}
-| INCR term {sp_0($2); mcontext_check M_int $2; to_Call_op_ (M_mixed [M_int ; M_none]) P_tight "++" [$2.any.expr] $1 $2}
-| DECR term {sp_0($2); mcontext_check M_int $2; to_Call_op_ (M_mixed [M_int ; M_none]) P_tight "--" [$2.any.expr] $1 $2}
-| term INCR {sp_0($2); mcontext_check M_int $1; to_Call_op_ (M_mixed [M_int ; M_none]) P_tight "++ post" [$1.any.expr] $1 $2}
-| term DECR {sp_0($2); mcontext_check M_int $1; to_Call_op_ (M_mixed [M_int ; M_none]) P_tight "-- post" [$1.any.expr] $1 $2}
-| NOT argexpr {warn_rule [Warn_normalized_expressions] "don't use \"not\", use \"!\" instead"; mcontext_check_unop_l M_bool $2; to_Call_op_ M_bool P_and "not" ($2.any.expr) $1 $2}
-
-/* Constructors for anonymous data */
-
-| ARRAYREF ARRAYREF_END {sp_0($2); new_pesp (M_ref M_array) P_expr (Ref(I_array, List[])) $1 $2}
-| arrayref_start ARRAYREF_END {(if $1.any = [] then sp_0 else sp_p)($2) ; new_pesp (M_ref M_array) P_expr (Ref(I_array, List $1.any)) $1 $2}
-| arrayref_start expr ARRAYREF_END {sp_same $2 $3; new_pesp (M_ref M_array) P_expr (Ref(I_array, List($1.any @ [$2.any.expr]))) $1 $3}
-| arrayref_start BRACKET expr BRACKET_END ARRAYREF_END {sp_same $2 $5; new_pesp (M_ref M_array) P_expr (Ref(I_array, List($1.any @ [hash_ref $3]))) $1 $5}
-
-| BRACKET BRACKET_END {new_pesp (M_ref M_hash) P_expr (Ref(I_hash, List [])) $1 $2} /* empty hash */
-| BRACKET_HASHREF expr BRACKET_END %prec PREC_HIGH {sp_p($3); new_pesp (M_ref M_hash) P_expr (hash_ref $2) $1 $3} /* { foo => "Bar" } */
-| SUB BRACKET BRACKET_END %prec PREC_HIGH {sp_n($2); sp_0($3); new_pesp (M_ref M_sub) P_expr (Anonymous_sub(None, Block [], pos_range $2 $3)) $1 $3}
-| SUB_WITH_PROTO BRACKET BRACKET_END %prec PREC_HIGH {sp_n($2); sp_0($3); new_pesp (M_ref M_sub) P_expr (Anonymous_sub(Some $1.any, Block [], pos_range $2 $3)) $1 $3}
-| SUB BRACKET lines BRACKET_END %prec PREC_HIGH {sp_n($2); new_pesp (M_ref M_sub) P_expr (anonymous_sub None $3 $4) $1 $4}
-| SUB_WITH_PROTO BRACKET lines BRACKET_END %prec PREC_HIGH {sp_n($2); new_pesp (M_ref M_sub) P_expr (anonymous_sub (Some $1.any) $3 $4) $1 $4}
-
-| termdo {new_1pesp P_tok $1.any $1}
-| REF term {new_pesp (M_ref $2.mcontext) P_expr (Ref(I_scalar, remove_call_with_same_para_special $2.any.expr)) $1 $2} /* \$x, \@y, \%z */
-| my_our %prec UNIOP {new_1pesp P_expr $1.any $1}
-| LOCAL term %prec UNIOP {sp_n($2); new_pesp (M_mixed [ $2.mcontext ; M_none ]) P_expr (to_Local $2) $1 $2}
-
-| parenthesized {new_1pesp $1.any.priority (List $1.any.expr) $1} /* (1, 2) */
-| parenthesized arrayref {sp_0($2); let is_slice = not (is_only_one_in_List $2.any) in new_pesp (if is_slice then M_list else M_unknown_scalar) P_tok (to_Deref_with(I_array, (if is_slice then I_array else I_scalar), List $1.any.expr, List $2.any)) $1 $2} /* list indexing or slicing */
-
-| variable {
- let e =
- match $1.any with
- | Deref(I_func, Ident _) ->
- call_with_same_para_special $1.any (* not the same as f(@_) *)
- | e -> e in
- new_1pesp P_tok e $1
- }
-
-| subscripted {new_1pesp P_tok $1.any $1}
-
-| array arrayref {new_pesp M_list P_expr (to_Deref_with(I_array, I_array, from_array $1, List $2.any)) $1 $2} /* array slice: @array[vals] */
-| array BRACKET expr BRACKET_END {sp_0($2); sp_0($3); sp_0($4); new_pesp M_list P_expr (to_Deref_with(I_hash, I_array, from_array $1, $3.any.expr)) $1 $4} /* hash slice: @hash{@keys} */
-
-/* function_calls */
-| ONE_SCALAR_PARA RAW_STRING {call_one_scalar_para P_uniop $1 [to_Raw_string $2] $1 $2}
-| ONE_SCALAR_PARA STRING {call_one_scalar_para P_uniop $1 [to_String true $2] $1 $2}
-| ONE_SCALAR_PARA variable {call_one_scalar_para P_uniop $1 [$2.any] $1 $2}
-| ONE_SCALAR_PARA restricted_subscripted {call_one_scalar_para P_uniop $1 [$2.any] $1 $2}
-| ONE_SCALAR_PARA parenthesized {call_one_scalar_para P_tok $1 $2.any.expr $1 $2}
-| ONE_SCALAR_PARA BRACKET lines BRACKET_END {sp_n($2); new_pesp M_unknown P_uniop (call(Deref(I_func, Ident(None, $1.any, raw_pos2pos $1.pos)), [anonymous_sub None $3 $4])) $1 $4} /* eval { foo } */
-| ONE_SCALAR_PARA diamond {call_one_scalar_para P_uniop $1 [$2.any] $1 $2}
-| ONE_SCALAR_PARA %prec PREC_LOW {call_one_scalar_para P_tok $1 [] $1 $1}
-| ONE_SCALAR_PARA word argexpr {check_parenthesized_first_argexpr_with_Ident $2.any $3; call_one_scalar_para P_uniop $1 [call(Deref(I_func, $2.any), $3.any.expr)] $1 $3} /* ref foo $a, $b */
-| ONE_SCALAR_PARA hash PKG_SCOPE {sp_0($3); call_one_scalar_para P_uniop $1 [ Call(Too_complex, [$2.any]) ] $1 $3} /* keys %main:: */
-| ONE_SCALAR_PARA BAREWORD {if $2.any = "_" && $1.any.[0] = '-' then new_pesp M_bool P_uniop Too_complex $1 $2 else die_rule "syntax error"} /* -e "foo" && -f _ */
-
-| ONE_SCALAR_PARA array arrayref {call_one_scalar_para P_uniop $1 [to_Deref_with(I_array, I_array, from_array $2, List $3.any)] $1 $3} /* array slice: @array[vals] */
-| ONE_SCALAR_PARA array BRACKET expr BRACKET_END {sp_0($3); sp_0($4); sp_0($5); call_one_scalar_para P_uniop $1 [to_Deref_with(I_hash, I_array, from_array $2, $4.any.expr)] $1 $5} /* hash slice: @hash{@keys} */
-
-| func parenthesized {sp_0($2); call_func $1 $2} /* &foo(@args) */
-| word argexpr {check_parenthesized_first_argexpr_with_Ident $1.any $2; call_no_paren $1 $2} /* foo $a, $b */
-| word BRACKET lines BRACKET_END MULT { die_with_rawpos $5.pos "I can't handle this correctly, please add parentheses" }
-| word BRACKET lines BRACKET_END COMMA argexpr %prec LSTOP {sp_n($2); new_pesp M_unknown P_call_no_paren (call(Deref(I_func, $1.any), Ref(I_hash, List (fst $3.any)) :: $6.any.expr)) $1 $6} /* bless { foo }, $bar */
-| word_paren parenthesized {sp_0($2); call_with_paren $1 $2} /* foo(@args) */
-| word BRACKET lines BRACKET_END listexpr %prec LSTOP {sp_n($2); call_and_context(Deref(I_func, $1.any), anonymous_sub None $3 $4 :: $5.any.expr) false (if $5.any.expr = [] then P_tok else P_call_no_paren) $1 $5} /* map { foo } @bar */
-| word BRACKET BRACKET expr BRACKET_END BRACKET_END listexpr %prec LSTOP {sp_n($2); sp_p($3); sp_p($4); sp_p($5); sp_p($6); new_pesp M_unknown (if $7.any.expr = [] then P_tok else P_call_no_paren) (call(Deref(I_func, $1.any), anonymous_sub None (new_esp (M_ref M_hash) ([ hash_ref $4 ], false) $3 $5) $6 :: $7.any.expr)) $1 $7} /* map { { foo } } @bar */
-| word BRACKET BRACKET expr BRACKET_END semi_colon BRACKET_END listexpr %prec LSTOP {sp_n($2); sp_p($3); sp_p($4); sp_p($5); sp_p($7); new_pesp M_unknown (if $8.any.expr = [] then P_tok else P_call_no_paren) (call(Deref(I_func, $1.any), anonymous_sub None (new_esp (M_ref M_hash) ([ hash_ref $4; Semi_colon ], true) $3 $6) $7 :: $8.any.expr)) $1 $8} /* map { { foo }; } @bar */
-
-| term ARROW word_or_scalar parenthesized {sp_0($2); sp_0($3); sp_0($4); if $4.any.expr = [] then warn [Warn_suggest_simpler] $4.pos "remove these unneeded parentheses"; new_pesp M_unknown P_tok (to_Method_call($1.any.expr, $3.any, $4.any.expr)) $1 $4} /* $foo->bar(list) */
-| term ARROW word_or_scalar {sp_0($2); sp_0($3); new_pesp M_unknown P_tok (to_Method_call($1.any.expr, $3.any, [])) $1 $3} /* $foo->bar */
-
-| NEW word { sp_n($2); new_pesp (M_ref M_unknown) P_expr (to_Method_call ($2.any, Ident(None, "new", get_pos $1), [])) $1 $2} /* new Class */
-| NEW word_paren parenthesized { sp_n($2); sp_0($3); new_pesp (M_ref M_unknown) P_expr (to_Method_call($2.any, Ident(None, "new", get_pos $1), $3.any.expr)) $1 $3} /* new Class(...) */
-| NEW word terminal { die_rule "you must parenthesize parameters: \"new Class(...)\" instead of \"new Class ...\"" }
-| NEW word variable { die_rule "you must parenthesize parameters: \"new Class(...)\" instead of \"new Class ...\"" }
-
-| PRINT { to_Call_op_ (M_mixed [M_int; M_none]) P_call_no_paren $1.any (var_STDOUT :: [ var_dollar_ (get_pos $1) ]) $1 $1}
-| PRINT argexpr {check_parenthesized_first_argexpr $1.any $2; to_Call_op_ (M_mixed [M_int; M_none]) P_call_no_paren $1.any (var_STDOUT :: $2.any.expr) $1 $2}
-| PRINT_TO_SCALAR { to_Call_op_ (M_mixed [M_int; M_none]) P_call_no_paren (fst $1.any) (var_STDOUT :: [ Deref(I_scalar, Ident(None, snd $1.any, get_pos $1)) ]) $1 $1}
-| PRINT_TO_SCALAR argexpr { to_Call_op_ (M_mixed [M_int; M_none]) P_call_no_paren (fst $1.any) (Deref(I_scalar, Ident(None, snd $1.any, get_pos $1)) :: $2.any.expr) $1 $2}
-| PRINT_TO_STAR { to_Call_op_ (M_mixed [M_int; M_none]) P_call_no_paren (fst $1.any) (Deref(I_star, Ident(None, snd $1.any, get_pos $1)) :: [ var_dollar_ (get_pos $1) ]) $1 $1}
-| PRINT_TO_STAR argexpr { to_Call_op_ (M_mixed [M_int; M_none]) P_call_no_paren (fst $1.any) (Deref(I_star, Ident(None, snd $1.any, get_pos $1)) :: $2.any.expr) $1 $2}
-
-| hash PKG_SCOPE {sp_0($2); new_pesp M_hash P_tok (Call(Too_complex, [$1.any])) $1 $2} /* %main:: */
-
-| terminal {$1}
-
-expr_bracket_end:
-| expr BRACKET_END { sp_p($2); new_esp (M_ref M_hash) (hash_ref $1) $1 $2 }
-| expr BRACKET_END ARROW bracket_subscript {sp_p($2); sp_0($3); new_esp M_unknown_scalar (to_Deref_with(I_hash, I_scalar, hash_ref $1, $4.any)) $1 $4} /* { foo }->{Bar} */
-
-terminal:
-| word {word_alone $1}
-| NUM {new_1pesp P_tok (Num($1.any, get_pos $1)) $1}
-| STRING {new_1pesp P_tok (to_String true $1) $1}
-| RAW_STRING {new_1pesp P_tok (to_Raw_string $1) $1}
-| REVISION {new_1pesp P_tok (to_Raw_string $1) $1}
-| COMMAND_STRING {to_Call_op_ (M_mixed[M_string; M_list]) P_tok "``" [to_String false $1] $1 $1}
-| QUOTEWORDS {let l = List.map (fun s -> Raw_string(s, raw_pos2pos $1.pos)) (words $1.any) in new_pesp (M_tuple (repeat M_string (List.length l))) P_tok (List [ List l ]) $1 $1}
-| HERE_DOC {new_1pesp P_tok (to_String false (new_1esp (fst $1.any) $1)) $1 }
-| RAW_HERE_DOC {new_1pesp P_tok (Raw_string(fst $1.any, raw_pos2pos (snd $1.any))) $1}
-| QR_PATTERN {to_Call_op_ M_string P_tok "qr//" (from_PATTERN $1) $1 $1}
-| PATTERN {to_Call_op_ M_array P_expr "m//" (var_dollar_ (get_pos $1) :: from_PATTERN $1) $1 $1}
-| PATTERN_SUBST {to_Call_op_ (M_mixed[M_none; M_int]) P_expr "s///" (var_dollar_ (get_pos $1) :: from_PATTERN_SUBST $1) $1 $1}
-| diamond {new_1pesp P_expr $1.any $1}
-
-diamond:
-| LT GT {sp_0($2); to_Call_op (M_mixed[M_string; M_list]) "<>" [] $1 $2}
-| LT term GT {sp_0($2); sp_0($3); to_Call_op (M_mixed[M_string; M_list]) "<>" [$2.any.expr] $1 $3}
-
-subscripted: /* Some kind of subscripted expression */
-| variable PKG_SCOPE bracket_subscript {sp_0($2); sp_0($3); new_esp M_unknown (Call(Too_complex, [$3.any])) $1 $3} /* $foo::{something} */
-| scalar bracket_subscript {sp_0($2); check_scalar_subscripted $1; new_esp M_unknown_scalar (to_Deref_with(I_hash , I_scalar, from_scalar $1, $2.any )) $1 $2} /* $foo{bar} */
-| scalar arrayref {sp_0($2); check_scalar_subscripted $1; new_esp M_unknown_scalar (to_Deref_with(I_array, I_scalar, from_scalar $1, only_one_array_ref $2)) $1 $2} /* $array[$element] */
-| term ARROW simple_subscript {sp_0($2); sp_0($3); new_esp $3.mcontext (to_Deref_with_arrow $2 (fst $3.any, I_scalar, $1.any.expr, snd $3.any)) $1 $3}
-| subscripted simple_subscript {sp_0($2); new_esp $2.mcontext (to_Deref_with(fst $2.any, I_scalar, $1.any, snd $2.any)) $1 $2}
-
-restricted_subscripted: /* Some kind of subscripted expression */
-| variable PKG_SCOPE bracket_subscript {sp_0($2); sp_0($3); new_esp M_unknown (Call(Too_complex, [$3.any])) $1 $3} /* $foo::{something} */
-| word_paren parenthesized {new_esp M_unknown (call(Deref(I_func, $1.any), $2.any.expr)) $1 $2}
-| scalar bracket_subscript {sp_0($2); check_scalar_subscripted $1; new_esp M_unknown_scalar (to_Deref_with(I_hash , I_scalar, from_scalar $1, $2.any )) $1 $2} /* $foo{bar} */
-| scalar arrayref {sp_0($2); check_scalar_subscripted $1; new_esp M_unknown_scalar (to_Deref_with(I_array, I_scalar, from_scalar $1, only_one_array_ref $2)) $1 $2} /* $array[$element] */
-| scalar ARROW simple_subscript {sp_0($2); sp_0($3); new_esp $3.mcontext (to_Deref_with_arrow $2 (fst $3.any, I_scalar, $1.any, snd $3.any)) $1 $3}
-| restricted_subscripted ARROW simple_subscript {sp_0($2); sp_0($3); new_esp $3.mcontext (to_Deref_with_arrow $2 (fst $3.any, I_scalar, $1.any, snd $3.any)) $1 $3} /* somehref->{bar} */
-| restricted_subscripted simple_subscript {sp_0($2); new_esp $2.mcontext (to_Deref_with(fst $2.any, I_scalar, $1.any, snd $2.any)) $1 $2}
-
-| restricted_subscripted ARROW word_or_scalar parenthesized {sp_0($2); sp_0($3); sp_0($4); if $4.any.expr = [] then warn [Warn_suggest_simpler] $4.pos "remove these unneeded parentheses"; new_esp M_unknown (to_Method_call($1.any, $3.any, $4.any.expr)) $1 $4} /* $foo->bar(list) */
-| restricted_subscripted ARROW word_or_scalar {sp_0($2); sp_0($3); new_esp M_unknown (to_Method_call($1.any, $3.any, [])) $1 $3} /* $foo->bar */
-
-simple_subscript:
-| bracket_subscript {new_esp M_unknown_scalar (I_hash, $1.any) $1 $1}
-| arrayref {new_esp M_unknown_scalar (I_array, only_one_array_ref $1) $1 $1}
-| parenthesized {new_esp M_unknown (I_func , List($1.any.expr)) $1 $1}
-
-
-arrayref:
-| arrayref_start ARRAYREF_END {sp_0($2); new_esp (M_ref M_array) $1.any $1 $2}
-| arrayref_start expr ARRAYREF_END {sp_0($3); new_esp (M_ref M_array) ($1.any @ [$2.any.expr]) $1 $3}
-| arrayref_start BRACKET expr BRACKET_END ARRAYREF_END {sp_p($2); sp_p($4); sp_0($5); new_esp (M_ref M_hash) ($1.any @ [hash_ref $3]) $1 $5}
-parenthesized:
-| parenthesized_start PAREN_END {sp_0_or_cr($2); new_pesp (if $1.any = [] then M_list else $1.mcontext) (if $1.any = [] then P_tok else P_paren P_comma) $1.any $1 $2}
-| parenthesized_start expr PAREN_END {sp_0_or_cr($3); (if $1.any = [] then sp_0_or_cr else sp_p)($2); new_pesp (if $1.any = [] then $2.mcontext else M_list) (P_paren (if $1.any = [] then $2.any.priority else P_comma)) ($1.any @ [(if $1.any = [] then prio_lo P_loose else prio_lo_after P_comma) $2]) $1 $3}
-| parenthesized_start BRACKET expr BRACKET_END PAREN_END {sp_p($4); sp_0_or_cr($5); new_pesp (if $1.any = [] then M_ref M_hash else M_list) (P_paren (if $1.any = [] then P_expr else P_comma)) ($1.any @ [hash_ref $3]) $1 $5}
-
-arrayref_start:
-| ARRAYREF {new_1esp [] $1 }
-| arrayref_start BRACKET expr BRACKET_END comma {sp_p($2); sp_p($3); sp_p($4); new_esp M_special ($1.any @ [hash_ref $3]) $1 $5}
-parenthesized_start:
-| PAREN {new_1esp [] $1 }
-| parenthesized_start BRACKET expr BRACKET_END comma {(if $1.any = [] then sp_0_or_cr else sp_p)($2); sp_p($3); sp_p($4); new_esp (M_ref M_hash) ($1.any @ [hash_ref $3]) $1 $5}
-
-my_our: /* Things that can be "my"'d */
-| my_our_paren PAREN_END {sp_0($2); new_esp (M_mixed [ $1.mcontext ; M_none ]) (My_our(sndfst $1.any, snd $1.any, get_pos $1)) $1 $2}
-| my_our_paren SCALAR_IDENT PAREN_END {check_my_our_paren $1 $2; new_esp (M_mixed [ mtuple_context_concat $1.mcontext M_unknown_scalar; M_none ]) (My_our(sndfst $1.any, snd $1.any @ [I_scalar, snd $2.any], pos_range $1 $3)) $1 $3}
-| my_our_paren HASH_IDENT PAREN_END {check_my_our_paren $1 $2; new_esp (M_mixed [ M_list ; M_none ]) (My_our(sndfst $1.any, snd $1.any @ [I_hash, snd $2.any], pos_range $1 $3)) $1 $3}
-| my_our_paren ARRAY_IDENT PAREN_END {check_my_our_paren $1 $2; new_esp (M_mixed [ M_list ; M_none ]) (My_our(sndfst $1.any, snd $1.any @ [I_array, snd $2.any], pos_range $1 $3)) $1 $3}
-| MY_OUR SCALAR_IDENT {new_esp (M_mixed [M_unknown_scalar; M_none]) (My_our($1.any, [I_scalar, snd $2.any], get_pos $2)) $1 $2}
-| MY_OUR HASH_IDENT {new_esp (M_mixed [M_hash ; M_none]) (My_our($1.any, [I_hash, snd $2.any], get_pos $2)) $1 $2}
-| MY_OUR ARRAY_IDENT {new_esp (M_mixed [M_array ; M_none]) (My_our($1.any, [I_array, snd $2.any], get_pos $2)) $1 $2}
-
-my_our_paren:
-| MY_OUR PAREN {sp_1($2); new_esp (M_tuple []) ((true, $1.any), []) $1 $2}
-| my_our_paren comma {if fstfst $1.any then die_rule "syntax error"; new_esp $1.mcontext ((true, sndfst $1.any), snd $1.any) $1 $2}
-| my_our_paren BAREWORD {check_my_our_paren $1 $2; if $2.any <> "undef" then die_rule "scalar expected"; new_esp (mtuple_context_concat $1.mcontext M_none) ((false, sndfst $1.any), snd $1.any @ [I_raw, $2.any]) $1 $2}
-| my_our_paren SCALAR_IDENT {check_my_our_paren $1 $2; new_esp (mtuple_context_concat $1.mcontext M_unknown_scalar) ((false, sndfst $1.any), snd $1.any @ [I_scalar, snd $2.any]) $1 $2}
-| my_our_paren HASH_IDENT {check_my_our_paren $1 $2; new_esp M_list ((false, sndfst $1.any), snd $1.any @ [I_hash, snd $2.any]) $1 $2}
-| my_our_paren ARRAY_IDENT {check_my_our_paren $1 $2; new_esp M_list ((false, sndfst $1.any), snd $1.any @ [I_array, snd $2.any]) $1 $2}
-
-termdo: /* Things called with "do" */
-| DO term %prec UNIOP { die_rule "\"do EXPR\" not allowed" } /* do $filename */
-| DO BRACKET lines BRACKET_END %prec PREC_HIGH {sp_n($2); new_esp $3.mcontext (lines_to_Block $3 $4) $1 $4} /* do { code */
-
-bracket_subscript:
-| BRACKET expr BRACKET_END {sp_0($1); sp_same $2 $3; check_hash_subscript $2; new_esp M_special (only_one_in_List $2) $1 $3}
-| COMPACT_HASH_SUBSCRIPT {sp_0($1); new_1esp (to_Raw_string $1) $1 }
-
-variable:
-| scalar {$1}
-| star {$1}
-| hash {$1}
-| array {$1}
-| arraylen {$1} /* $#x, $#{ something } */
-| func {$1} /* &foo; */
-
-word:
-| bareword { $1 }
-| RAW_IDENT { new_1esp (to_Ident $1) $1 }
-
-comma: COMMA {new_esp M_special true $1 $1} | RIGHT_ARROW {sp_p($1); new_1esp false $1 }
-
-semi_colon: SEMI_COLON {sp_0($1); $1}
-
-word_or_scalar:
-| word {$1}
-| scalar {$1}
-| word_paren {$1}
-| MULT_L_STR { new_1esp (Ident(None, "x", get_pos $1)) $1 }
-| FOR { new_1esp (Ident(None, $1.any, get_pos $1)) $1 }
-| ONE_SCALAR_PARA { new_1esp (Ident(None, $1.any, get_pos $1)) $1 }
-
-bareword:
-| NEW { new_1esp (Ident(None, "new", get_pos $1)) $1 }
-| BAREWORD { new_1esp (Ident(None, $1.any, get_pos $1)) $1 }
-
-word_paren:
-| BAREWORD_PAREN { new_1esp (Ident(None, $1.any, get_pos $1)) $1 }
-| RAW_IDENT_PAREN { new_1esp (to_Ident $1) $1 }
-| PO_COMMENT word_paren { po_comment($1); new_esp M_special $2.any $1 $2 }
-
-
-arraylen: ARRAYLEN_IDENT {new_esp M_int (deref_arraylen (to_Ident $1)) $1 $1} | ARRAYLEN scalar {sp_0($2); new_esp M_int (deref_arraylen $2.any ) $1 $1 } | ARRAYLEN bracket_subscript {new_esp M_int (deref_arraylen $2.any) $1 $2}
-scalar: SCALAR_IDENT {new_esp M_unknown_scalar (Deref(I_scalar, to_Ident $1)) $1 $1} | DOLLAR scalar {sp_0($2); new_esp M_unknown_scalar (Deref(I_scalar, $2.any)) $1 $1 } | DOLLAR bracket_subscript {new_esp M_unknown_scalar (deref_raw I_scalar $2.any) $1 $2} | DOLLAR BRACKET BRACKET expr BRACKET_END BRACKET_END {sp_0($2); sp_0($3); sp_p($5); sp_0($6); new_esp M_unknown_scalar (Deref(I_scalar, hash_ref $4)) $1 $6}
-func: FUNC_IDENT {new_esp M_unknown (Deref(I_func , to_Ident $1)) $1 $1} | AMPERSAND scalar {sp_0($2); new_esp M_unknown (Deref(I_func , $2.any)) $1 $1 } | AMPERSAND bracket_subscript {new_esp M_unknown (deref_raw I_func $2.any) $1 $2}
-array: ARRAY_IDENT {new_esp M_array (Deref(I_array , to_Ident $1)) $1 $1} | AT scalar {sp_0($2); new_esp M_array (Deref(I_array , $2.any)) $1 $1 } | AT bracket_subscript {new_esp M_array (deref_raw I_array $2.any) $1 $2}
-hash: HASH_IDENT {new_esp M_hash (Deref(I_hash , to_Ident $1)) $1 $1} | PERCENT scalar {sp_0($2); new_esp M_hash (Deref(I_hash , $2.any)) $1 $1 } | PERCENT bracket_subscript {new_esp M_hash (deref_raw I_hash $2.any) $1 $2}
-star: STAR_IDENT {new_esp M_unknown (Deref(I_star , to_Ident $1)) $1 $1} | STAR scalar {sp_0($2); new_esp M_unknown (Deref(I_star , $2.any)) $1 $1 } | STAR bracket_subscript {new_esp M_unknown (deref_raw I_star $2.any) $1 $2}
-
-expr_or_empty: {default_esp (Block [])} | expr {new_1esp $1.any.expr $1 }
-
-%%
-
-prog_ref := Some prog
-;;
diff --git a/perl_checker.src/parser_helper.ml b/perl_checker.src/parser_helper.ml
deleted file mode 100644
index 43d60a4..0000000
--- a/perl_checker.src/parser_helper.ml
+++ /dev/null
@@ -1,1409 +0,0 @@
-open Types
-open Common
-open Printf
-
-let bpos = -1, -1
-
-let raw_pos2pos(a, b) = !Info.current_file, a, b
-let raw_pos_range { pos = (a1, b1) } { pos = (a2, b2) } = (if a1 = -1 then a2 else a1), (if b2 = -1 then b1 else b2)
-let pos_range esp1 esp2 = raw_pos2pos (raw_pos_range esp1 esp2)
-let get_pos pesp = raw_pos2pos pesp.pos
-let get_pos_start { pos = (start, _) } = start
-let get_pos_end { pos = (_, end_) } = end_
-let var_dollar_ pos = Deref(I_scalar, Ident(None, "_", pos))
-let var_STDOUT = Deref(I_star, Ident(None, "STDOUT", raw_pos2pos bpos))
-
-let new_any mcontext any spaces pos = { mcontext = mcontext ; any = any ; spaces = spaces ; pos = pos }
-let new_any_ any spaces pos = new_any M_unknown any spaces pos
-let new_esp mcontext e esp_start esp_end = new_any mcontext e esp_start.spaces (raw_pos_range esp_start esp_end)
-let new_1esp e esp = new_any esp.mcontext e esp.spaces esp.pos
-let new_pesp mcontext prio e esp_start esp_end = new_any mcontext { priority = prio ; expr = e } esp_start.spaces (raw_pos_range esp_start esp_end)
-let new_1pesp prio e esp = new_any esp.mcontext { priority = prio ; expr = e } esp.spaces esp.pos
-let default_esp e = new_any M_unknown e Space_none bpos
-let default_pesp prio e = new_any M_unknown { priority = prio ; expr = e } Space_none bpos
-
-let split_name_or_fq_name full_ident =
- match split_at2 ':'':' full_ident with
- | [] -> internal_error "split_ident"
- | [ident] -> None, ident
- | l ->
- let fql, name = split_last l in
- let fq = String.concat "::" fql in
- Some fq, name
-
-let is_var_dollar_ = function
- | Deref(I_scalar, Ident(None, "_", _)) -> true
- | _ -> false
-let is_var_number_match = function
- | Deref(I_scalar, Ident(None, s, _)) -> String.length s = 1 && s.[0] <> '0' && char_is_number s.[0]
- | _ -> false
-
-let non_scalar_context context = context = I_hash || context = I_array
-let is_scalar_context context = context = I_scalar
-
-let rec is_not_a_scalar = function
- | Deref_with(_, context, _, _)
- | Deref(context, _) -> non_scalar_context context
- | List []
- | List(_ :: _ :: _) -> true
- | Call(Deref(I_func, Ident(None, "map", _)), _)
- | Call(Deref(I_func, Ident(None, "grep", _)), _) -> true
- | Call_op("?:", [ _cond ; a; b ], _) -> is_not_a_scalar a || is_not_a_scalar b
- | _ -> false
-
-let is_a_scalar = function
- | Ref _
- | Num _
- | Raw_string _
- | String _
- | Call(Deref(I_func, Ident(None, "N", _)), _) -> true
- | My_our(_, [ context, _ ], _)
- | Deref_with(_, context, _, _)
- | Deref(context, _) -> is_scalar_context context
- | _ -> false
-
-let is_a_string = function
- | String _ | Raw_string _ -> true
- | _ -> false
-
-let is_parenthesized = function
- | List[]
- | List[List _] -> true
- | _ -> false
-
-let un_parenthesize = function
- | List[List[e]] -> e
- | List[e] -> e
- | _ -> internal_error "un_parenthesize"
-
-let rec un_parenthesize_full = function
- | List[e] -> un_parenthesize_full e
- | e -> e
-
-let rec un_parenthesize_full_l = function
- | [ List l ] -> un_parenthesize_full_l l
- | l -> l
-
-let is_always_true = function
- | Num(n, _) -> float_of_string n <> 0.
- | Raw_string(s, _) -> s <> ""
- | String(l, _) -> l <> []
- | Ref _ -> true
- | _ -> false
-
-let is_always_false = function
- | Num(n, _) -> float_of_string n = 0.
- | Raw_string(s, _) -> s = ""
- | String(l, _) -> l = []
- | List [] -> true
- | Ident(None, "undef", _) -> true
- | _ -> false
-
-let rec is_lvalue = function
- | Call(Deref(I_func, Ident(None, f, _)), _) -> List.mem f [ "substr" ]
-
- | Call_op("?:", [ _ ; a ; b ], _) -> is_lvalue a && is_lvalue b
-
- | Call_op("local", l, _)
- | List [ List l ]
- -> List.for_all is_lvalue l
-
- | My_our _
- | Deref(_, _)
- | Deref_with(_, _, _, _)
- | Ident(None, "undef", _)
- -> true
-
- | _ -> false
-
-let not_complex e =
- if is_parenthesized e then true else
- let rec not_complex_ op = function
- | Call_op("?:", _, _) -> false
- | Call_op(op', l, _) -> op <> op' && List.for_all (not_complex_ op') l
- | e -> not (is_parenthesized e)
- in not_complex_ "" (un_parenthesize_full e)
-
-let not_simple = function
- | Num _ | Ident _ | Deref(_, Ident _) -> false
- | _ -> true
-
-let context2s = function
- | I_scalar -> "$"
- | I_hash -> "%"
- | I_array -> "@"
- | I_func -> "&"
- | I_raw -> ""
- | I_star -> "*"
-let variable2s(context, ident) = context2s context ^ ident
-
-let rec string_of_fromparser = function
- | Semi_colon -> ";"
- | Undef -> "undef"
- | Num(num, _) -> num
-
- | Raw_string(s, _) -> "\"" ^ s ^ "\""
- | String(l, _) ->
- let l' = List.map (fun (s, e) ->
- s ^ if e = List[] then "" else string_of_fromparser e
- ) l in
- "\"" ^ String.concat "" l' ^ "\""
-
- | Ident(None, s, _) -> s
- | Ident(Some fq, s, _) -> fq ^ "::" ^ s
- | My_our(myour, l, _) -> myour ^ "(" ^ String.concat "," (List.map (fun (context, s) -> context2s context ^ s) l) ^ ")"
-
- | Anonymous_sub(_, e, _) -> "sub { " ^ string_of_fromparser e ^ " }"
- | Ref(_, e) -> "\\" ^ string_of_fromparser e
- | Deref(context, e) -> context2s context ^ string_of_fromparser e
-
- | Diamond(None) -> "<>"
- | Diamond(Some e) -> "<" ^ string_of_fromparser e ^ ">"
-
- | Sub_declaration(name, _prototype, body, Real_sub_declaration) ->
- "sub " ^ string_of_fromparser name ^ " { " ^ string_of_fromparser body ^ " }"
-
- | Sub_declaration(name, _prototype, body, Glob_assign) ->
- "*" ^ string_of_fromparser name ^ " = sub { " ^ string_of_fromparser body ^ " };"
-
- | Deref_with(_, _, _e1, _e2) ->
- internal_error "todo"
-
- | Package(p) -> "package " ^ string_of_fromparser p
-
- | Use(e, []) -> "use " ^ string_of_fromparser e
- | Use(e, l) -> "use " ^ string_of_fromparser e ^ "(" ^ lstring_of_fromparser l
-
- | List l -> lstring_of_fromparser_parentheses l
- | Block l -> "{ " ^ lstring_of_fromparser l ^ " }"
- | Call_op(op, l, _) -> op ^ lstring_of_fromparser_parentheses l
-
- | Call(e, l) -> string_of_fromparser e ^ lstring_of_fromparser l
-
- | Method_call(obj, meth, l) ->
- let para = if l = [] then "" else lstring_of_fromparser_parentheses l in
- string_of_fromparser obj ^ "->" ^ string_of_fromparser meth ^ para
-
- | Label(e) -> e ^ ": "
-
- | Perl_checker_comment _ -> ""
- | Too_complex -> "XXX"
-
-and lstring_of_fromparser l = String.concat ", " (List.map string_of_fromparser l)
-and lstring_of_fromparser_parentheses l = "(" ^ lstring_of_fromparser l ^ ")"
-
-let rec is_same_fromparser a b =
- match a, b with
- | Undef, Undef -> true
- | Ident(fq1, s1, _), Ident(fq2, s2, _) -> fq1 = fq2 && s1 = s2
- | Num(s1, _), Num(s2, _)
- | Raw_string(s1, _), Raw_string(s2, _) -> s1 = s2
-
- | String(l1, _), String(l2, _) ->
- for_all2_ (fun (s1, e1) (s2, e2) -> s1 = s2 && is_same_fromparser e1 e2) l1 l2
-
- | Ref(c1, e1), Ref(c2, e2)
- | Deref(c1, e1), Deref(c2, e2) -> c1 = c2 && is_same_fromparser e1 e2
-
- | Deref_with(c1, c_1, e1, e_1), Deref_with(c2, c_2, e2, e_2) -> c1 = c2 && c_1 = c_2 && is_same_fromparser e1 e2 && is_same_fromparser e_1 e_2
-
- | Diamond(None), Diamond(None) -> true
- | Diamond(Some e1), Diamond(Some e2) -> is_same_fromparser e1 e2
-
- | List(l1), List(l2) -> for_all2_ is_same_fromparser l1 l2
-
- | Call_op(op1, l1, _), Call_op(op2, l2, _) -> op1 = op2 && for_all2_ is_same_fromparser l1 l2
- | Call(e1, l1), Call(e2, l2) -> is_same_fromparser e1 e2 && for_all2_ is_same_fromparser l1 l2
-
- | Method_call(e1, m1, l1), Method_call(e2, m2, l2) ->
- is_same_fromparser e1 e2 && is_same_fromparser m1 m2 && for_all2_ is_same_fromparser l1 l2
-
- | _ -> false
-
-let from_scalar esp =
- match esp.any with
- | Deref(I_scalar, ident) -> ident
- | _ -> internal_error "from_scalar"
-
-let from_array esp =
- match esp.any with
- | Deref(I_array, ident) -> ident
- | _ -> internal_error "from_array"
-
-let rec get_pos_from_expr = function
- | Anonymous_sub(_, _, pos)
- | String(_, pos)
- | Call_op(_, _, pos)
- | Perl_checker_comment(_, pos)
- | My_our(_, _, pos)
- | Raw_string(_, pos)
- | Num(_, pos)
- | Ident(_, _, pos)
- -> pos
-
- | Package e
- | Ref(_, e)
- | Deref(_, e)
- | Sub_declaration(e, _, _, _)
- | Deref_with(_, _, e, _)
- | Use(e, _)
- | Call(e, _)
- | Method_call(_, e, _)
- -> get_pos_from_expr e
-
- | Diamond(option_e)
- -> if option_e = None then raw_pos2pos bpos else get_pos_from_expr (some option_e)
-
- | List l
- | Block l
- -> if l = [] then raw_pos2pos bpos else get_pos_from_expr (List.hd l)
-
- | Semi_colon
- | Too_complex
- | Undef
- | Label _
- -> raw_pos2pos bpos
-
-let msg_with_rawpos (start, end_) msg = Info.pos2sfull_current start end_ ^ msg
-let die_with_rawpos raw_pos msg = failwith (msg_with_rawpos raw_pos msg)
-let warn warn_types raw_pos msg = if Flags.are_warning_types_set warn_types then print_endline_flush (msg_with_rawpos raw_pos msg)
-
-let die_rule msg = die_with_rawpos (Parsing.symbol_start(), Parsing.symbol_end()) msg
-let warn_rule warn_types msg = warn warn_types (Parsing.symbol_start(), Parsing.symbol_end()) msg
-
-let warn_verb warn_types pos msg = if not !Flags.quiet then warn warn_types (pos, pos) msg
-let warn_too_many_space start = warn_verb [Warn_white_space] start "you should have only one space here"
-let warn_no_space start = warn_verb [Warn_white_space] start "you should have a space here"
-let warn_cr start = warn_verb [Warn_white_space] start "you should not have a carriage-return (\\n) here"
-let warn_space start = warn_verb [Warn_white_space] start "you should not have a space here"
-
-let rec prio_less = function
- | P_none, _ | _, P_none -> internal_error "prio_less"
-
- | P_paren_wanted prio1, prio2
- | prio1, P_paren_wanted prio2 -> prio_less(prio1, prio2)
-
- | P_ternary, P_or -> false
- | P_ternary, P_and -> false
-
- | _, P_loose -> true
- | P_loose, _ -> false
- | _, P_or -> true
- | P_or, _ -> false
-
- | _, P_and -> true
- | P_and, _ -> false
- | _, P_call_no_paren -> true
- | P_call_no_paren, _ -> false
- | _, P_comma -> true
- | P_comma, _ -> false
- | _, P_assign -> true
- | P_assign, _ -> false
- | _, P_ternary -> true
- | P_ternary, _ -> false
-
- | _, P_tight_or -> true
- | P_tight_or, _ -> false
- | _, P_tight_and -> true
- | P_tight_and, _ -> false
-
- | P_bit, P_bit -> true
- | P_bit, _ -> false
-
- | _, P_expr -> true
- | P_expr, _ -> false
-
- | _, P_eq -> true
- | P_eq, _ -> false
- | _, P_cmp -> true
- | P_cmp, _ -> false
- | _, P_uniop -> true
- | P_uniop, _ -> false
- | _, P_add -> true
- | P_add, _ -> false
- | _, P_mul -> true
- | P_mul, _ -> false
- | _, P_tight -> true
- | P_tight, _ -> false
-
- | _, P_paren _ -> true
- | P_paren _, _ -> true
- | P_tok, _ -> true
-
-let prio_lo_check pri_out pri_in pos expr =
- if prio_less(pri_in, pri_out) then
- (match pri_in with
- | P_paren (P_paren_wanted _) -> ()
- | P_paren pri_in' ->
- if pri_in' <> pri_out &&
- prio_less(pri_in', pri_out) && not_complex (un_parenthesize expr) then
- warn [Warn_suggest_simpler] pos "unneeded parentheses"
- | _ -> ())
- else
- (match expr with
- | Call(Deref(I_func, Ident(None, f, _)), _) when f <> "delete" && pri_in = P_uniop && pri_out = P_add
- -> () (* ugly special case since we don't parse uniop correctly (eg: -d $_ . "foo" *)
- | Call_op ("print", [Deref (I_star, Ident (None, "STDOUT", _)); (Deref(I_scalar, _) as ident)], _) ->
- warn [Warn_traps] pos (sprintf "use parentheses: replace \"print %s ...\" with \"print(%s ...)\"" (string_of_fromparser ident) (string_of_fromparser ident))
- | _ -> warn [Warn_traps] pos "missing parentheses (needed for clarity)")
-
-let prio_lo pri_out in_ = prio_lo_check pri_out in_.any.priority in_.pos in_.any.expr ; in_.any.expr
-
-let prio_lo_after pri_out in_ =
- if in_.any.priority = P_call_no_paren then in_.any.expr else prio_lo pri_out in_
-
-let prio_lo_concat esp = prio_lo P_mul { esp with any = { esp.any with priority = P_paren_wanted esp.any.priority } }
-
-let hash_ref esp = Ref(I_hash, prio_lo P_loose esp)
-
-let sp_0 esp =
- match esp.spaces with
- | Space_none -> ()
- | Space_0 -> ()
- | Space_1
- | Space_n -> warn_space (get_pos_start esp)
- | Space_cr -> warn_cr (get_pos_start esp)
-
-let sp_0_or_cr esp =
- match esp.spaces with
- | Space_none -> ()
- | Space_0 -> ()
- | Space_1
- | Space_n -> warn_space (get_pos_start esp)
- | Space_cr -> ()
-
-let sp_1 esp =
- match esp.spaces with
- | Space_none -> ()
- | Space_0 -> warn_no_space (get_pos_start esp)
- | Space_1 -> ()
- | Space_n -> warn_too_many_space (get_pos_start esp)
- | Space_cr -> warn_cr (get_pos_start esp)
-
-let sp_n esp =
- match esp.spaces with
- | Space_none -> ()
- | Space_0 -> warn_no_space (get_pos_start esp)
- | Space_1 -> ()
- | Space_n -> ()
- | Space_cr -> warn_cr (get_pos_start esp)
-
-let sp_p esp =
- match esp.spaces with
- | Space_none -> ()
- | Space_0 -> warn_no_space (get_pos_start esp)
- | Space_1 -> ()
- | Space_n -> ()
- | Space_cr -> ()
-
-let sp_cr esp =
- match esp.spaces with
- | Space_none -> ()
- | Space_0
- | Space_1
- | Space_n -> warn_verb [Warn_white_space] (get_pos_start esp) "you should have a carriage-return (\\n) here"
- | Space_cr -> ()
-
-let sp_same esp1 esp2 =
- if esp1.spaces <> Space_0 then sp_p esp2
- else if esp2.spaces <> Space_0 then sp_p esp1
-
-let function_to_context word_alone = function
- | "map" | "grep" | "grep_index" | "map_index" | "uniq" | "uniq_" -> M_array
- | "partition" -> M_tuple [ M_ref M_array ; M_ref M_array ]
- | "find" -> M_unknown_scalar
- | "any" | "every" -> M_bool
- | "find_index" -> M_int
- | "each_index" -> M_none
- | "N" | "N_" -> M_string
-
- | "chop" | "chomp" | "push" | "unshift" -> M_none
- | "hex" | "length" | "time" | "fork" | "getppid" -> M_int
- | "eof" | "wantarray" -> M_int
- | "stat" | "lstat" -> M_list
- | "arch" | "quotemeta" | "join" | "lc" | "lcfirst" | "uc" | "ucfirst" -> M_string
-
- | "split" -> M_array
- | "shift" | "pop" -> M_unknown_scalar
- | "die" | "return" | "redo" | "next" | "last" -> M_unknown
- | "caller" -> M_mixed [M_string ; M_list]
-
- | "ref" -> M_ref M_unknown_scalar
- | "undef" -> if word_alone then M_undef else M_none
- | _ -> M_unknown
-
-let word_alone esp =
- let word = esp.any in
- let mcontext, e = match word with
- | Ident(None, f, pos) ->
- let e = match f with
- | "length" | "stat" | "lstat" | "chop" | "chomp" | "quotemeta" | "lc" | "lcfirst" | "uc" | "ucfirst" ->
- Call(Deref(I_func, word), [var_dollar_ pos])
-
- | "split" -> Call(Deref(I_func, word), [ Raw_string(" ", pos) ; var_dollar_ pos ])
- | "die" -> Call(Deref(I_func, word), [ Deref(I_scalar, Ident(None, "@", raw_pos2pos bpos)) ])
- | "return" | "eof" | "caller"
- | "redo" | "next" | "last" ->
- Deref(I_func, word)
-
- | "hex" | "ref" ->
- warn_rule [Warn_complex_expressions] (sprintf "please use \"%s $_\" instead of \"%s\"" f f) ;
- Call(Deref(I_func, word), [ Raw_string(" ", pos) ; var_dollar_ pos ])
- | "time" | "wantarray" | "fork" | "getppid" | "arch" ->
- warn_rule [Warn_complex_expressions] (sprintf "please use %s() instead of %s" f f) ;
- Deref(I_func, word)
- | _ -> word
- in
- function_to_context true f, e
- | _ -> M_unknown, word
- in
- new_pesp mcontext P_tok e esp esp
-
-let check_parenthesized_first_argexpr word esp =
- let want_space = word.[0] = '-' in
- if word = "return" then () else
- match esp.any.expr with
- | [ Call_op(_, (e' :: l), _) ]
- | e' :: l ->
- if is_parenthesized e' then
- if l = [] then
- (if want_space then sp_n else sp_0) esp
- else
- (* eg: join (" ", @l) . "\n" *)
- die_with_rawpos (get_pos_start esp, get_pos_start esp) "please remove the space before the function call"
- else
- sp_p esp
- | _ ->
- if word = "time" then die_rule "please use time() instead of time";
- sp_p esp
-
-let check_parenthesized_first_argexpr_with_Ident ident esp =
- if esp.any.priority = P_tok then ();
- (match ident with
- | Ident(Some _, _, _) ->
- (match esp.any.expr with
- | [e] when is_parenthesized e -> ()
- | _ -> warn_rule [Warn_suggest_simpler] "use parentheses around argument (otherwise it might cause syntax errors if the package is \"require\"d and not \"use\"d")
- | Ident(None, word, _) when List.mem word ["ref" ; "readlink"] ->
- if esp.any.priority <> P_tok then warn_rule [Warn_complex_expressions] "use parentheses around argument"
- | _ -> ());
- check_parenthesized_first_argexpr (string_of_fromparser ident) esp
-
-let check_hash_subscript esp =
- let can_be_raw_string = function
- | "" | "x" | "y" -> false (* special case for {'y'} otherwise the emacs mode goes wild, special case for {'x'} to have the same as {'y'} (since they usually go together) *)
- | s ->
- char_is_alpha s.[0] && (String.length s = 1 || string_forall_with char_is_alphanumerical_ 1 s)
- in
- match esp.any.expr with
- | List [String ([(s, List [])], _)] when can_be_raw_string s -> warn [Warn_suggest_simpler] esp.pos (sprintf "{\"%s\"} can be written {%s}" s s)
- | List [Raw_string(s, _)] when can_be_raw_string s -> warn [Warn_suggest_simpler] esp.pos (sprintf "{'%s'} can be written {%s}" s s)
- | _ -> ()
-
-let check_arrow_needed arrow = function
- | Deref_with(I_array, I_scalar, List [List [Call _]], _) -> () (* "->" needed for (f())[0]->{XX} *)
- | Deref_with _ -> warn [Warn_suggest_simpler] arrow.pos "the arrow \"->\" is unneeded"
- | _ -> ()
-
-let check_scalar_subscripted esp =
- match esp.any with
- | Deref(I_scalar, Deref _) -> warn_rule [Warn_complex_expressions] "for complex dereferencing, use \"->\""
- | _ -> ()
-
-let negatable_ops = collect (fun (a, b) -> [ a, b ; b, a ]) [
- "==", "!=" ;
- "eq", "ne" ;
-]
-
-let check_negatable_expr esp =
- match un_parenthesize_full esp.any.expr with
- | Call_op("m//", var :: _, _) when not (is_var_dollar_ var) ->
- warn_rule [Warn_suggest_simpler] "!($var =~ /.../) is better written $var !~ /.../"
- | Call_op("!m//", var :: _, _) when not (is_var_dollar_ var) ->
- warn_rule [Warn_suggest_simpler] "!($var !~ /.../) is better written $var =~ /.../"
- | Call_op(op, _, _) ->
- (try
- let neg_op = List.assoc op negatable_ops in
- warn_rule [Warn_suggest_simpler] (Printf.sprintf "!($foo %s $bar) is better written $foo %s $bar" op neg_op)
- with Not_found -> ())
- | _ -> ()
-
-let check_ternary_paras(cond, a, b) =
- let rec dont_need_short_circuit_rec = function
- | Num _
- | Raw_string _
- | String ([(_, List [])], _)
- -> true
- | Call(Deref(I_func, Ident(None, "N", _)), [ List(String _ :: l) ])
- | Call_op(".", l, _)
- | Ref(I_hash, List l)
- | List l -> List.for_all dont_need_short_circuit_rec l
- | _ -> false
- in
- let rec dont_need_short_circuit = function
- | Ref(_, Deref(_, Ident _))
- | Deref(_, Ident _) -> true
- | Ref(I_hash, List l)
- | List l -> List.for_all dont_need_short_circuit l
- | e -> dont_need_short_circuit_rec e
- in
- let check_ternary_para = function
- | List [] -> warn_rule [Warn_suggest_simpler; Warn_MDK_Common] "you may use if_() here\n beware that the short-circuit semantic of ?: is not kept\n if you want to keep the short-circuit behaviour, replace () with @{[]} and there will be no warning anymore"
- | _ -> ()
- in
- if dont_need_short_circuit a || is_same_fromparser cond a then check_ternary_para b;
- if dont_need_short_circuit b || is_same_fromparser cond b then check_ternary_para a;
- if is_same_fromparser cond a && is_a_scalar a && is_a_scalar b then warn_rule [Warn_suggest_simpler] "you can replace \"$foo ? $foo : $bar\" with \"$foo || $bar\"";
- [ cond; a; b ]
-
-let check_unneeded_var_dollar_ esp =
- if is_var_dollar_ esp.any.expr then warn [Warn_suggest_simpler] esp.pos "\"$_ =~ /regexp/\" can be written \"/regexp/\"" else
- if is_var_number_match esp.any.expr then warn [Warn_complex_expressions] esp.pos "do not use the result of a match (eg: $1) to match another pattern"
-let check_unneeded_var_dollar_not esp =
- if is_var_dollar_ esp.any.expr then warn [Warn_suggest_simpler] esp.pos "\"$_ !~ /regexp/\" can be written \"!/regexp/\"" else
- if is_var_number_match esp.any.expr then warn [Warn_complex_expressions] esp.pos "do not use the result of a match (eg: $1) to match another pattern"
-let check_unneeded_var_dollar_s esp =
- let expr = esp.any.expr in
- if is_var_dollar_ expr then warn [Warn_suggest_simpler] esp.pos "\"$_ =~ s/regexp/.../\" can be written \"s/regexp/.../\"" else
- if is_var_number_match expr then warn [Warn_traps] esp.pos "do not modify the result of a match (eg: $1)" else
- let expr = match expr with
- | List [List [Call_op("=", [ expr; _], _)]] -> expr (* check $xx in ($xx = ...) =~ ... *)
- | _ -> expr in
- if is_a_string expr || not (is_a_scalar expr) then warn [Warn_complex_expressions] esp.pos "you can only use s/// on a variable"
-
-let check_my esp = if esp.any <> "my" then die_rule "syntax error"
-let check_foreach esp = if esp.any = "for" then warn [Warn_normalized_expressions] esp.pos "write \"foreach\" instead of \"for\""
-let check_for esp = if esp.any = "foreach" then warn [Warn_normalized_expressions] esp.pos "write \"for\" instead of \"foreach\""
-let check_for_foreach esp arg =
- match arg.any.expr with
- | List [ Deref(I_scalar, _) ] ->
- if esp.any = "foreach" then warn [Warn_normalized_expressions] esp.pos "you are using the special trick to locally set $_ with a value, for this please use \"for\" instead of \"foreach\""
- | List [ Deref_with(context, I_scalar, _, _) ] when context <> I_func ->
- if esp.any = "foreach" then warn [Warn_normalized_expressions] esp.pos "you are using the special trick to locally set $_ with a value, for this please use \"for\" instead of \"foreach\""
- | List [ Deref(I_hash, _) ] ->
- warn [Warn_traps] esp.pos "foreach with a hash is usually an error"
- | _ ->
- if esp.any = "for" then warn [Warn_normalized_expressions] esp.pos "write \"foreach\" instead of \"for\""
-
-let check_block_expr has_semi_colon last_expr esp_last esp_BRACKET_END =
- sp_p esp_BRACKET_END ;
-
- if esp_BRACKET_END.spaces = Space_cr then
- (if not has_semi_colon then warn_verb [Warn_white_space] (get_pos_end esp_last) "missing \";\"")
- else
- (if last_expr = Semi_colon then warn_verb [Warn_white_space] (get_pos_end esp_last) "spurious \";\" before closing block")
-
-let check_block_lines esp_lines esp_BRACKET_END =
- match fst esp_lines.any with
- | [] ->
- sp_0_or_cr esp_BRACKET_END
- | l ->
- (if List.hd l = Semi_colon then sp_0 else sp_p) esp_lines ;
- check_block_expr (snd esp_lines.any) (last l) esp_lines esp_BRACKET_END
-
-let check_unless_else elsif else_ =
- if elsif.any <> [] then warn [Warn_complex_expressions] elsif.pos "don't use \"elsif\" with \"unless\" (replace \"unless\" with \"if\")";
- if else_.any <> [] then warn [Warn_complex_expressions] else_.pos "don't use \"else\" with \"unless\" (replace \"unless\" with \"if\")"
-
-let check_my_our_paren { any = ((comma_closed, _), l) } after_esp =
- (if l = [] then sp_0 else sp_1) after_esp ;
- if not comma_closed then die_rule "syntax error"
-
-let check_simple_pattern = function
- | [ String([ st, List [] ], _); Raw_string("", _) ] ->
- if String.length st > 2 &&
- st.[0] = '^' && st.[String.length st - 1] = '$' then
- let st = skip_n_char_ 1 1 st in
- if string_forall_with char_is_alphanumerical_ 0 st then
- warn_rule [Warn_suggest_simpler] (sprintf "\"... =~ /^%s$/\" is better written \"... eq '%s'\"" st st)
- | _ -> ()
-
-let rec only_one esp =
- match esp.any with
- | [List l'] -> only_one { esp with any = l' }
- | [e] -> e
- | [] -> die_with_rawpos esp.pos "you must give one argument"
- | _ -> die_with_rawpos esp.pos "you must give only one argument"
-
-let only_one_array_ref esp =
- let e = only_one esp in
- (match e with
- | Call_op("last_array_index", [Deref(I_array, e)], _) ->
- warn [Warn_suggest_simpler] esp.pos (sprintf "you can replace $#%s with -1" (string_of_fromparser e))
- | _ -> ());
- e
-
-let only_one_in_List esp =
- match esp.any.expr with
- | List l -> only_one { esp with any = l }
- | e -> e
-
-let rec is_only_one_in_List = function
- | [List l] -> is_only_one_in_List l
- | [_] -> true
- | _ -> false
-
-let maybe_to_Raw_string = function
- | Ident(None, s, pos) -> Raw_string(s, pos)
- | Ident(Some fq, s, pos) -> Raw_string(fq ^ "::" ^ s, pos)
- | e -> e
-
-let to_List = function
- | [e] -> e
- | l -> List l
-
-let deref_arraylen e = Call_op("last_array_index", [Deref(I_array, e)], raw_pos2pos bpos)
-let deref_raw context e =
- let e = match e with
- | Raw_string(s, pos) ->
- let fq, ident = split_name_or_fq_name s in
- Ident(fq, ident, pos)
- | Deref(I_scalar, (Ident _ as ident)) ->
- warn_rule [Warn_suggest_simpler] (sprintf "%s{$%s} can be written %s$%s" (context2s context) (string_of_fromparser ident) (context2s context) (string_of_fromparser ident));
- e
- | _ -> e
- in Deref(context, e)
-
-let to_Ident { any = (fq, name); pos = pos } = Ident(fq, name, raw_pos2pos pos)
-let to_Raw_string { any = s; pos = pos } = Raw_string(s, raw_pos2pos pos)
-let to_Method_call (object_, method_, para) =
- match method_ with
- | Ident(Some "SUPER", name, pos) -> Method_call(maybe_to_Raw_string object_, Raw_string(name, pos), para)
- | Ident(Some _, _, _) -> Call(Deref(I_func, method_), maybe_to_Raw_string object_ :: para)
- | _ -> Method_call(maybe_to_Raw_string object_, maybe_to_Raw_string method_, para)
-let to_Deref_with(from_context, to_context, ref_, para) =
- if is_not_a_scalar ref_ then warn_rule [] "bad deref";
- Deref_with(from_context, to_context, ref_, para)
-
-let to_Deref_with_arrow arrow (from_context, to_context, ref_, para) =
- if from_context != I_func then check_arrow_needed arrow ref_ ;
- to_Deref_with(from_context, to_context, ref_, para)
-
-let lines_to_Block esp_lines esp_BRACKET_END =
- check_block_lines esp_lines esp_BRACKET_END;
- Block (fst esp_lines.any)
-
-let to_Local esp =
- let l =
- match esp.any.expr with
- | List[List l] -> l
- | e -> [e]
- in
- let local_vars, local_exprs = fpartition (function
- | Deref(I_star as context, Ident(None, ident, _))
- | Deref(I_scalar as context, Ident(None, ("_" as ident), _)) ->
- Some(context, ident)
- | Deref(I_scalar, Ident _)
- | Deref(I_array, Ident _)
- | Deref(I_star, Ident _)
- | Deref_with(I_hash, I_scalar, Ident _, _)
- | Deref_with(I_hash, I_scalar, Deref(I_scalar, _), _)
- | Deref_with(I_hash, I_scalar, Deref_with(I_hash, I_scalar, Ident _, _), _)
- | Deref_with(I_hash, I_scalar, Deref_with(I_hash, I_scalar, Deref(I_scalar, Ident _), _), _) ->
- None
- | _ -> die_with_rawpos esp.pos "bad argument to \"local\""
- ) l in
- if local_vars = [] then Call_op("local", local_exprs, raw_pos2pos esp.pos)
- else if local_exprs = [] then My_our("local", local_vars, raw_pos2pos esp.pos)
- else die_with_rawpos esp.pos "bad argument to \"local\""
-
-let sub_declaration (name, proto) body sub_kind = Sub_declaration(name, proto, Block body, sub_kind)
-let anonymous_sub proto lines bracket_end = Anonymous_sub (proto, lines_to_Block lines bracket_end, raw_pos2pos lines.pos)
-let call_with_same_para_special f = Call(f, [Deref(I_star, (Ident(None, "_", raw_pos2pos bpos)))])
-let remove_call_with_same_para_special = function
- | Call(f, [Deref(I_star, (Ident(None, "_", _)))]) -> f
- | e -> e
-
-let check_My_under_condition msg = function
- | List [ My_our("my", _, _) ] ->
- warn_rule [Warn_traps] "this is stupid"
- | List [ Call_op("=", [ My_our("my", _, _); _ ], _) ] ->
- warn_rule [Warn_traps] msg
- | _ -> ()
-
-let cook_call_op op para pos =
- (match op with
- | "le" | "ge" | "eq" | "ne" | "gt" | "lt" | "cmp" ->
- if List.exists (function Num _ -> true | _ -> false) para then
- warn_rule [Warn_traps] (sprintf "you should use a number operator, not the string operator \"%s\" (or replace the number with a string)" op)
- | "." ->
- if List.exists (function Call(Deref(I_func, Ident(None, "N_", _)), _) -> true | _ -> false) para then
- warn_rule [Warn_MDK_Common; Warn_traps] "N_(\"xxx\") . \"yyy\" is dumb since the string \"xxx\" will never get translated"
- | _ -> ());
-
- (match op, para with
- | "if", List [Call_op ("=", [ _; e ], _)] :: _ when is_always_true e || is_always_false e ->
- warn_rule [Warn_traps] "are you sure you did not mean \"==\" instead of \"=\"?"
-
- | "foreach", [ _; Block [ expr ; Semi_colon ] ]
- | "foreach", [ _; Block [ expr ] ] ->
- (match expr with
- | Call_op("if infix", [ List [ Call(Deref(I_func, Ident(None, "push", _)), [ Deref(I_array, Ident _) as l ; Deref(I_scalar, Ident(None, "_", _)) ]) ] ; _ ], _) ->
- let l = string_of_fromparser l in
- warn_rule [Warn_suggest_functional] (sprintf "use \"push %s, grep { ... } ...\" instead of \"foreach (...) { push %s, $_ if ... }\"\n or sometimes \"%s = grep { ... } ...\"" l l l)
- | Call_op("if infix", [ List [ Call(Deref(I_func, Ident(None, "push", _)), [ Deref(I_array, Ident _) as l; _ ]) ] ; _ ], _) ->
- let l = string_of_fromparser l in
- warn_rule [Warn_suggest_functional] (sprintf "use \"push %s, map { ... ? ... : () } ...\" instead of \"foreach (...) { push %s, ... if ... }\"\n or sometimes \"%s = map { ... ? ... : () } ...\"\n or sometimes \"%s = map { if_(..., ...) } ...\"" l l l l)
-
- | Call_op ("if", [ _; Block [ List [ Call_op("=", [Deref(I_scalar, _) as ret; Deref(I_scalar, Ident(None, "_", _)) ], _) ];
- Semi_colon;
- List [ Deref(I_func, Ident(None, "last", _)) ];
- Semi_colon ] ], _) ->
- warn_rule [Warn_suggest_functional; Warn_MDK_Common] (sprintf "use \"%s = find { ... } ...\"" (string_of_fromparser ret))
-
- | List [ Call(Deref(I_func, Ident(None, "push", _)), [ Deref(I_array, Ident _) as l; _ ]) ] ->
- let l = string_of_fromparser l in
- warn_rule [Warn_suggest_functional] (sprintf "use \"push %s, map { ... } ...\" instead of \"foreach (...) { push %s, ... }\"\n or sometimes \"%s = map { ... } ...\"" l l l)
- | _ -> ())
-
- | "=", [My_our _; Ident(None, "undef", _)] ->
- warn [Warn_suggest_simpler] pos "no need to initialize variable, it's done by default"
- | "=", [My_our _; List[]] ->
- if Info.is_on_same_line_current pos then warn [Warn_suggest_simpler] pos "no need to initialize variables, it's done by default"
-
- | "=", [ Deref_with(I_array, I_scalar, id, Deref(I_array, id_)); _ ] when is_same_fromparser id id_ ->
- warn_rule [Warn_suggest_simpler] "\"$a[@a] = ...\" is better written \"push @a, ...\""
-
- | "=", [ Deref(I_star, String ([(sf1, List [])], _)); _ ] ->
- warn_rule [Warn_help_perl_checker] (sprintf "write *{'%s'} instead of *{\"%s\"}" sf1 sf1)
-
- | "||=", List [ List _ ] :: _
- | "&&=", List [ List _ ] :: _ -> warn_rule [Warn_complex_expressions] "remove the parentheses"
- | "||=", e :: _
- | "&&=", e :: _ -> if is_not_a_scalar e then warn_rule [Warn_traps] (sprintf "\"%s\" is only useful with a scalar" op)
-
- | "==", [Call_op("last_array_index", _, _); Num(n, _)] ->
- warn_rule [Warn_suggest_simpler] (sprintf "$#x == %s is better written @x == %d" n (1 + int_of_string n))
- | "==", [Call_op("last_array_index", _, _); Call_op("- unary", [Num (n, _)], _)] ->
- warn_rule [Warn_suggest_simpler] (sprintf "$#x == -%s is better written @x == %d" n (1 - int_of_string n))
-
-
- | "||", e :: _ when is_always_true e -> warn_rule [Warn_strange] "<constant> || ... is the same as <constant>"
- | "&&", e :: _ when is_always_false e -> warn_rule [Warn_strange] "<constant> && ... is the same as <constant>"
- | "||", e :: _ when is_always_false e -> warn_rule [Warn_strange] "<constant> || ... is the same as ..."
- | "&&", e :: _ when is_always_true e -> warn_rule [Warn_strange] "<constant> && ... is the same as ..."
-
- | "or", e :: _ when is_always_true (un_parenthesize_full e) -> warn_rule [Warn_strange] "<constant> or ... is the same as <constant>"
- | "and", e :: _ when is_always_false (un_parenthesize_full e) -> warn_rule [Warn_strange] "<constant> and ... is the same as <constant>"
- | "or", e :: _ when is_always_false (un_parenthesize_full e) -> warn_rule [Warn_strange] "<constant> or ... is the same as ..."
- | "and", e :: _ when is_always_true (un_parenthesize_full e) -> warn_rule [Warn_strange] "<constant> and ... is the same as ..."
-
- | "or", [ List [ Deref(I_scalar, id) ]; List [ Call_op("=", [ Deref(I_scalar, id_); _], _) ] ] when is_same_fromparser id id_ ->
- warn_rule [Warn_suggest_simpler] "\"$foo or $foo = ...\" can be written \"$foo ||= ...\""
-
- | "and", [ _cond ; expr ] -> check_My_under_condition "replace \"<cond> and my $foo = ...\" with \"my $foo = <cond> && ...\"" expr
- | "or", [ _cond ; expr ] -> check_My_under_condition "replace \"<cond> or my $foo = ...\" with \"my $foo = !<cond> && ...\"" expr
-
- | _ -> ());
-
- match op, para with
- | "=", [ Deref(I_star, (Ident _ as f1)); Deref(I_star, (Ident _ as f2)) ] ->
- let s1, s2 = string_of_fromparser f1, string_of_fromparser f2 in
- warn [Warn_complex_expressions] pos (sprintf "\"*%s = *%s\" is better written \"*%s = \\&%s\"" s1 s2 s1 s2) ;
- sub_declaration (f1, None) [ call_with_same_para_special(Deref(I_func, f2)) ] Glob_assign
- | "=", [ Deref(I_star, Raw_string(sf1, pos_f1)); Deref(I_star, (Ident _ as f2)) ] ->
- let s2 = string_of_fromparser f2 in
- warn [Warn_help_perl_checker] pos (sprintf "\"*{'%s'} = *%s\" is better written \"*{'%s'} = \\&%s\"" sf1 s2 sf1 s2) ;
- sub_declaration (Ident(None, sf1, pos_f1), None) [ call_with_same_para_special(Deref(I_func, f2)) ] Glob_assign
-
- | "=", [ Deref(I_star, (Ident _ as f1)); Ref(I_scalar, Deref(I_func, (Ident _ as f2))) ] ->
- sub_declaration (f1, None) [ call_with_same_para_special(Deref(I_func, f2)) ] Glob_assign
- | "=", [ Deref(I_star, Raw_string(sf1, pos_f1)); Ref(I_scalar, Deref(I_func, (Ident _ as f2))) ] ->
- sub_declaration (Ident(None, sf1, pos_f1), None) [ call_with_same_para_special(Deref(I_func, f2)) ] Glob_assign
-
- | "=", [ Deref(I_star, (Ident _ as f1)); (Anonymous_sub(proto, sub, _)) ] ->
- sub_declaration (f1, proto) [ sub ] Glob_assign
-
- | _ -> Call_op(op, para, raw_pos2pos pos)
-
-let to_Call_op mcontext op para esp_start esp_end =
- let pos = raw_pos_range esp_start esp_end in
- new_any mcontext (cook_call_op op para pos) esp_start.spaces pos
-let to_Call_op_ mcontext prio op para esp_start esp_end =
- let pos = raw_pos_range esp_start esp_end in
- new_any mcontext { priority = prio ; expr = cook_call_op op para pos } esp_start.spaces pos
-let to_Call_assign_op_ mcontext prio op left right esp_left esp_end =
- if not (is_lvalue left) then warn [Warn_strange] esp_left.pos "invalid lvalue";
- to_Call_op_ mcontext prio op [ left ; right ] esp_left esp_end
-
-let followed_by_comma expr true_comma =
- if true_comma then expr else
- match split_last expr with
- | l, Ident(None, s, pos) -> l @ [Raw_string(s, pos)]
- | _ -> expr
-
-
-let pot_strings = Hashtbl.create 16
-let po_comments = ref []
-let po_comment esp = lpush po_comments esp.any
-
-let check_format_a_la_printf s pos =
- let rec check_format_a_la_printf_ contexts i =
- try
- let i' = String.index_from s i '%' in
- try
- let contexts =
- match s.[i' + 1] with
- | '%' -> contexts
- | 'd' -> M_int :: contexts
- | 's' | 'c' -> M_string :: contexts
- | c -> warn [Warn_strange] (pos + i', pos + i') (sprintf "invalid command %%%c" c); contexts
- in
- check_format_a_la_printf_ contexts (i' + 2)
- with Invalid_argument _ -> warn [Warn_strange] (pos + i', pos + i') "invalid command %" ; contexts
- with Not_found -> contexts
- in check_format_a_la_printf_ [] 0
-
-let generate_pot file =
- let fd = open_out file in
- output_string fd
-("# SOME DESCRIPTIVE TITLE.
-# Copyright (C) YEAR Free Software Foundation, Inc.
-# FIRST AUTHOR <EMAIL@ADDRESS>, YEAR.
-#
-#, fuzzy
-msgid \"\"
-msgstr \"\"
-\"Project-Id-Version: PACKAGE VERSION\\n\"
-\"POT-Creation-Date: " ^ input_line (Unix.open_process_in "date '+%Y-%m-%d %H:%M%z'") ^ "\\n\"
-\"PO-Revision-Date: YEAR-MO-DA HO:MI+ZONE\\n\"
-\"Last-Translator: FULL NAME <EMAIL@ADDRESS>\\n\"
-\"Language-Team: LANGUAGE <LL@li.org>\\n\"
-\"MIME-Version: 1.0\\n\"
-\"Content-Type: text/plain; charset=CHARSET\\n\"
-\"Content-Transfer-Encoding: 8-bit\\n\"
-
-") ;
-
- let rec print_formatted_char = function
- | '"' -> output_char fd '\\'; output_char fd '"'
- | '\t' -> output_char fd '\\'; output_char fd 't'
- | '\\' -> output_char fd '\\'; output_char fd '\\'
- | '\n' -> output_string fd "\\n\"\n\""
- | c -> output_char fd c
- in
- let sorted_pot_strings = List.sort (fun (_, pos_a) (_, pos_b) -> compare pos_a pos_b)
- (Hashtbl.fold (fun k (v, _) l -> (k,v) :: l) pot_strings [] ) in
- List.iter (fun (s, _) ->
- match Hashtbl.find_all pot_strings s with
- | [] -> ()
- | l ->
- List.iter (fun _ -> Hashtbl.remove pot_strings s) l ;
-
- List.iter (fun po_comment -> output_string fd ("#. " ^ po_comment ^ "\n")) (collect snd l);
-
- let pos_l = List.sort compare (List.map fst l) in
- fprintf fd "#: %s\n" (String.concat " " (List.map Info.pos2s_for_po pos_l)) ;
- output_string fd "#, c-format\n" ;
-
- output_string fd (if String.contains s '\n' then "msgid \"\"\n\"" else "msgid \"") ;
- String.iter print_formatted_char s ;
- output_string fd "\"\n" ;
- output_string fd "msgstr \"\"\n\n"
- ) sorted_pot_strings ;
- close_out fd
-
-let check_system_call = function
- | "mkdir" :: l ->
- let has_p = List.exists (str_begins_with "-p") l in
- let has_m = List.exists (str_begins_with "-m") l in
- if has_p && has_m then ()
- else if has_p then warn_rule [Warn_suggest_simpler; Warn_MDK_Common] "you can replace system(\"mkdir -p ...\") with mkdir_p(...)"
- else if has_m then warn_rule [Warn_suggest_simpler; Warn_MDK_Common] "you can replace system(\"mkdir -m <mode> ...\") with mkdir(..., <mode>)"
- else warn_rule [Warn_suggest_simpler; Warn_MDK_Common] "you can replace system(\"mkdir ...\") with mkdir(...)"
- | _ -> ()
-
-let call_raw force_non_builtin_func (e, para) =
- let check_anonymous_block f = function
- | [ Anonymous_sub _ ; Deref (I_hash, _) ] ->
- warn_rule [Warn_strange] ("a hash is not a valid parameter to function " ^ f)
-
- | Anonymous_sub _ :: _ -> ()
- | _ -> warn_rule [Warn_complex_expressions] (sprintf "always use \"%s\" with a block (eg: %s { ... } @list)" f f)
- in
-
- match e with
- | Deref(I_func, Ident(None, f, _)) ->
- (match f with
- | "join" ->
- (match un_parenthesize_full_l para with
- | e :: _ when not (is_a_scalar e) -> warn_rule [Warn_traps] "first argument of join() must be a scalar";
- | [_] -> warn_rule [Warn_traps] "not enough parameters"
- | [_; e] when is_a_scalar e -> warn_rule [Warn_traps] "join('...', $foo) is the same as $foo"
- | _ -> ())
-
- | "length" ->
- if para = [] then warn_rule [Warn_complex_expressions] (sprintf "please use \"%s $_\" instead of \"%s\"" f f) else
- if is_not_a_scalar (List.hd para) then warn_rule [Warn_traps] "never use \"length @l\", it returns the length of the string int(@l)" ;
-
- | "open" ->
- (match para with
- | [ List(Ident(None, name, _) :: _) ]
- | Ident(None, name, _) :: _ ->
- if not (List.mem name [ "STDIN" ; "STDOUT" ; "STDERR" ]) then
- warn_rule [Warn_complex_expressions] (sprintf "use a scalar instead of a bareword (eg: occurrences of %s with $%s)" name name)
- | _ -> ())
-
- | "N" | "N_" ->
- (match para with
- | [ List(String([ s, List [] ], (_, pos_offset, _ as pos)) :: para) ] ->
- if !Flags.generate_pot then (
- Hashtbl.add pot_strings s (pos, !po_comments) ;
- po_comments := []
- ) ;
- let contexts = check_format_a_la_printf s pos_offset in
- if f = "N" then
- if List.length para < List.length contexts then
- warn_rule [Warn_traps; Warn_MDK_Common] "not enough parameters"
- else if List.length para > List.length contexts then
- warn_rule [Warn_traps; Warn_MDK_Common] "too many parameters" ;
- (*if String.contains s '\t' then warn_rule "tabulation in translated string must be written \\\\t";*)
- (*if count_matching_char s '\n' > 10 then warn_rule "long string";*)
- | [ List(String _ :: _) ] -> die_rule "don't use interpolated translated string, use %s or %d instead"
- | _ -> die_rule (sprintf "%s() must be used with a string" f))
-
- | "if_" ->
- (match para with
- | [ List [ _ ] ] -> warn_rule [Warn_traps; Warn_MDK_Common] "not enough parameters";
- | _ -> ())
-
- | "map" ->
- (match para with
-
- | Anonymous_sub(None, Block [ List [ Call(Deref(I_func, Ident(None, "if_", _)),
- [ List [ _ ; Deref(I_scalar, Ident(None, "_", _)) ] ]) ] ], _) :: _ ->
- warn_rule [Warn_suggest_simpler; Warn_MDK_Common] "you can replace \"map { if_(..., $_) }\" with \"grep { ... }\""
- | _ -> check_anonymous_block f para)
-
- | "grep" ->
- (match para with
- | [ Anonymous_sub(None, Block [ List [ Call_op("not", [ Call(Deref(I_func, Ident(None, "member", _)), [ List(Deref(I_scalar, Ident(None, "_", _)) :: _) ]) ], _) ] ], _); _ ] ->
- warn_rule [Warn_suggest_simpler; Warn_MDK_Common] "you can replace \"grep { !member($_, ...) } @l\" with \"difference2([ @l ], [ ... ])\""
- | [ Anonymous_sub(None, Block [ List [ Call(Deref(I_func, Ident(None, "member", _)), [ List(Deref(I_scalar, Ident(None, "_", _)) :: _) ]) ] ], _); _ ] ->
- warn_rule [Warn_suggest_simpler; Warn_MDK_Common] "you can replace \"grep { member($_, ...) } @l\" with \"intersection([ @l ], [ ... ])\""
- | _ -> check_anonymous_block f para)
-
- | "any" ->
- (match para with
- [Anonymous_sub (None, Block
- [ List [ Call_op("eq", [Deref(I_scalar, Ident(None, "_", _)); _ ], _) ] ],
- _); _ ] ->
- warn_rule [Warn_suggest_simpler; Warn_MDK_Common] "you can replace \"any { $_ eq ... } @l\" with \"member(..., @l)\""
- | _ -> check_anonymous_block f para)
-
- | "grep_index" | "map_index" | "partition" | "uniq_"
- | "find"
- | "every"
- | "find_index"
- | "each_index" -> check_anonymous_block f para
-
- | "member" ->
- (match para with
- [ List [ _; Call(Deref(I_func, Ident(None, "keys", _)), _) ] ] ->
- warn_rule [Warn_suggest_simpler; Warn_MDK_Common] "you can replace \"member($xxx, keys %yyy)\" with \"exists $yyy{$xxx}\""
- | _ -> ())
-
- | "pop" | "shift" ->
- (match para with
- | []
- | [ Deref(I_array, _) ]
- | [ List [ Deref(I_array, _) ] ] -> ()
- | _ -> warn_rule [Warn_traps] (f ^ " is expecting an array and nothing else"))
-
- | "push" | "unshift" ->
- (match para with
- | Deref(I_array, _) :: l
- | [ List (Deref(I_array, _) :: l) ] ->
- if l = [] then warn_rule [Warn_traps] ("you must give some arguments to " ^ f)
- | _ -> warn_rule [Warn_traps] (f ^ " is expecting an array"))
-
- | "system" ->
- let fake_string_option_from_expr = function
- | String(l, _) -> Some(String.concat "" (List.map fst l))
- | Raw_string(s, _) -> Some s
- | _ -> None
- in
- (match un_parenthesize_full_l para with
- | [ e ] ->
- (match fake_string_option_from_expr e with
- | Some s ->
- if List.exists (String.contains s) [ '\'' ; char_quote ] &&
- not (List.exists (String.contains s) [ '<' ; '>' ; '&' ; ';']) then
- warn_rule [Warn_complex_expressions] "instead of quoting parameters you should give a list of arguments";
- check_system_call (split_at ' ' s)
- | None -> ())
- | l ->
- let l' = filter_some_with fake_string_option_from_expr l in
- check_system_call l')
- | _ -> ()
- );
-
- let para' = match f with
- | "no" ->
- (match para with
- | [ Ident(_, _, pos) as s ] -> Some [ Raw_string(string_of_fromparser s, pos) ]
- | [ Call(Deref(I_func, (Ident(_, _, pos) as s)), l) ] -> Some(Raw_string(string_of_fromparser s, pos) :: l)
- | _ -> die_rule "use \"no PACKAGE <para>\"")
- | "undef" ->
- (match para with
- | [ Deref(I_star, ident) ] -> Some [ Deref(I_func, ident) ]
- | _ -> None)
-
- | "goto" ->
- (match para with
- | [ Ident(None, s, pos) ] -> Some [ Raw_string(s, pos) ]
- | _ -> None)
-
- | "last" | "next" | "redo" when not force_non_builtin_func ->
- (match para with
- | [ Ident(None, s, pos) ] -> Some [ Raw_string(s, pos) ]
- | _ -> die_rule (sprintf "%s must be used with a raw string" f))
-
- | "split" ->
- (match para with
- | [ List(Call_op("m//", Deref(I_scalar, Ident(None, "_", _)) :: pattern, pos) :: l) ]
- | Call_op("m//", Deref(I_scalar, Ident(None, "_", _)) :: pattern, pos) :: l ->
- Some(Call_op("qr//", pattern, pos) :: l)
- | _ -> None)
-
- | _ -> None
- in Call(e, some_or para' para)
- | _ -> Call(e, para)
-
-let call(e, para) = call_raw false (e, para)
-
-let check_return esp_func esp_para =
- match esp_func.any with
- | Ident(None, "return", _) ->
- prio_lo_check P_call_no_paren esp_para.any.priority esp_para.pos (List esp_para.any.expr)
- | _ -> ()
-
-let call_and_context(e, para) force_non_builtin_func priority esp_start esp_end =
- let context =
- match e with
- | Deref(I_func, Ident(None, f, _)) -> function_to_context false f
- | _ -> M_unknown
- in
- new_pesp context priority (call_raw force_non_builtin_func (e, para)) esp_start esp_end
-
-let call_no_paren esp_func esp_para = check_return esp_func esp_para; call_and_context(Deref(I_func, esp_func.any), esp_para.any.expr) false P_call_no_paren esp_func esp_para
-let call_with_paren esp_func esp_para = check_return esp_func esp_para; call_and_context (Deref(I_func, esp_func.any), esp_para.any.expr) false P_tok esp_func esp_para
-
-let call_func esp_func esp_para =
- call_and_context(esp_func.any, esp_para.any.expr) true P_tok esp_func esp_para
-
-let call_one_scalar_para prio { any = e ; pos = pos } para esp_start esp_end =
- let para' =
- match para with
- | [] ->
- if e = "shift" || e = "pop" then
- [] (* can't decide here *)
- else
- (if not (List.mem e [ "length" ]) then warn_rule [Warn_complex_expressions] (sprintf "please use \"%s $_\" instead of \"%s\"" e e) ;
- [var_dollar_ (raw_pos2pos pos)])
- | _ -> para
- in
- new_pesp M_unknown prio (call(Deref(I_func, Ident(None, e, raw_pos2pos pos)), para')) esp_start esp_end
-
-
-let (current_lexbuf : Lexing.lexbuf option ref) = ref None
-
-let rec list2tokens l =
- let rl = ref l in
- fun lexbuf ->
- match !rl with
- | [] -> internal_error "list2tokens"
- | ((start, end_), e) :: l ->
- (* HACK: fake a normal lexbuf *)
- lexbuf.Lexing.lex_start_p <- { Lexing.dummy_pos with Lexing.pos_cnum = start } ;
- lexbuf.Lexing.lex_curr_p <- { Lexing.dummy_pos with Lexing.pos_cnum = end_ } ;
- rl := l ; e
-
-let parse_tokens parse tokens lexbuf_opt =
- if lexbuf_opt <> None then current_lexbuf := lexbuf_opt ;
- if tokens = [] then [] else
- parse (list2tokens tokens) (some !current_lexbuf)
-
-let parse_interpolated parse l =
- let l' = List.map (fun (s, tokens) -> s, to_List(parse_tokens parse tokens None)) l in
- match split_last l' with
- | pl, ("", List []) -> pl
- | _ -> l'
-
-let to_String parse strict { any = l ; pos = pos } =
- let l' = parse_interpolated parse l in
- (match l' with
- | [ "", List [Deref(I_scalar, Ident(None, ident, _))]] ->
- if ident <> "!" && strict then warn [Warn_suggest_simpler] pos (sprintf "%s is better written without the double quotes" (variable2s(I_scalar, ident)))
- | [ "", List [Deref(I_hash, _)]] ->
- warn [Warn_traps] pos "don't use a hash in string context"
- | [ "", List [Deref(I_array, _)]]
- | [ "", List [Deref_with(I_array, I_array, _, _)]] -> (* for slices like: "@m3[1..$#m3]" *)
- ()
- | [("", _)] ->
- if strict then warn [Warn_suggest_simpler] pos "double quotes are unneeded"
- | _ -> ());
- String(l', raw_pos2pos pos)
-
-let from_PATTERN parse { any = (s, opts) ; pos = pos } =
- let re = parse_interpolated parse s in
- (match List.rev re with
- | (s, List []) :: _ ->
- if str_ends_with s ".*" then
- warn_rule [Warn_suggest_simpler] (sprintf "you can remove \"%s\" at the end of your regexp" ".*")
- else if str_ends_with s ".*$" then
- warn_rule [Warn_suggest_simpler] (sprintf "you can remove \"%s\" at the end of your regexp" ".*$")
- | _ -> ());
- let pattern = [ String(re, raw_pos2pos pos) ; Raw_string(opts, raw_pos2pos pos) ] in
- check_simple_pattern pattern;
- pattern
-
-let from_PATTERN_SUBST parse { any = (s1, s2, opts) ; pos = pos } =
- [ String(parse_interpolated parse s1, raw_pos2pos pos) ;
- String(parse_interpolated parse s2, raw_pos2pos pos) ;
- Raw_string(opts, raw_pos2pos pos) ]
-
-
-let rec mcontext2s = function
- | M_none -> "()"
-
- | M_bool -> "bool"
-
- | M_int -> "int"
- | M_float -> "float"
- | M_string -> "string"
- | M_ref c -> "ref(" ^ mcontext2s c ^ ")"
- | M_revision -> "revision"
- | M_undef -> "undef"
- | M_sub -> "sub"
- | M_unknown_scalar -> "scalar"
-
- | M_tuple l -> "tuple(" ^ String.concat ", " (List.map mcontext2s l) ^ ")"
- | M_list -> "list"
- | M_array -> "array"
- | M_hash -> "hash"
-
- | M_special -> "special"
- | M_unknown -> "unknown"
- | M_mixed l -> String.concat " | " (List.map mcontext2s l)
-
-let rec mcontext_lower c1 c2 =
- match c1, c2 with
- | M_special, _ | _, M_special -> internal_error "M_special in mcontext_compare"
-
- | M_unknown, _
- | _, M_unknown -> true
-
- | M_mixed l, c -> List.exists (fun a -> mcontext_lower a c) l
- | c, M_mixed l -> List.exists (mcontext_lower c) l
-
- | M_none, M_none | M_sub, M_sub | M_hash, M_hash | M_hash, M_bool -> true
- | M_none, _ | M_sub, _ | M_hash, _ -> false
-
- | _, M_list -> true
-
- | M_list, M_bool
- | M_list, M_tuple _
-
- (* M_unknown_scalar is M_mixed [ M_int ; M_float ; M_string ; M_bool ; M_ref _ ; M_revision ; M_undef ] *)
- | M_unknown_scalar, M_int | M_unknown_scalar, M_float | M_unknown_scalar, M_string | M_unknown_scalar, M_bool
- | M_unknown_scalar, M_ref _ | M_unknown_scalar, M_revision | M_unknown_scalar, M_undef | M_unknown_scalar, M_unknown_scalar
-
- | M_array, M_array | M_array, M_int | M_array, M_float | M_array, M_bool | M_array, M_unknown_scalar | M_array, M_tuple _
- | M_int, M_int | M_int, M_float | M_int, M_string | M_int, M_bool | M_int, M_unknown_scalar
- | M_float, M_float | M_float, M_string | M_float, M_bool | M_float, M_unknown_scalar
- | M_string, M_string | M_string, M_bool | M_string, M_unknown_scalar
- | M_bool, M_bool | M_bool, M_unknown_scalar
-
- | M_ref _, M_unknown_scalar
- | M_revision, M_revision | M_revision, M_unknown_scalar
- | M_undef, M_undef | M_undef, M_unknown_scalar
-
- -> true
-
- | M_tuple t1, M_tuple t2 ->
- List.length t1 = List.length t2 && for_all2_true mcontext_lower t1 t2
-
- | M_tuple [c], M_int | M_tuple [c], M_float | M_tuple [c], M_string | M_tuple [c], M_bool
- | M_tuple [c], M_ref _ | M_tuple [c], M_revision | M_tuple [c], M_undef | M_tuple [c], M_unknown_scalar
- -> mcontext_lower c c2
-
-(* | M_ref a, M_ref b -> mcontext_lower a b *)
-
- | _ -> false
-
-let mcontext_is_scalar = function
- | M_unknown -> false
- | c -> mcontext_lower c M_unknown_scalar
-
-let mcontext_to_scalar = function
- | M_array -> M_int
- | c -> if mcontext_is_scalar c then c else M_unknown_scalar
-
-let mcontext_merge_raw c1 c2 =
- match c1, c2 with
- | M_unknown, _ | _, M_unknown -> Some M_unknown
- | M_unknown_scalar, c when mcontext_is_scalar c -> Some M_unknown_scalar
- | c, M_unknown_scalar when mcontext_is_scalar c -> Some M_unknown_scalar
- | M_mixed _, _ | _, M_mixed _ -> internal_error "mcontext_merge_raw"
- | _ ->
- if mcontext_lower c1 c2 then Some c2 else
- if mcontext_lower c2 c1 then Some c1 else
- if c1 = c2 then Some c1 else
- None
-
-let rec mcontext_lmerge_add l = function
- | M_mixed l2 -> List.fold_left mcontext_lmerge_add [] (l2 @ l)
- | c ->
- let rec add_to = function
- | [] -> [c]
- | M_mixed subl :: l -> add_to (subl @ l)
- | c2 :: l ->
- match mcontext_merge_raw c c2 with
- | Some c' -> c' :: l
- | None -> c2 :: add_to l
- in add_to l
-
-let mcontext_lmerge l =
- match List.fold_left mcontext_lmerge_add [] l with
- | [] -> internal_error "mcontext_lmerge"
- | [c] -> c
- | l -> M_mixed l
-
-let mcontext_merge c1 c2 = mcontext_lmerge [ c1 ; c2 ]
-
-let mcontext_lmaybe esp = if esp.any = [] then [] else [esp.mcontext]
-
-let mcontext_check_raw wanted_mcontext mcontext =
- if not (mcontext_lower mcontext wanted_mcontext) then
- warn_rule [Warn_context] (sprintf "context %s is not compatible with context %s" (mcontext2s mcontext) (mcontext2s wanted_mcontext))
-
-let mcontext_check wanted_mcontext esp =
- (match wanted_mcontext with
- | M_list | M_array | M_float | M_mixed [M_array; M_none] | M_tuple _ -> ()
- | _ ->
- match un_parenthesize_full esp.any.expr with
- | Call(Deref(I_func, Ident(None, "grep", _)), _) ->
- warn_rule [Warn_suggest_simpler; Warn_help_perl_checker] (if wanted_mcontext = M_bool then
- "in boolean context, use \"any\" instead of \"grep\"" else
- "you may use \"find\" instead of \"grep\"")
- | _ -> ());
- mcontext_check_raw wanted_mcontext esp.mcontext
-
-let mcontext_check_unop_l wanted_mcontext esp =
- mcontext_check wanted_mcontext { esp with any = { esp.any with expr = List esp.any.expr } }
-
-let mcontext_check_non_none esp =
- if esp.mcontext = M_none then warn_rule [Warn_context] "() context not accepted here"
-
-let mcontext_check_none msg expr esp =
- let rec mcontext_check_none_rec msg expr = function
- | M_none | M_unknown -> ()
- | M_mixed l when List.exists (fun c -> c = M_none) l -> ()
- | M_tuple l ->
- (match expr with
- | [Block [List l_expr]]
- | [List l_expr]
- | [List l_expr ; Semi_colon] ->
- let rec iter = function
- | e::l_expr, mcontext::l ->
- mcontext_check_none_rec (if l = [] then msg else "value is dropped") [e] mcontext ;
- iter (l_expr, l)
- | [], [] -> ()
- | _ -> internal_error "mcontext_check_none"
- in iter (un_parenthesize_full_l l_expr, l)
- | _ -> internal_error "mcontext_check_none")
- | _ ->
- match expr with
- | [List [Num("1", _)]; Semi_colon] -> () (* allow "1;" for package return value. It would be much better to check we are at toplevel, but hell i don't want to wire this information up to here *)
- | [List [Call_op ("<>", [Ident (None, "STDIN", _)], _)]; Semi_colon] -> () (* allow <STDIN> to ask "press return" *)
- | [List [Call(Deref(I_func, Ident(None, "map", _)), _)]; Semi_colon] -> warn_rule [Warn_void] "if you don't use the return value, use \"foreach\" instead of \"map\""
- | _ -> warn [Warn_void] esp.pos msg
- in
- mcontext_check_none_rec msg expr esp.mcontext
-
-(* only returns M_float when there is at least one float *)
-let mcontext_float_or_int l =
- List.iter (mcontext_check_raw M_float) l;
- if List.mem M_float l then M_float else M_int
-
-let mcontext_op_assign left right =
- mcontext_check_non_none right;
-
- let left_mcontext =
- match left.mcontext with
- | M_mixed [ c ; M_none ] -> c
- | c -> c
- in
-
- let wanted_mcontext = match left_mcontext with
- | M_array -> M_list
- | M_hash -> M_mixed [ M_hash ; M_list ]
- | m -> m
- in
- mcontext_check wanted_mcontext right;
-
- let return_mcontext =
- match left_mcontext with
- | M_tuple _ -> M_array
- | c -> c
- in
- mcontext_merge return_mcontext M_none
-
-let mtuple_context_concat c1 c2 =
- match c1, c2 with
- | M_array, _ | _, M_array
- | M_hash, _ | _, M_hash -> M_list
- | M_tuple l, _ -> M_tuple (l @ [c2])
- | _ -> M_tuple [c1 ; c2]
-
-let call_op_if_infix left right esp_start esp_end =
- (match left, right with
- | List [Call_op("=", [Deref(context, _); _], _)], _ when non_scalar_context context -> ()
- | List [Call_op("=", [v; _], _)],
- List [Call_op("not", [v'], _)] when is_same_fromparser v v' ->
- warn_rule [Warn_suggest_simpler] "\"$foo = ... if !$foo\" can be written \"$foo ||= ...\""
- | _ -> ());
-
- mcontext_check_none "value is dropped" [left] esp_start;
- (match right with
- | List [ Num("0", _)] -> () (* allow my $x if 0 *)
- | _ -> check_My_under_condition "replace \"my $foo = ... if <cond>\" with \"my $foo = <cond> && ...\"" left);
-
- let pos = raw_pos_range esp_start esp_end in
- new_any M_none (Call_op("if infix", [ left ; right], raw_pos2pos pos)) esp_start.spaces pos
-
-let call_op_unless_infix left right esp_start esp_end =
- (match left, right with
- | List [Call_op("=", [Deref(context, _); _], _)], _ when non_scalar_context context -> ()
- | List [Call_op("=", [v; _], _)], List [v'] when is_same_fromparser v v' ->
- warn_rule [Warn_suggest_simpler] "\"$foo = ... unless $foo\" can be written \"$foo ||= ...\""
- | _ -> ());
- (match right with
- | List [Call_op(op, _, _)] ->
- (match op with
- | "&&" | "||" | "not" | "ne" | "?:" -> warn_rule [Warn_complex_expressions] "don't use \"unless\" when the condition is complex, use \"if\" instead"
- | _ -> ());
- | _ -> ());
-
- mcontext_check_none "value is dropped" [left] esp_start;
- check_My_under_condition "replace \"my $foo = ... unless <cond>\" with \"my $foo = !<cond> && ...\"" left;
-
- let pos = raw_pos_range esp_start esp_end in
- new_any M_none (Call_op("unless infix", [ left ; right], raw_pos2pos pos)) esp_start.spaces pos
-
-let symops pri para_context return_context op_str left op right =
- sp_same op right;
- let skip_context_check =
- (op_str = "==" || op_str = "!=") && (match left.any.expr, right.any.expr with
- | Deref(I_array, _), List [] -> true (* allow @l == () and @l != () *)
- | _ -> false)
- in
- if op_str <> "==" && op_str <> "!=" && para_context = M_float then
- (match un_parenthesize_full left.any.expr with
- | Call_op("last_array_index", _, _) -> warn_rule [Warn_complex_expressions] "change your expression to use @xxx instead of $#xxx"
- | _ -> ());
-
- if not skip_context_check then
- (mcontext_check para_context left ; mcontext_check para_context right) ;
- to_Call_op_ return_context pri op_str [prio_lo pri left; prio_lo_after pri right] left right
diff --git a/perl_checker.src/parser_helper.mli b/perl_checker.src/parser_helper.mli
deleted file mode 100644
index e820703..0000000
--- a/perl_checker.src/parser_helper.mli
+++ /dev/null
@@ -1,314 +0,0 @@
-val bpos : int * int
-val raw_pos2pos : 'a * 'b -> string * 'a * 'b
-val raw_pos_range :
- 'a Types.any_spaces_pos -> 'b Types.any_spaces_pos -> int * int
-val pos_range :
- 'a Types.any_spaces_pos -> 'b Types.any_spaces_pos -> string * int * int
-val get_pos : 'a Types.any_spaces_pos -> string * int * int
-val get_pos_start : 'a Types.any_spaces_pos -> int
-val get_pos_end : 'a Types.any_spaces_pos -> int
-val var_dollar_ : Types.pos -> Types.fromparser
-val var_STDOUT : Types.fromparser
-val new_any :
- Types.maybe_context ->
- 'a -> Types.spaces -> int * int -> 'a Types.any_spaces_pos
-val new_any_ : 'a -> Types.spaces -> int * int -> 'a Types.any_spaces_pos
-val new_esp :
- Types.maybe_context ->
- 'a ->
- 'b Types.any_spaces_pos ->
- 'c Types.any_spaces_pos -> 'a Types.any_spaces_pos
-val new_1esp : 'a -> 'b Types.any_spaces_pos -> 'a Types.any_spaces_pos
-val new_pesp :
- Types.maybe_context ->
- Types.priority ->
- 'a ->
- 'b Types.any_spaces_pos ->
- 'c Types.any_spaces_pos -> 'a Types.prio_anyexpr Types.any_spaces_pos
-val new_1pesp :
- Types.priority ->
- 'a -> 'b Types.any_spaces_pos -> 'a Types.prio_anyexpr Types.any_spaces_pos
-val default_esp : 'a -> 'a Types.any_spaces_pos
-val default_pesp :
- Types.priority -> 'a -> 'a Types.prio_anyexpr Types.any_spaces_pos
-val split_name_or_fq_name : string -> string option * string
-val is_var_dollar_ : Types.fromparser -> bool
-val is_var_number_match : Types.fromparser -> bool
-val non_scalar_context : Types.context -> bool
-val is_scalar_context : Types.context -> bool
-val is_not_a_scalar : Types.fromparser -> bool
-val is_a_scalar : Types.fromparser -> bool
-val is_a_string : Types.fromparser -> bool
-val is_parenthesized : Types.fromparser -> bool
-val un_parenthesize : Types.fromparser -> Types.fromparser
-val un_parenthesize_full : Types.fromparser -> Types.fromparser
-val un_parenthesize_full_l : Types.fromparser list -> Types.fromparser list
-val is_always_true : Types.fromparser -> bool
-val is_always_false : Types.fromparser -> bool
-val is_lvalue : Types.fromparser -> bool
-val not_complex : Types.fromparser -> bool
-val not_simple : Types.fromparser -> bool
-val context2s : Types.context -> string
-val variable2s : Types.context * string -> string
-val string_of_fromparser : Types.fromparser -> string
-val lstring_of_fromparser : Types.fromparser list -> string
-val lstring_of_fromparser_parentheses : Types.fromparser list -> string
-val is_same_fromparser : Types.fromparser -> Types.fromparser -> bool
-val from_scalar : Types.fromparser Types.any_spaces_pos -> Types.fromparser
-val from_array : Types.fromparser Types.any_spaces_pos -> Types.fromparser
-val get_pos_from_expr : Types.fromparser -> Types.pos
-val msg_with_rawpos : int * int -> string -> string
-val die_with_rawpos : int * int -> string -> 'a
-val warn : Types.warning list -> int * int -> string -> unit
-val die_rule : string -> 'a
-val warn_rule : Types.warning list -> string -> unit
-val warn_verb : Types.warning list -> int -> string -> unit
-val warn_too_many_space : int -> unit
-val warn_no_space : int -> unit
-val warn_cr : int -> unit
-val warn_space : int -> unit
-val prio_less : Types.priority * Types.priority -> bool
-val prio_lo_check :
- Types.priority -> Types.priority -> int * int -> Types.fromparser -> unit
-val prio_lo :
- Types.priority ->
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos ->
- Types.fromparser
-val prio_lo_after :
- Types.priority ->
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos ->
- Types.fromparser
-val prio_lo_concat :
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos ->
- Types.fromparser
-val hash_ref :
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos ->
- Types.fromparser
-val sp_0 : 'a Types.any_spaces_pos -> unit
-val sp_0_or_cr : 'a Types.any_spaces_pos -> unit
-val sp_1 : 'a Types.any_spaces_pos -> unit
-val sp_n : 'a Types.any_spaces_pos -> unit
-val sp_p : 'a Types.any_spaces_pos -> unit
-val sp_cr : 'a Types.any_spaces_pos -> unit
-val sp_same : 'a Types.any_spaces_pos -> 'b Types.any_spaces_pos -> unit
-val function_to_context : bool -> string -> Types.maybe_context
-val word_alone :
- Types.fromparser Types.any_spaces_pos ->
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos
-val check_parenthesized_first_argexpr :
- string ->
- Types.fromparser list Types.prio_anyexpr Types.any_spaces_pos -> unit
-val check_parenthesized_first_argexpr_with_Ident :
- Types.fromparser ->
- Types.fromparser list Types.prio_anyexpr Types.any_spaces_pos -> unit
-val check_hash_subscript :
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos -> unit
-val check_arrow_needed : 'a Types.any_spaces_pos -> Types.fromparser -> unit
-val check_scalar_subscripted : Types.fromparser Types.any_spaces_pos -> unit
-val negatable_ops : (string * string) list
-val check_negatable_expr :
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos -> unit
-val check_ternary_paras :
- Types.fromparser * Types.fromparser * Types.fromparser ->
- Types.fromparser list
-val check_unneeded_var_dollar_ :
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos -> unit
-val check_unneeded_var_dollar_not :
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos -> unit
-val check_unneeded_var_dollar_s :
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos -> unit
-val check_my : string Types.any_spaces_pos -> unit
-val check_foreach : string Types.any_spaces_pos -> unit
-val check_for : string Types.any_spaces_pos -> unit
-val check_for_foreach :
- string Types.any_spaces_pos ->
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos -> unit
-val check_block_expr :
- bool ->
- Types.fromparser ->
- 'a Types.any_spaces_pos -> 'b Types.any_spaces_pos -> unit
-val check_block_lines :
- (Types.fromparser list * bool) Types.any_spaces_pos ->
- 'a Types.any_spaces_pos -> unit
-val check_unless_else :
- 'a list Types.any_spaces_pos -> 'b list Types.any_spaces_pos -> unit
-val check_my_our_paren :
- ((bool * 'a) * 'b list) Types.any_spaces_pos ->
- 'c Types.any_spaces_pos -> unit
-val check_simple_pattern : Types.fromparser list -> unit
-val only_one : Types.fromparser list Types.any_spaces_pos -> Types.fromparser
-val only_one_array_ref :
- Types.fromparser list Types.any_spaces_pos -> Types.fromparser
-val only_one_in_List :
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos ->
- Types.fromparser
-val is_only_one_in_List : Types.fromparser list -> bool
-val maybe_to_Raw_string : Types.fromparser -> Types.fromparser
-val to_List : Types.fromparser list -> Types.fromparser
-val deref_arraylen : Types.fromparser -> Types.fromparser
-val deref_raw : Types.context -> Types.fromparser -> Types.fromparser
-val to_Ident :
- (string option * string) Types.any_spaces_pos -> Types.fromparser
-val to_Raw_string : string Types.any_spaces_pos -> Types.fromparser
-val to_Method_call :
- Types.fromparser * Types.fromparser * Types.fromparser list ->
- Types.fromparser
-val to_Deref_with :
- Types.context * Types.context * Types.fromparser * Types.fromparser ->
- Types.fromparser
-val to_Deref_with_arrow :
- 'a Types.any_spaces_pos ->
- Types.context * Types.context * Types.fromparser * Types.fromparser ->
- Types.fromparser
-val lines_to_Block :
- (Types.fromparser list * bool) Types.any_spaces_pos ->
- 'a Types.any_spaces_pos -> Types.fromparser
-val to_Local :
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos ->
- Types.fromparser
-val sub_declaration :
- Types.fromparser * string option ->
- Types.fromparser list -> Types.sub_declaration_kind -> Types.fromparser
-val anonymous_sub :
- string option ->
- (Types.fromparser list * bool) Types.any_spaces_pos ->
- 'a Types.any_spaces_pos -> Types.fromparser
-val call_with_same_para_special : Types.fromparser -> Types.fromparser
-val remove_call_with_same_para_special : Types.fromparser -> Types.fromparser
-val check_My_under_condition : string -> Types.fromparser -> unit
-val cook_call_op :
- string -> Types.fromparser list -> int * int -> Types.fromparser
-val to_Call_op :
- Types.maybe_context ->
- string ->
- Types.fromparser list ->
- 'a Types.any_spaces_pos ->
- 'b Types.any_spaces_pos -> Types.fromparser Types.any_spaces_pos
-val to_Call_op_ :
- Types.maybe_context ->
- Types.priority ->
- string ->
- Types.fromparser list ->
- 'a Types.any_spaces_pos ->
- 'b Types.any_spaces_pos ->
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos
-val to_Call_assign_op_ :
- Types.maybe_context ->
- Types.priority ->
- string ->
- Types.fromparser ->
- Types.fromparser ->
- 'a Types.any_spaces_pos ->
- 'b Types.any_spaces_pos ->
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos
-val followed_by_comma :
- Types.fromparser list -> bool -> Types.fromparser list
-val pot_strings : (string, (string * int * int) * string list) Hashtbl.t
-val po_comments : string list ref
-val po_comment : string Types.any_spaces_pos -> unit
-val check_format_a_la_printf : string -> int -> Types.maybe_context list
-val generate_pot : string -> unit
-val check_system_call : string list -> unit
-val call_raw :
- bool -> Types.fromparser * Types.fromparser list -> Types.fromparser
-val call : Types.fromparser * Types.fromparser list -> Types.fromparser
-val check_return :
- Types.fromparser Types.any_spaces_pos ->
- Types.fromparser list Types.prio_anyexpr Types.any_spaces_pos -> unit
-val call_and_context :
- Types.fromparser * Types.fromparser list ->
- bool ->
- Types.priority ->
- 'a Types.any_spaces_pos ->
- 'b Types.any_spaces_pos ->
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos
-val call_no_paren :
- Types.fromparser Types.any_spaces_pos ->
- Types.fromparser list Types.prio_anyexpr Types.any_spaces_pos ->
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos
-val call_with_paren :
- Types.fromparser Types.any_spaces_pos ->
- Types.fromparser list Types.prio_anyexpr Types.any_spaces_pos ->
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos
-val call_func :
- Types.fromparser Types.any_spaces_pos ->
- Types.fromparser list Types.prio_anyexpr Types.any_spaces_pos ->
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos
-val call_one_scalar_para :
- Types.priority ->
- string Types.any_spaces_pos ->
- Types.fromparser list ->
- 'a Types.any_spaces_pos ->
- 'b Types.any_spaces_pos ->
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos
-val current_lexbuf : Lexing.lexbuf option ref
-val list2tokens : ((int * int) * 'a) list -> Lexing.lexbuf -> 'a
-val parse_tokens :
- ((Lexing.lexbuf -> 'a) -> Lexing.lexbuf -> 'b list) ->
- ((int * int) * 'a) list -> Lexing.lexbuf option -> 'b list
-val parse_interpolated :
- ((Lexing.lexbuf -> 'a) -> Lexing.lexbuf -> Types.fromparser list) ->
- (string * ((int * int) * 'a) list) list -> (string * Types.fromparser) list
-val to_String :
- ((Lexing.lexbuf -> 'a) -> Lexing.lexbuf -> Types.fromparser list) ->
- bool ->
- (string * ((int * int) * 'a) list) list Types.any_spaces_pos ->
- Types.fromparser
-val from_PATTERN :
- ((Lexing.lexbuf -> 'a) -> Lexing.lexbuf -> Types.fromparser list) ->
- ((string * ((int * int) * 'a) list) list * string) Types.any_spaces_pos ->
- Types.fromparser list
-val from_PATTERN_SUBST :
- ((Lexing.lexbuf -> 'a) -> Lexing.lexbuf -> Types.fromparser list) ->
- ((string * ((int * int) * 'a) list) list *
- (string * ((int * int) * 'a) list) list * string)
- Types.any_spaces_pos -> Types.fromparser list
-val mcontext2s : Types.maybe_context -> string
-val mcontext_lower : Types.maybe_context -> Types.maybe_context -> bool
-val mcontext_is_scalar : Types.maybe_context -> bool
-val mcontext_to_scalar : Types.maybe_context -> Types.maybe_context
-val mcontext_merge_raw :
- Types.maybe_context -> Types.maybe_context -> Types.maybe_context option
-val mcontext_lmerge_add :
- Types.maybe_context list -> Types.maybe_context -> Types.maybe_context list
-val mcontext_lmerge : Types.maybe_context list -> Types.maybe_context
-val mcontext_merge :
- Types.maybe_context -> Types.maybe_context -> Types.maybe_context
-val mcontext_lmaybe :
- 'a list Types.any_spaces_pos -> Types.maybe_context list
-val mcontext_check_raw : Types.maybe_context -> Types.maybe_context -> unit
-val mcontext_check :
- Types.maybe_context ->
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos -> unit
-val mcontext_check_unop_l :
- Types.maybe_context ->
- Types.fromparser list Types.prio_anyexpr Types.any_spaces_pos -> unit
-val mcontext_check_non_none : 'a Types.any_spaces_pos -> unit
-val mcontext_check_none :
- string -> Types.fromparser list -> 'a Types.any_spaces_pos -> unit
-val mcontext_float_or_int : Types.maybe_context list -> Types.maybe_context
-val mcontext_op_assign :
- 'a Types.any_spaces_pos ->
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos ->
- Types.maybe_context
-val mtuple_context_concat :
- Types.maybe_context -> Types.maybe_context -> Types.maybe_context
-val call_op_if_infix :
- Types.fromparser ->
- Types.fromparser ->
- 'a Types.any_spaces_pos ->
- 'b Types.any_spaces_pos -> Types.fromparser Types.any_spaces_pos
-val call_op_unless_infix :
- Types.fromparser ->
- Types.fromparser ->
- 'a Types.any_spaces_pos ->
- 'b Types.any_spaces_pos -> Types.fromparser Types.any_spaces_pos
-val symops :
- Types.priority ->
- Types.maybe_context ->
- Types.maybe_context ->
- string ->
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos ->
- 'a Types.any_spaces_pos ->
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos ->
- Types.fromparser Types.prio_anyexpr Types.any_spaces_pos
diff --git a/perl_checker.src/perl_checker.html.pl b/perl_checker.src/perl_checker.html.pl
deleted file mode 100644
index e90d2eb..0000000
--- a/perl_checker.src/perl_checker.html.pl
+++ /dev/null
@@ -1,168 +0,0 @@
-$s = <<'EOF';
-<head><title>perl_checker</title></head>
-<h1>Goals of perl_checker</h1>
-
-<ul>
-<li> for beginners in perl:
- based on what the programmer is writing,
- <ul>
- <li> suggest better or more standard ways to do the same
- <li> detect wrong code
- <br>
- =&gt; a kind of automatic teacher
- </ul>
-
-<li> for senior programmers:
- detect typos, unused variables, check number
- of parameters, global analysis to check method calls...
-
-<li> enforce the same perl style by enforcing a subset of perl of features.
- In perl <a href="http://c2.com/cgi/wiki?ThereIsMoreThanOneWayToDoIt">There is more than one way to do it</a>.
- In perl_checker's subset of Perl, there is not too many ways to do it.
- This is especially useful for big projects.
- (NB: the subset is chosen to keep a good expressivity)
-</ul>
-
-<h1>Compared to <a href="http://perlcritic.tigris.org/">Perl-Critic</a>
-
-<ul>
-<li>perl_checker use its own OCaml-written perl parser, which is in no way as robust as <a href="http://www.perl.com/pub/a/2005/06/09/ppi.html">PPI</a>.
- A PPI require is to be able to parse non finished perl documents.
- perl_checker is a checker, and it is not a big deal to die horribly on a weird perl expression, telling the programmer what to write instead.
-
-<li>perl_checker is <b>much</b> faster (more than 100 times) (ML pattern matching rulez)
-
-<li>perl_checker checks a lot more things than perlcritic: undeclared variables, unknown functions, unknown methods...
-
-<li>and of course perl_checker checks are different from the Conways's <a href="http://www.oreilly.com/catalog/perlbp/">Perl Best Practices</a>
-</ul>
-
-<h1>Get it</h1>
-
-<a href="http://cvs.mandriva.com/cgi-bin/cvsweb.cgi/soft/perl-MDK-Common/perl_checker.src/">CVS source</a>
-
-<h1>Implemented features</h1>
-
-<dl>
- <dt>white space normalization
- <dd>enforce a similar coding style. In many languages you can find a coding
- style document (eg: <a href="http://www.gnu.org/prep/standards_23.html">the GNU one</a>).
-
- TESTS=force_layout.t
-
- </dd>
- <dt>disallow <i>complex</i> expressions
- <dd>perl_checker try to ban some weird-not-used-a-lot features.
-
- TESTS=syntax_restrictions.t
-
- </dd>
- <dt>suggest simpler expressions
- <dd>when there is a simpler way to write an expression, suggest it. It can
- also help detecting errors.
-
- TESTS=suggest_better.t
-
- </dd>
- <dt>context checks
- <dd>Perl has types associated with variables names, the so-called "context".
- Some expressions mixing contexts are stupid, perl_checker detects them.
-
- TESTS=context.t
-
- </dd>
- <dt>function call check
- <dd>detection of unknown functions or mismatching prototypes (warning: since
- perl is a dynamic language, some spurious warnings may occur when a function
- is defined using stashes).
-
- TESTS=prototype.t
-
- </dd>
- <dt>method call check
- <dd>detection of unknown methods or mismatching prototypes. perl_checker
- doesn't have any idea what the object type is, it simply checks if a method
- with that name and that number of parameters exists.
-
- TESTS=method.t
-
- </dd>
- <dt>return value check
- <dd>dropping the result of a functionnally <i>pure</i> function is stupid.
- using the result of a function returning void is stupid too.
-
- TESTS=return_value.t
-
- </dd>
- <dt>detect some Perl traps
- <dd>some Perl expressions are stupid, and one gets a warning when running
- them with <tt>perl -w</tt>. The drawback are <tt>perl -w</tt> is the lack of
- code coverage, it only detects expressions which are evaluated.
-
- TESTS=various_errors.t
-
-</dl>
-
-<h1>Todo</h1>
-
-Functionalities that would be nice:
-<ul>
- <li> add flow analysis
- <li> maybe a "soft typing" type analysis
- <li> detect places where imperative code can be replaced with
- functional code (already done for some <b>simple</b> loops)
- <li> check the number of returned values when checking prototype compliance
-</ul>
-EOF
-
-my $_rationale = <<'EOF';
-<h1>Rationale</h1>
-
-Perl is a big language, there is <a
-href="http://c2.com/cgi/wiki?ThereIsMoreThanOneWayToDoIt">ThereIsMoreThanOneWayToDoIt</a>.
-It has advantages but also some drawbacks for team project:
-<ul>
- <li> it is hard to learn every special rules. Automatically enforced syntax
- coding rules help learning incrementally
-EOF
-
-use lib ('test', '..');
-use read_t;
-sub get_example {
- my ($file) = @_;
- my @tests = read_t::read_t("test/$file");
- $file =~ s|test/||;
- qq(<p><a name="$file"><table border=1 cellpadding=3>\n) .
- join('', map {
- my $lines = join("<br>", map { "<tt>" . html_quote($_) . "</tt>" } @{$_->{lines}});
- my $logs = join("<br>", map { html_quote($_) } @{$_->{logs}});
- " <tr><td>\n", $lines, "</td><td>", $logs, "</td></tr>\n";
- } @tests) .
- "</table></a>\n";
-}
-
-sub anchor_to_examples {
- my ($s) = @_;
- $s =~ s!TESTS=(\S+)!(<a href="#$1">examples</a>)!g;
- $s;
-}
-sub fill_in_examples {
- my ($s) = @_;
- $s =~ s!TESTS=(\S+)!get_example($1)!ge;
- $s;
-}
-
-$s =~ s!<h1>Implemented features</h1>(.*)<h1>!
- "<h1>Implemented features</h1>" . anchor_to_examples($1) .
- "<h1>Examples</h1>" . fill_in_examples($1) .
- "<h1>"!se;
-
-print $s;
-
-sub html_quote {
- local $_ = $_[0];
- s/</&lt;/g;
- s/>/&gt;/g;
- s/^(\s*)/"&nbsp;" x length($1)/e;
- $_;
-}
diff --git a/perl_checker.src/perl_checker.ml b/perl_checker.src/perl_checker.ml
deleted file mode 100644
index 4459e30..0000000
--- a/perl_checker.src/perl_checker.ml
+++ /dev/null
@@ -1,183 +0,0 @@
-open Types
-open Common
-open Tree
-open Global_checks
-
-let search_basedir file_name nb =
- let dir = Filename.dirname file_name in
- let config = Config_file.read dir in
- let nb = some_or config.Config_file.basedir nb in
- updir dir nb
-
-let basedir = ref ""
-let set_basedir per_files file =
- if !basedir = "" then
- let nb = List.length (split_at2 ':'':' (List.hd file.packages).package_name) - 1 in
- let dir = search_basedir file.file_name nb in
- lpush Tree.use_lib dir ;
- Config_file.read_any dir 1 ;
- read_packages_from_cache per_files dir ;
- if !Flags.verbose then print_endline_flush ("basedir is " ^ dir);
- basedir := dir
-
-let rec parse_file from_basedir require_name per_files file =
- try
- if !Flags.verbose then print_endline_flush ("parsing " ^ file) ;
- let build_time = Unix.time() in
- let command =
- match !Flags.expand_tabs with
- | Some width -> "expand -t " ^ string_of_int width
- | None -> "cat" in
- let channel = Unix.open_process_in (Printf.sprintf "%s \"%s\"" command file) in
- let lexbuf = Lexing.from_channel channel in
- try
- Info.start_a_new_file file ;
- let tokens = Lexer.get_token Lexer.token lexbuf in
- if not Build.debugging then ignore (Unix.close_process_in channel) ;
- let t = Parser_helper.parse_tokens Parser.prog tokens (Some lexbuf) in
- let per_file = get_global_info_from_package from_basedir require_name build_time t in
- set_basedir per_files per_file ;
- Global_checks.add_file_to_files per_files per_file ;
-
- let required_packages = collect (fun package -> package.required_packages) per_file.packages in
- required_packages, per_files
- with Failure s -> (
- print_endline_flush s ;
- exit 1
- )
- with
- | Not_found -> internal_error "runaway Not_found"
-
-and parse_package_if_needed per_files (package_name, pos) =
- if List.mem package_name !Config_file.ignored_packages then [], per_files else
- let splitted = split_at2 ':'':' package_name in
- let rel_file = String.concat "/" splitted ^ ".pm" in
-
- (*print_endline_flush ("wondering about " ^ package_name) ;*)
- try
- let dir = findfile (Build.fake_packages_dir :: !use_lib) rel_file in
- let file = Info.file_to_absolute_file (dir ^ "/" ^ rel_file) in
- Config_file.read_any (Filename.dirname file) (List.length splitted) ;
- let already_done =
- try
- let per_file = Hashtbl.find per_files file in
- Some (collect (fun pkg -> pkg.required_packages) per_file.packages)
- with Not_found -> None in
- match already_done with
- | Some required_packages -> required_packages, per_files
- | None -> parse_file (dir = !basedir) (Some package_name) per_files file
- with Not_found ->
- print_endline_flush (Info.pos2sfull pos ^ Printf.sprintf "can't find package %s" package_name) ;
- [], per_files
-
-let rec parse_required_packages state already_done = function
- | [] -> state, already_done
- | e :: l ->
- if List.mem e already_done then
- parse_required_packages state already_done l
- else
- let el, state = parse_package_if_needed state e in
- parse_required_packages state (e :: already_done) (el @ l)
-
-
-let parse_options =
- let args_r = ref [] in
- let restrict_to_files = ref false in
-
- let pot_file = ref "" in
- let package_dependencies_graph_file = ref "" in
- let generate_pot_chosen file =
- Flags.generate_pot := true ;
- Flags.expand_tabs := None ;
- pot_file := file
- in
- let options = [
- "-v", Arg.Set Flags.verbose, " be verbose" ;
- "-q", Arg.Set Flags.quiet, " be quiet" ;
- "-t", Arg.Int (fun i -> Flags.expand_tabs := Some i), " set the tabulation width (default is 8)" ;
- "--restrict-to-files", Arg.Set restrict_to_files, " only display warnings concerning the file(s) given on command line" ;
- "--no-cache", Arg.Set Flags.no_cache, " do not use cache" ;
- "--generate-pot", Arg.String generate_pot_chosen, "" ;
- "--generate-package-dependencies-graph", Arg.String (fun f -> package_dependencies_graph_file := f),
- "\n" ;
-
- "--check-unused-global-vars", Arg.Set Flags.check_unused_global_vars, " disable unused global functions & variables check" ^
- "\nBasic warnings:";
- "--no-check-white-space", Arg.Clear Flags.check_white_space, " disable white space check" ;
- "--no-suggest-simpler", Arg.Clear Flags.check_suggest_simpler, " disable simpler code suggestion" ;
- "--no-suggest-functional", Arg.Clear Flags.suggest_functional, " disable Functional Programming suggestions" ^
- "\nNormalisation warnings:";
- "--no-check-strange", Arg.Clear Flags.check_strange, " disable strange code check" ;
- "--no-check-complex-expressions", Arg.Clear Flags.check_complex_expressions, " disable complex expressions check" ;
- "--no-check-normalized-expressions", Arg.Clear Flags.normalized_expressions, " don't warn about non normalized expressions" ;
- "--no-help-perl-checker", Arg.Clear Flags.check_help_perl_checker, " beware, perl_checker doesn't understand all perl expressions, so those warnings *are* important" ^
- "\nCommon warnings:";
- "--no-check-void", Arg.Clear Flags.check_void, " disable dropped value check" ;
- "--no-check-names", Arg.Clear Flags.check_names, " disable variable & function usage check" ;
- "--no-check-prototypes", Arg.Clear Flags.check_prototypes, " disable prototypes check" ;
- "--no-check-import-export", Arg.Clear Flags.check_import_export, " disable inter modules check" ^
- "\nImportant warnings:";
- "--no-check-context", Arg.Clear Flags.check_context, " disable context check" ;
- "--no-check-traps", Arg.Clear Flags.check_traps, " disable traps (errors) check" ^
- "\n";
-
- ] in
- let usage = "Usage: perl_checker [<options>] <files>\nOptions are:" in
- Arg.parse options (lpush args_r) usage;
-
- let files = if !args_r = [] && Build.debugging then ["../t.pl"] else !args_r in
- let files = List.map Info.file_to_absolute_file files in
-
- let required_packages, per_files = collect_withenv (parse_file true None) (default_per_files()) files in
- let required_packages = uniq_ (fun (a,_) (b,_) -> a = b) required_packages in
-
- if !Flags.generate_pot then Parser_helper.generate_pot !pot_file else (
-
- let per_files, required_packages =
- fluid_let Flags.quiet (!restrict_to_files || !Flags.quiet)
- (fun () ->
- parse_required_packages per_files [] required_packages) in
- let l_required_packages = List.map fst required_packages in
-
- write_packages_cache per_files !basedir ;
-
- (* removing non needed files from per_files (those files come from the cache) *)
- List.iter (fun k ->
- let per_file = Hashtbl.find per_files k in
- if per_file.require_name <> None && not (List.mem (some per_file.require_name) l_required_packages) && not (List.mem per_file.file_name files) then
- Hashtbl.remove per_files k
- ) (hashtbl_keys per_files);
-
- let state = default_state per_files in
-
- Hashtbl.iter (fun _ per_file -> List.iter (add_package_to_state state) per_file.packages) per_files ;
-
- let state =
- let global_vars_declared = Hashtbl.create 16 in
- let package_name_to_file_name = hashtbl_collect (fun _ per_file -> List.map (fun pkg -> pkg.package_name, per_file.file_name) per_file.packages) per_files in
- Hashtbl.iter (fun _ pkg ->
- let file_name = List.assoc pkg.package_name package_name_to_file_name in
- fluid_let Flags.quiet (!restrict_to_files || !Flags.quiet)
- (fun () -> get_vars_declaration global_vars_declared file_name pkg)
- ) state.per_packages ;
- arrange_global_vars_declared global_vars_declared state
- in
-
- let state = Global_checks.get_methods_available state in
-
- let l = hashtbl_values per_files in
- let l = if !restrict_to_files then List.filter (fun file -> List.mem file.file_name files) l else l in
-
- let l = uniq (collect (fun file -> List.map (fun pkg -> pkg.package_name) file.packages) l) in
- let l = List.map (Hashtbl.find state.per_packages) l in
-
- (* HACK: skip ignored_packages. Some package may have appeared in ignored_packages due to the xs bootstrap hack *)
- let l = List.filter (fun pkg -> not (List.mem pkg.package_name !Config_file.ignored_packages)) l in
-
- List.iter (Global_checks.check_tree state) l;
-
- if !Flags.check_unused_global_vars then List.iter Global_checks.check_unused_vars l;
-
- if !package_dependencies_graph_file <> "" then generate_package_dependencies_graph state !package_dependencies_graph_file
-
- )
diff --git a/perl_checker.src/perl_checker.mli b/perl_checker.src/perl_checker.mli
deleted file mode 100644
index 8b13789..0000000
--- a/perl_checker.src/perl_checker.mli
+++ /dev/null
@@ -1 +0,0 @@
-
diff --git a/perl_checker.src/print.ml b/perl_checker.src/print.ml
deleted file mode 100644
index e69de29..0000000
--- a/perl_checker.src/print.ml
+++ /dev/null
diff --git a/perl_checker.src/print.mli b/perl_checker.src/print.mli
deleted file mode 100644
index 8b13789..0000000
--- a/perl_checker.src/print.mli
+++ /dev/null
@@ -1 +0,0 @@
-
diff --git a/perl_checker.src/test/.cvsignore b/perl_checker.src/test/.cvsignore
deleted file mode 100644
index 9f6633c..0000000
--- a/perl_checker.src/test/.cvsignore
+++ /dev/null
@@ -1,2 +0,0 @@
-.pl
-.perl_checker.cache
diff --git a/perl_checker.src/test/Makefile b/perl_checker.src/test/Makefile
deleted file mode 100644
index abe816c..0000000
--- a/perl_checker.src/test/Makefile
+++ /dev/null
@@ -1,3 +0,0 @@
-
-test:
- for i in *.t; do ./test_it $$i || exit 1; done
diff --git a/perl_checker.src/test/context.t b/perl_checker.src/test/context.t
deleted file mode 100644
index 081abcc..0000000
--- a/perl_checker.src/test/context.t
+++ /dev/null
@@ -1,41 +0,0 @@
-foreach (%h) {} context hash is not compatible with context list
- foreach with a hash is usually an error
-
-map { 'xxx' } %h a hash is not a valid parameter to function map
-
-$xxx = ('yyy', 'zzz') context tuple(string, string) is not compatible with context scalar
-
-@l ||= 'xxx' "||=" is only useful with a scalar
-
-length @l never use "length @l", it returns the length of the string int(@l)
-
-%h . 'yyy' context hash is not compatible with context string
-
-'xxx' > 'yyy' context string is not compatible with context float
- context string is not compatible with context float
-
-
-1 cmp 2 you should use a number operator, not the string operator "cmp" (or replace the number with a string)
-
-$xxx == undef context undef is not compatible with context float
-
-my ($xxx) = 1 context int is not compatible with context tuple(scalar)
-
-($xxx, $yyy) = 1 context int is not compatible with context tuple(scalar, scalar)
-
-($xxx, $yyy) = (1, 2, 3) context tuple(int, int, int) is not compatible with context tuple(scalar, scalar)
-
-@l eq '3' context array is not compatible with context string
-
-qw(a b) > 2 context tuple(string, string) is not compatible with context float
-
-%h > 0 context hash is not compatible with context float
-
-%h eq 0 context hash is not compatible with context string
- you should use a number operator, not the string operator "eq" (or replace the number with a string)
-
-@l == ()
-
-$xxx = { xxx() }->{xxx};
-
-$xxx = { xxx() }->{$xxx};
diff --git a/perl_checker.src/test/force_layout.t b/perl_checker.src/test/force_layout.t
deleted file mode 100644
index bb5494e..0000000
--- a/perl_checker.src/test/force_layout.t
+++ /dev/null
@@ -1,23 +0,0 @@
-sub xxx you should not have a carriage-return (\n) here
-{}
-
-xxx you should not have a carriage-return (\n) here
- ($xxx);
-
-xxx( $xxx) you should not have a space here
-
-$xxx ++ you should not have a space here
-
-my($_xxx, $_yyy) you should have a space here
-
-xxx ($xxx) you should not have a space here
-
-'foo'.'bar' you should have a space here
-
-if ($xxx) { missing ";"
- xxx()
-}
-
-if ($xxx) { unneeded ";"
- xxx();
-};
diff --git a/perl_checker.src/test/method.t b/perl_checker.src/test/method.t
deleted file mode 100644
index e59e858..0000000
--- a/perl_checker.src/test/method.t
+++ /dev/null
@@ -1,11 +0,0 @@
-bad->yyy unknown package bad
-
-pkg->bad unknown method bad starting in package pkg
-
-$xxx->bad unknown method bad
-
-$xxx->m1 not enough parameters
-
-$xxx->m0('zzz') too many parameters
-
-$xxx->m0_or_2('zzz') not enough or too many parameters
diff --git a/perl_checker.src/test/prototype.t b/perl_checker.src/test/prototype.t
deleted file mode 100644
index 6e56aae..0000000
--- a/perl_checker.src/test/prototype.t
+++ /dev/null
@@ -1,23 +0,0 @@
-
-sub xxx { 'yyy' } if the function doesn't take any parameters, please use the empty prototype.
- example "sub foo() { ... }"
-
-sub xxx { an non-optional argument must not follow an optional argument
- my ($o_xxx, $yyy) = @_;
- ($o_xxx, $yyy);
-}
-
-sub xxx { an array must be the last variable in a prototype
- my (@xxx, $yyy) = @_;
- @xxx, $yyy;
-}
-
-bad() unknown function bad
-
-sub f0() {} too many parameters
-f0('yyy')
-
-sub f2 { my ($x, $_y) = @_; $x } not enough parameters
-f2('yyy')
-
-N("xxx %s yyy") not enough parameters
diff --git a/perl_checker.src/test/read_t.pm b/perl_checker.src/test/read_t.pm
deleted file mode 100644
index a07c041..0000000
--- a/perl_checker.src/test/read_t.pm
+++ /dev/null
@@ -1,28 +0,0 @@
-package read_t;
-
-use lib '../..';
-use MDK::Common;
-
-sub read_t {
- my ($file) = @_;
-
- my @tests;
- my ($column_width, $line_number, @lines, @logs);
- foreach (cat_($file), "\n") {
- if (/^$/) {
- push @tests, { line_number => $line_number, lines => [ @lines ], logs => [ @logs ] } if @lines;
- @lines = @logs = ();
- } else {
- $column_width ||= length(first(/(.{20}\s+)/));
- my ($line, $log) = $column_width > 25 && /(.{$column_width})(.*)/ ? (chomp_($1) . "\n", $2) : ($_, '');
- $line =~ s/[ \t]*$//;
- push @lines, $line;
- push @logs, $log;
- }
- $line_number++;
- }
- @tests;
-}
-
-1;
-
diff --git a/perl_checker.src/test/return_value.t b/perl_checker.src/test/return_value.t
deleted file mode 100644
index b4786f5..0000000
--- a/perl_checker.src/test/return_value.t
+++ /dev/null
@@ -1,23 +0,0 @@
-if ($xxx or $yyy) {} value should be dropped
- context () is not compatible with context bool
-
-if ($xxx and $yyy) {} value should be dropped
- context () is not compatible with context bool
-
-$xxx && yyy(); value is dropped
-
-`xxx`; value is dropped
-
-/(.*)/; value is dropped
-
-'xxx'; value is dropped
-
-'xxx' if $xxx; value is dropped
-
-map { xxx($_) } @l; if you don't use the return value, use "foreach" instead of "map"
-
-$xxx = chomp; () context not accepted here
- context () is not compatible with context scalar
-
-$xxx = push @l, 1 () context not accepted here
- context () is not compatible with context scalar
diff --git a/perl_checker.src/test/suggest_better.t b/perl_checker.src/test/suggest_better.t
deleted file mode 100644
index d76abeb..0000000
--- a/perl_checker.src/test/suggest_better.t
+++ /dev/null
@@ -1,112 +0,0 @@
-@{$xxx} @{$xxx} can be written @$xxx
-
-$h{"yyy"} {"yyy"} can be written {yyy}
-
-"$xxx" $xxx is better written without the double quotes
-
-$xxx->{yyy}->{zzz} the arrow "->" is unneeded
-
-"xxx\$xxx" you can replace "xxx\$xxx" with 'xxx$xxx', that way you don't need to escape <$>
-
-"xxx\$xxx" you can replace "xxx\$xxx" with 'xxx$xxx', that way you don't need to escape <$>
-
-"xxx\"$xxx" you can replace "xxx\"xxx" with qq(xxx"xxx), that way you don't need to escape <">
-
-/xxx\'xxx/ you can replace \' with '
-
-/xxx\;xxx/ you can replace \; with ;
-
-/\// change the delimit character / to get rid of this escape
-
-{ nop(); } spurious ";" before closing block
-
-+1 don't use unary +
-
-return ($xxx) unneeded parentheses
-
-if (($xxx eq $yyy) || $zzz) {} unneeded parentheses
-
-if (($xxx =~ /yyy/) || $zzz) {} unneeded parentheses
-
-nop() foreach ($xxx, $yyy); unneeded parentheses
-
-($xxx) ||= 'xxx' remove the parentheses
-
-$o->m0() remove these unneeded parentheses
-
-$o = xxx() if !$o; "$foo = ... if !$foo" can be written "$foo ||= ..."
-
-$o = xxx() unless $o; "$foo = ... unless $foo" can be written "$foo ||= ..."
-
-$o or $o = xxx(); "$foo or $foo = ..." can be written "$foo ||= ..."
-
-$_ =~ s/xxx/yyy/ "$_ =~ s/regexp/.../" can be written "s/regexp/.../"
-
-$xxx =~ /^yyy$/ "... =~ /^yyy$/" is better written "... eq 'yyy'"
-
-/xxx.*/ you can remove ".*" at the end of your regexp
-
-/xxx.*$/ you can remove ".*$" at the end of your regexp
-
-/[^\s]/ you can replace [^\s] with \S
-
-/[^\w]/ you can replace [^\w] with \W
-
-$xxx ? $xxx : $yyy you can replace "$foo ? $foo : $bar" with "$foo || $bar"
-
-my @l = (); no need to initialize variables, it's done by default
-
-$l[$#l] you can replace $#l with -1
-
-$#l == 0 $#x == 0 is better written @x == 1
-
-$#l == -1 $#x == -1 is better written @x == 0
-
-$#l < 0 change your expression to use @xxx instead of $#xxx
-
-$l[@l] = 1 "$a[@a] = ..." is better written "push @a, ..."
-
-xxx(@_) replace xxx(@_) with &xxx
-
-member($xxx, keys %h) you can replace "member($xxx, keys %yyy)" with "exists $yyy{$xxx}"
-
-!($xxx =~ /.../) !($var =~ /.../) is better written $var !~ /.../
-
-!($xxx == 1) !($foo == $bar) is better written $foo != $bar
-
-!($xxx eq 'foo') !($foo eq $bar) is better written $foo ne $bar
-
-grep { !member($_, qw(a b c)) } @l you can replace "grep { !member($_, ...) } @l" with "difference2([ @l ], [ ... ])"
-
-any { $_ eq 'foo' } @l you can replace "any { $_ eq ... } @l" with "member(..., @l)"
-
-foreach (@l) { use "push @l2, grep { ... } ..." instead of "foreach (...) { push @l2, $_ if ... }"
- push @l2, $_ if yyy($_); or sometimes "@l2 = grep { ... } ..."
-}
-
-foreach (@l) { use "push @l2, map { ... } ..." instead of "foreach (...) { push @l2, ... }"
- push @l2, yyy($_); or sometimes "@l2 = map { ... } ..."
-}
-
-foreach (@l) { use "push @l2, map { ... ? ... : () } ..." instead of "foreach (...) { push @l2, ... if ... }"
- push @l2, yyy($_) if zzz($_); or sometimes "@l2 = map { ... ? ... : () } ..."
-} or sometimes "@l2 = map { if_(..., ...) } ..."
-
-foreach (@l) { use "$xxx = find { ... } ..."
- if (xxx($_)) {
- $xxx = $_;
- last;
- }
-}
-
-if (grep { xxx() } @l) {} in boolean context, use "any" instead of "grep"
-
-$xxx = grep { xxx() } @l; you may use "find" instead of "grep"
-
-$xxx ? $yyy : () you may use if_() here
- beware that the short-circuit semantic of ?: is not kept
- if you want to keep the short-circuit behaviour, replace () with @{[]} and there will be no warning anymore
-
-system(qq(foo "$xxx")) instead of quoting parameters you should give a list of arguments
-
-system("mkdir", $xxx) you can replace system("mkdir ...") with mkdir(...)
diff --git a/perl_checker.src/test/syntax_restrictions.t b/perl_checker.src/test/syntax_restrictions.t
deleted file mode 100644
index de7bf77..0000000
--- a/perl_checker.src/test/syntax_restrictions.t
+++ /dev/null
@@ -1,70 +0,0 @@
-$xxx <<= 2 don't use "<<=", use the expanded version instead
-
-m@xxx@ don't use m@...@, replace @ with / ! , or |
-
-s:xxx:yyy: don't use s:...:, replace : with / ! , or |
-
-qw/a b c/ don't use qw/.../, use qw(...) instead
-
-qw{a b c} don't use qw{...}, use qw(...) instead
-
-q{xxx} don't use q{...}, use q(...) instead
-
-qq{xxx} don't use qq{...}, use qq(...) instead
-
-qx(xxx) don't use qx(...), use `...` instead
-
--xxx don't use -xxx, use '-xxx' instead
-
-not $xxx don't use "not", use "!" instead
-
-$1 =~ /xxx/ do not use the result of a match (eg: $1) to match another pattern
-
-$xxx =~ "yyy" use a regexp, not a string
-
-xxx() =~ s/xxx/yyy/ you can only use s/// on a variable
-
-$1 =~ /xxx/ do not use the result of a match (eg: $1) to match another pattern
-
-grep /xxx/, @l always use "grep" with a block (eg: grep { ... } @list)
-
-for (@l) {} write "foreach" instead of "for"
-
-foreach ($xxx = 0; $xxx < 9; $xxx++) {} write "for" instead of "foreach"
-
-foreach $xxx (@l) {} don't use for without "my"ing the iteration variable
-
-foreach ($xxx) {} you are using the special trick to locally set $_ with a value, for this please use "for" instead of "foreach"
-
-unless ($xxx) {} else {} don't use "else" with "unless" (replace "unless" with "if")
-
-unless ($xxx) {} elsif ($yyy) {} don't use "elsif" with "unless" (replace "unless" with "if")
-
-zzz() unless $xxx || $yyy; don't use "unless" when the condition is complex, use "if" instead
-
-$$xxx{yyy} for complex dereferencing, use "->"
-
-wantarray please use wantarray() instead of wantarray
-
-eval please use "eval $_" instead of "eval"
-
-local *F; open F, "foo"; use a scalar instead of a bareword (eg: occurrences of F with $F)
-
-$xxx !~ s/xxx/yyy/ use =~ instead of !~ and negate the return value
-
-pkg::nop $xxx; use parentheses around argument (otherwise it might cause syntax errors if the package is "require"d and not "use"d
-
-new foo $xxx you must parenthesize parameters: "new Class(...)" instead of "new Class ..."
-
-*xxx = *yyy "*xxx = *yyy" is better written "*xxx = \&yyy"
-
-$_xxx = 1 variable $_xxx must not be used
- (variable with name _XXX are reserved for unused variables)
-
-sub f2 { my ($x, $_y) = @_; $x } not enough parameters
-f2(@l); # ok
-f2(xxx()); # bad
-
-$xxx = <<"EOF"; Don't use <<"MARK", use <<MARK instead
-foo
-EOF
diff --git a/perl_checker.src/test/test_it b/perl_checker.src/test/test_it
deleted file mode 100755
index a89c2c5..0000000
--- a/perl_checker.src/test/test_it
+++ /dev/null
@@ -1,113 +0,0 @@
-#!/usr/bin/perl
-
-use lib '../..';
-use MDK::Common;
-use read_t;
-
-my ($file) = @ARGV;
-my @tests = read_t::read_t($file);
-
-output('pkg3.pm', <<'EOF');
-package pkg3;
-our @ISA = qw(Exporter);
-our %EXPORT_TAGS = (
- missing_fs => [ qw(f f0) ],
-);
-our @EXPORT_OK = qw(f);
-EOF
-
-my $header = <<'EOF';
-package pkg;
-use lib "../..";
-sub new {}
-sub m0 { my ($_o) = @_; 0 }
-sub m1 { my ($_o, $a) = @_; $a }
-sub m2 { my ($_o, $_a, $b) = @_; $b }
-sub m0_or_2 { my ($_o, $_a, $b) = @_; $b }
-package pkg2;
-sub new {}
-sub m0_or_2 { my ($_o) = @_; 0 }
-
-package my_pkg;
-sub nop {}
-sub xxx { @_ }
-sub yyy { @_ }
-sub zzz { @_ }
-sub pkg::nop {}
-sub N { $_[0] }
-sub N_ { $_[0] }
-my ($xxx, $yyy, $zzz, $o, @l, @l2, %h);
-xxx($xxx, $yyy, $zzz, $o, @l, @l2, %h);
-use MDK::Common;
-
-EOF
-
-my $oo_header = <<'EOF';
-EOF
-
-my $local = <<'EOF';
-{
- local $_;
-EOF
-
-my $local_trailer = <<'EOF';
-
- xxx($xxx, $yyy, $zzz, $o, @l, @l2, %h);
-}
-EOF
-
-my $new;
-foreach my $test (@tests) {
- my @l = @{$test->{lines}};
-
- pop @l while $l[-1] =~ /^\s*$/;
- if (@l == 1) {
- if ($l[-1] !~ /(;|[\s{]\})\s*$/) {
- $l[-1] =~ s/^(.*?)(\s*$)/xxx($1);$2/;
- } else {
- # no comma for:
- # - prefix for/foreach/...
- # - already a comma
- # - a block { ... }
- my $no_comma = $l[-1] =~ /(^\s*(for|foreach|if|unless|while|sub)\s)|(;\s+$)|(^{.*}\s*$)/;
- my $opt_comma = $no_comma ? '' : ';';
- $l[-1] =~ s/(\s+$)/$opt_comma nop();$1/;
- }
- }
- if (! any { /^(sub|use) / } @l) {
- @l = ($local, @l, $local_trailer);
- }
- if (any { /->\w/ } @l) {
- @l = ($oo_header, $header, @l);
- } else {
- @l = ($header, @l);
- }
- output('.pl', @l);
- my @raw_log = `../perl_checker .pl`;
- die "@raw_log in .pl ($file):\n" . join('', @{$test->{lines}}) if any { /^syntax error$/ } @raw_log;
-
- my $f;
- my @log = grep {
- if (/^File "(.*)", line /) {
- $f = $1;
- 0;
- } else {
- $f eq '.pl';
- }
- } @raw_log;
-
- foreach my $i (0 .. max(int @{$test->{lines}}, int @log) - 1) {
- my $s = $test->{lines}[$i];
- $s =~ s/\s+$//;
- $new .= sprintf "%-40s %s", $s, $log[$i] || "\n";
- }
- $new .= "\n";
-}
-output("$file.new", $new);
-if (system('diff', '-buB', $file, "$file.new") == 0) {
- unlink "$file.new", '.pl', 'pkg3.pm';
- exit 0;
-} else {
- warn "*" x 80, "\nnot same\n";
- exit 1;
-}
diff --git a/perl_checker.src/test/various_errors.t b/perl_checker.src/test/various_errors.t
deleted file mode 100644
index 48a8ece..0000000
--- a/perl_checker.src/test/various_errors.t
+++ /dev/null
@@ -1,61 +0,0 @@
-local $xxx ||= $yyy applying ||= on a new initialized variable is wrong
-
-$1 =~ s/xxx/yyy/ do not modify the result of a match (eg: $1)
-
-$xxx[1, 2] you must give only one argument
-
-$xxx[] you must give one argument
-
-my $_x = 'xxx' if $xxx; replace "my $foo = ... if <cond>" with "my $foo = <cond> && ..."
-
-$xxx or my $_x = 'xxx'; replace "<cond> or my $foo = ..." with "my $foo = !<cond> && ..."
-
-'' || 'xxx' <constant> || ... is the same as ...
-
-if ($xxx = '') {} are you sure you did not mean "==" instead of "="?
-
-N("xxx$yyy") don't use interpolated translated string, use %s or %d instead
-
-if ($xxx && $yyy = xxx()) {} invalid lvalue
-
-1 + 2 >> 3 missing parentheses (needed for clarity)
-
-$xxx ? $yyy = 1 : $zzz = 2; missing parentheses (needed for clarity)
- invalid lvalue
-
-N_("xxx") . 'yyy' N_("xxx") . "yyy" is dumb since the string "xxx" will never get translated
-
-join(@l) first argument of join() must be a scalar
-
-join(',', 'foo') join('...', $foo) is the same as $foo
-
-if_($xxx) not enough parameters
-
-push @l you must give some arguments to push
-
-push $xxx, 1 push is expecting an array
-
-pop $xxx pop is expecting an array and nothing else
-
-my (@l2, $xxx) = @l; @l2 takes all the arguments, $xxx is undef in any case
-
-$bad undeclared variable $bad
-
-{ my $a } unused variable $a
-
-my $xxx; yyy($xxx); my $xxx; redeclared variable $xxx
-
-{ my $xxx; $xxx = 1 } variable $xxx assigned, but not read
-
-$a undeclared variable $a
-
-use bad; can't find package bad
-
-use pkg3 ':bad'; package pkg3 doesn't export tag :bad
-bad(); unknown function bad
-
-use pkg3 ':missing_fs'; name &f is not defined in package pkg3
-f(); name &f0 is not defined in package pkg3
-
-use pkg3 'f'; name &f is not defined in package pkg3
-f();
diff --git a/perl_checker.src/tree.ml b/perl_checker.src/tree.ml
deleted file mode 100644
index 16fd0e4..0000000
--- a/perl_checker.src/tree.ml
+++ /dev/null
@@ -1,443 +0,0 @@
-open Types
-open Common
-open Printf
-open Config_file
-open Parser_helper
-
-type special_export = Re_export_all | Fake_export_all
-
-type exports = {
- export_ok : (context * string) list ;
- export_auto : (context * string) list ;
- export_tags : (string * (context * string) list) list ;
- special_export : special_export option ;
- }
-
-type uses = (string * ((context * string) list option * pos)) list
-
-type prototype = {
- proto_nb_min : int ;
- proto_nb_max : int option ;
- }
-
-type variable_used = Access_none | Access_write_only | Access_various
-
-type per_package = {
- package_name : string ; has_package_name : bool ;
- vars_declared : (context * string, pos * variable_used ref * prototype option) Hashtbl.t;
- imported : ((context * string) * (string * variable_used ref * prototype option)) list option ref;
- exports : exports ;
- uses : uses ;
- required_packages : (string * pos) list ;
- body : fromparser list;
- isa : (string * pos) list option ;
- }
-
-type per_file = {
- file_name : string ;
- require_name : string option ;
- lines_starts : int list ;
- build_time : float ;
- packages : per_package list ;
- from_basedir : bool ;
- }
-
-let anonymous_package_count = ref 0
-let empty_exports = { export_ok = []; export_auto = []; export_tags = []; special_export = None }
-let use_lib = ref (List.map Info.file_to_absolute_file (readlines (Unix.open_process_in "perl -le 'print foreach @INC'")))
-
-let ignore_package pkg =
- if !Flags.verbose then print_endline_flush ("ignoring package " ^ pkg);
- lpush ignored_packages pkg
-
-let die_with_pos pos msg = failwith (Info.pos2sfull pos ^ msg)
-let warn_with_pos warn_types pos msg = if Flags.are_warning_types_set warn_types then print_endline_flush (Info.pos2sfull pos ^ msg)
-
-let s2context s =
- match s.[0] with
- | '$' -> I_scalar, skip_n_char 1 s
- | '%' -> I_hash , skip_n_char 1 s
- | '@' -> I_array , skip_n_char 1 s
- | '&' -> I_func , skip_n_char 1 s
- | '*' -> I_star , skip_n_char 1 s
- | _ -> I_raw, s
-
-
-let get_current_package t =
- match t with
- | Package(Ident _ as ident) :: body ->
- let rec bundled_packages packages current_package found_body = function
- | [] -> List.rev ((Some current_package, List.rev found_body) :: packages)
- | Package(Ident _ as ident) :: body ->
- let packages = (Some current_package, List.rev found_body) :: packages in
- bundled_packages packages (string_of_fromparser ident) [] body
- | instr :: body ->
- bundled_packages packages current_package (instr :: found_body) body
- in
- bundled_packages [] (string_of_fromparser ident) [] body
- | _ ->
- if str_ends_with !Info.current_file ".pm" then warn_with_pos [Warn_normalized_expressions] (!Info.current_file, 0, 0) (sprintf "module %s does not have \"package xxxx;\" on its first line" (Info.absolute_file_to_file !Info.current_file)) ;
- [ None, t ]
-
-let from_qw_raw = function
- | String([s, List []], pos) -> [ s, pos ]
- | String(_, pos) ->
- warn_with_pos [] pos "not recognised yet" ;
- []
- | Raw_string(s, pos) ->
- [ s, pos ]
- | List [] -> []
- | List [ List l ] ->
- some_or (l_option2option_l (List.map (function
- | String([s, List []], pos)
- | Raw_string(s, pos) -> Some(s, pos)
- | Ident(_, _, pos) as ident -> Some(string_of_fromparser ident, pos)
- | e -> warn_with_pos [] (get_pos_from_expr e) "not recognised yet"; None
- ) l)) []
- | e -> warn_with_pos [] (get_pos_from_expr e) "not recognised yet"; []
-
-let from_qw e =
- List.map (fun (s, pos) ->
- let context, s' = s2context s in
- let context =
- match context with
- | I_raw -> if s'.[0] = ':' then I_raw else I_func
- | I_func -> warn_with_pos [Warn_import_export] pos "weird, exported name with a function context especially given"; I_func
- | _ -> context
- in context, s'
- ) (from_qw_raw e)
-
-let get_exported t =
- List.fold_left (fun exports e ->
- match e with
- | List [ Call_op("=", [ Deref(I_array, Ident(None, "EXPORT", _)); Call _ ], pos) ]
- | List [ Call_op("=", [ My_our("our", [(I_array, "EXPORT")], _); Call _ ], pos) ] ->
- if exports.special_export = None then warn_with_pos [Warn_import_export] pos "unrecognised @EXPORT" ;
- exports
-
- | List [ Call_op("=", [ Deref(I_array, Ident(None, "EXPORT", _)); v ], pos)]
- | List [ Call_op("=", [ My_our("our", [(I_array, "EXPORT")], _); v ], pos)] ->
- if exports.export_auto <> [] then warn_with_pos [Warn_import_export] pos "weird, @EXPORT set twice" ;
- { exports with export_auto = from_qw v }
-
- | Perl_checker_comment("RE-EXPORT-ALL", _) -> { exports with special_export = Some Re_export_all }
- | Perl_checker_comment("EXPORT-ALL", _) -> { exports with special_export = Some Fake_export_all }
-
- | List [ Call_op("=", [ Deref(I_array, Ident(None, "EXPORT_OK", _)); v ], pos)]
- | List [ Call_op("=", [ My_our("our", [(I_array, "EXPORT_OK")], _); v ], pos)] ->
- if exports.export_ok <> [] then warn_with_pos [Warn_import_export] pos "weird, @EXPORT_OK set twice" ;
- (match v with
- | Call(Deref(I_func, Ident(None, "map", _)),
- [ Anonymous_sub(_, Block [List [Deref(I_array, Deref(I_scalar, Ident (None, "_", _)))]], _);
- Call(Deref(I_func, Ident(None, "values", _)), [ Deref(I_hash, Ident(None, "EXPORT_TAGS", _))])]) ->
- { exports with export_ok = collect snd exports.export_tags }
- | _ -> { exports with export_ok = from_qw v })
-
- | List [ Call_op("=", [ Deref(I_hash, Ident(None, "EXPORT_TAGS", _)); v ], pos)]
- | List [ Call_op("=", [ My_our("our", [(I_hash, "EXPORT_TAGS")], _); v ], pos)] ->
- (try
- let export_tags =
- match v with
- | List [ List l ] ->
- List.map (function
- | Raw_string(tag, _), Ref(I_array, List [List [v]]) ->
- let para =
- match v with
- | Deref(I_array, Ident(None, "EXPORT_OK", _)) -> exports.export_ok
- | _ -> from_qw v
- in
- ":" ^ tag, para
- | _ -> raise Not_found
- ) (group_by_2 l)
- | _ -> raise Not_found
- in
- if exports.export_tags <> [] then warn_with_pos [Warn_import_export] pos "weird, %EXPORT_TAGS set twice" ;
- { exports with export_tags = export_tags }
- with _ ->
- warn_with_pos [Warn_import_export] pos "unrecognised %EXPORT_TAGS" ;
- exports)
-
- (* $EXPORT_TAGS{all} = [ map { @$_ } values %EXPORT_TAGS ]; *)
- | List [Call_op("=", [
- Deref_with(I_hash, I_scalar, Ident(None, "EXPORT_TAGS", _), Raw_string("all", _));
- Ref(I_array,
- List[List[
- Call(Deref(I_func, Ident(None, "map", _)),
- [Anonymous_sub(_, Block [List [Deref(I_array, Deref(I_scalar, Ident(None, "_", _)))]], _);
- Call(Deref(I_func, Ident(None, "values", _)), [Deref(I_hash, Ident(None, "EXPORT_TAGS", _))])])
- ]])
- ], _)] ->
- { exports with export_tags = (":all", collect snd exports.export_tags) :: exports.export_tags }
-
- | List (My_our _ :: _) ->
- let _,_ = e,e in
- exports
- | _ -> exports
- ) empty_exports t
-
-let uses_external_package = function
- | "vars" | "Exporter" | "diagnostics" | "strict" | "warnings" | "lib" | "POSIX" | "Gtk" | "Storable"
- | "Config" | "Socket" | "IO::Socket" | "DynaLoader" | "Data::Dumper" | "Time::localtime" | "Expect" -> true
- | _ -> false
-
-let get_uses t =
- List.fold_left (fun uses e ->
- match e with
- | Use(Ident(None, "lib", _), [libs]) ->
- use_lib := List.map Info.file_to_absolute_file (List.map snd (from_qw libs)) @ !use_lib ;
- uses
- | Use(Ident(None, "base", _), classes) ->
- let l = List.map (fun (pkg, pos) -> (pkg, (None, pos))) (collect from_qw_raw classes) in
- l @ uses
- | Use(Ident(_, _, pos) as pkg, l) ->
- let package = string_of_fromparser pkg in
- if uses_external_package package then
- uses
- else
- let para = match l with
- | [] -> None
- | [ Num(_, _) ] -> None (* don't care about the version number *)
- | _ -> Some(collect from_qw l)
- in
- (package, (para, pos)) :: uses
- | _ -> uses
- ) [] t
-
-let get_isa t =
- List.fold_left (fun (isa, exporter) e ->
- match e with
- | Use(Ident(None, "base", pos), classes) ->
- if isa <> None || exporter <> None then die_with_pos pos "\"use base\" and \"@ISA\" must be used once only";
- Some (collect from_qw_raw classes), None
- | List [ Call_op("=", [ Deref(I_array, Ident(None, "ISA", pos)) ; classes ], _) ]
- | List [ Call_op("=", [ My_our("our", [ I_array, "ISA" ], pos) ; classes ], _) ] ->
- if isa <> None || exporter <> None then die_with_pos pos "\"use base\" and \"@ISA\" must be used once only";
- let special, l = List.partition (fun (s, _) -> s = "DynaLoader" || s = "Exporter") (from_qw_raw classes) in
- let exporter = if List.mem_assoc "Exporter" special then Some pos else None in
- let isa = if l = [] && special <> [] then None else Some l in
- isa, exporter
- | _ -> isa, exporter
- ) (None, None) t
-
-let read_xs_extension_from_c global_vars_declared file_name package pos =
- try
- let cfile = Filename.chop_extension file_name ^ ".c" in
- let prefix = "newXS(\"" ^ package.package_name ^ "::" in
- ignore (fold_lines (fun in_bootstrap s ->
- if in_bootstrap then
- (try
- let offset = strstr s prefix + String.length prefix in
- let end_ = String.index_from s offset '"' in
- let ident = String.sub s offset (end_ - offset) in
- match split_name_or_fq_name ident with
- | None, ident -> Hashtbl.replace package.vars_declared (I_func, ident) (pos, ref Access_none, None)
- | Some fq, ident ->
- let fq = package.package_name ^ "::" ^ fq in
- Hashtbl.replace global_vars_declared (I_func, fq, ident) (pos, None)
- with Not_found -> ());
- in_bootstrap || str_contains s "XS_VERSION_BOOTCHECK"
- ) false (open_in cfile));
- if !Flags.verbose then print_endline_flush (sprintf "using xs symbols from %s" cfile) ;
- true
- with Invalid_argument _ | Sys_error _ -> false
-
-let findfile dirs f = List.find (fun dir -> Sys.file_exists (dir ^ "/" ^ f)) dirs
-
-let read_xs_extension_from_so global_vars_declared package pos =
- try
- let splitted = split_at2 ':'':' package.package_name in
- let rel_file = String.concat "/" ("auto" :: splitted @ [ last splitted ]) ^ ".so" in
- let so = (findfile !use_lib rel_file) ^ "/" ^ rel_file in
- let channel = Unix.open_process_in (Printf.sprintf "nm --defined-only -D \"%s\"" so) in
- if !Flags.verbose then print_endline_flush (sprintf "using shared-object symbols from %s" so) ;
- fold_lines (fun () s ->
- let s = skip_n_char 11 s in
- if str_begins_with "XS_" s then
- let s = skip_n_char 3 s in
- let len = String.length s in
- let rec find_package_name accu i =
- try
- let i' = String.index_from s i '_' in
- let accu = String.sub s i (i'-i) :: accu in
- if i' + 1 < len && s.[i'+1] = '_' then
- find_package_name accu (i' + 2)
- else
- List.rev accu, skip_n_char (i'+1) s
- with Not_found -> List.rev accu, skip_n_char i s
- in
- let fq, name = find_package_name [] 0 in
- Hashtbl.replace global_vars_declared (I_func, String.concat "::" fq, name) (pos, None)
- ) () channel;
- if not Build.debugging then ignore (Unix.close_process_in channel) ;
- true
- with Not_found -> false
-
-let has_proto perl_proto body =
- match perl_proto with
- | Some "" -> Some([], raw_pos2pos bpos, [body])
- | _ ->
- match body with
- | Block [] ->
- Some([ I_array, "_empty" ], raw_pos2pos bpos, [])
- | Block (List [Call_op ("=", [My_our ("my", mys, mys_pos); Deref(I_array, Ident(None, "_", _))], _pos)] :: body) ->
- Some(mys, mys_pos, body)
- | _ -> None
-
-let get_proto perl_proto body =
- map_option (fun (mys, pos, _) ->
- let scalars, others = break_at (fun (context, _) -> context <> I_scalar) mys in
- (match others with
- | (I_array, _) :: _ :: _ -> warn_with_pos [Warn_prototypes] pos "an array must be the last variable in a prototype"
- | (I_hash, _) :: _ :: _ -> warn_with_pos [Warn_prototypes] pos "an hash must be the last variable in a prototype"
- | _ -> ());
- let is_optional (_, s) =
- String.length s > 2 && (s.[0] = 'o' || s.[0] = 'b') && s.[1] = '_' ||
- String.length s > 3 && s.[0] = '_' && (s.[1] = 'o' || s.[1] = 'b') && s.[2] = '_'
- in
- let must_have, optional = break_at is_optional scalars in
- if not (List.for_all is_optional optional) then
- warn_with_pos [Warn_prototypes] pos "an non-optional argument must not follow an optional argument";
- let min = List.length must_have in
- { proto_nb_min = min; proto_nb_max = if others = [] then Some(min + List.length optional) else None }
- ) (has_proto perl_proto body)
-
-let get_vars_declaration global_vars_declared file_name package =
- List.iter (function
- | Sub_declaration(Ident(None, name, pos), perl_proto, body, _) ->
- Hashtbl.replace package.vars_declared (I_func, name) (pos, ref Access_none, get_proto perl_proto body)
- | Sub_declaration(Ident(Some fq, name, pos), perl_proto, body, _) ->
- Hashtbl.replace global_vars_declared (I_func, fq, name) (pos, get_proto perl_proto body)
-
- | List [ Call_op("=", [My_our("our", ours, pos); _], _) ]
- | List [ Call_op("=", [My_our("local", ([ I_scalar, "_" ] as ours), pos); _], _) ]
- | List [ My_our("our", ours, pos) ]
- | My_our("our", ours, pos) ->
- List.iter (fun (context, name) -> Hashtbl.replace package.vars_declared (context, name) (pos, ref Access_none, None)) ours
-
- | Use(Ident(None, "vars", pos), [ours]) ->
- List.iter (fun (context, name) -> Hashtbl.replace package.vars_declared (context, name) (pos, ref Access_none, None)) (from_qw ours)
- | Use(Ident(None, "vars", pos), _) ->
- die_with_pos pos "usage: use vars qw($var func)"
-
- | List [ Method_call(Raw_string(pkg, pos), Raw_string("bootstrap", _), _) ] ->
- if pkg <> package.package_name then
- warn_with_pos [Warn_import_export] pos "strange bootstrap (the package name is not the same as the current package)"
- else
- if not (read_xs_extension_from_c global_vars_declared file_name package pos) then
- if not (read_xs_extension_from_so global_vars_declared package pos) then
- ignore_package pkg
- | _ -> ()
- ) package.body
-
-let rec fold_tree f env e =
- match f env e with
- | Some env -> env
- | None ->
- match e with
- | Anonymous_sub(_, e', _)
- | Ref(_, e')
- | Deref(_, e')
- -> fold_tree f env e'
-
- | Diamond(e')
- -> fold_tree_option f env e'
-
- | String(l, _)
- -> List.fold_left (fun env (_, e) -> fold_tree f env e) env l
-
- | Sub_declaration(e1, _, e2, _)
- | Deref_with(_, _, e1, e2)
- ->
- let env = fold_tree f env e1 in
- let env = fold_tree f env e2 in
- env
-
- | Use(_, l)
- | List l
- | Block l
- | Call_op(_, l, _)
- -> List.fold_left (fold_tree f) env l
-
- | Call(e', l)
- ->
- let env = fold_tree f env e' in
- List.fold_left (fold_tree f) env l
-
- | Method_call(e1, e2, l)
- ->
- let env = fold_tree f env e1 in
- let env = fold_tree f env e2 in
- List.fold_left (fold_tree f) env l
-
- | _ -> env
-
-and fold_tree_option f env = function
- | None -> env
- | Some e -> fold_tree f env e
-
-
-let get_global_info_from_package from_basedir require_name build_time t =
- let current_packages = get_current_package t in
- let packages = List.map (fun (current_package, t) ->
- let exports = get_exported t in
- let exporting_something() = exports.export_ok <> [] || exports.export_auto <> [] || exports.export_tags <> [] || exports.special_export = Some Re_export_all in
-
- let package_name =
- match current_package with
- | None ->
- if exporting_something() then
- die_with_pos (!Info.current_file, 0, 0) "file with no package name wants to export!"
- else
- (incr anonymous_package_count ; sprintf "anonymous%d" !anonymous_package_count)
- | Some name -> name
- in
- let isa, exporter = get_isa t in
- (match exporter with
- | None ->
- if exporting_something() then warn_with_pos [Warn_import_export] (!Info.current_file, 0, 0) "you must have \"@ISA = qw(Exporter)\" to EXPORT something"
- | Some pos ->
- if not (exporting_something()) then warn_with_pos [Warn_import_export] pos "Inheritating from Exporter without EXPORTing anything");
-
- let uses = List.rev (get_uses t) in
- let required_packages = List.map (fun (s, (_, pos)) -> s, pos) uses in
- let required_packages = List.fold_left (fold_tree (fun l ->
- function
- | Perl_checker_comment(s, pos) when str_begins_with "require " s ->
- Some((skip_n_char 8 s, pos) :: l)
- | Call(Deref(I_func, Ident (None, "require", pos)), [Ident _ as pkg]) ->
- let package = string_of_fromparser pkg in
- if uses_external_package package then None else Some((package, pos) :: l)
- | Call(Deref(I_func, Ident (None, "require", pos)), [Raw_string(pkg, _)])
- when not (String.contains pkg '/') && Filename.check_suffix pkg ".pm" ->
- let package = Filename.chop_suffix pkg ".pm" in
- if uses_external_package package then None else Some((package, pos) :: l)
- | _ -> None)
- ) required_packages t in
- {
- package_name = package_name;
- has_package_name = current_package <> None ;
- exports = exports ;
- imported = ref None ;
- vars_declared = Hashtbl.create 16 ;
- uses = uses ;
- required_packages = required_packages ;
- body = t ;
- isa = isa ;
- }
- ) current_packages in
-
- let require_name = match require_name with
- | Some require_name -> Some require_name
- | None -> match packages with
- | [ pkg ] when pkg.has_package_name -> Some pkg.package_name
- | _ -> None
- in
- {
- file_name = !Info.current_file ;
- require_name = require_name ;
- lines_starts = !Info.current_file_lines_starts ;
- build_time = build_time ;
- packages = packages ;
- from_basedir = from_basedir ;
- }
-
diff --git a/perl_checker.src/tree.mli b/perl_checker.src/tree.mli
deleted file mode 100644
index 3cdf219..0000000
--- a/perl_checker.src/tree.mli
+++ /dev/null
@@ -1,57 +0,0 @@
-open Types
-
-type special_export = Re_export_all | Fake_export_all
-
-type exports = {
- export_ok : (context * string) list;
- export_auto : (context * string) list;
- export_tags : (string * (context * string) list) list;
- special_export : special_export option;
-}
-
-
-type uses = (string * ((context * string) list option * pos)) list
-
-type prototype = {
- proto_nb_min : int ;
- proto_nb_max : int option ;
- }
-
-type variable_used = Access_none | Access_write_only | Access_various
-
-type per_package = {
- package_name : string ; has_package_name : bool ;
- vars_declared : (context * string, pos * variable_used ref * prototype option) Hashtbl.t;
- imported : ((context * string) * (string * variable_used ref * prototype option)) list option ref;
- exports : exports ;
- uses : uses ;
- required_packages : (string * pos) list ;
- body : fromparser list;
- isa : (string * pos) list option ;
- }
-
-type per_file = {
- file_name : string ;
- require_name : string option ;
- lines_starts : int list ;
- build_time : float ;
- packages : per_package list ;
- from_basedir : bool ;
- }
-
-val empty_exports : exports
-val ignore_package : string -> unit
-val use_lib : string list ref
-val uses_external_package : string -> bool
-val findfile : string list -> string -> string
-
-val get_global_info_from_package : bool -> string option -> float -> fromparser list -> per_file
-
-val has_proto : string option -> fromparser -> ((context * string) list * pos * fromparser list) option
-val get_vars_declaration : (context * string * string, pos * prototype option) Hashtbl.t -> string -> per_package -> unit
-
-val die_with_pos : string * int * int -> string -> 'a
-val warn_with_pos : Types.warning list -> string * int * int -> string -> unit
-
-val fold_tree : ('a -> fromparser -> 'a option) -> 'a -> fromparser -> 'a
-val from_qw : fromparser -> (context * string) list
diff --git a/perl_checker.src/types.mli b/perl_checker.src/types.mli
deleted file mode 100644
index 5f23d3a..0000000
--- a/perl_checker.src/types.mli
+++ /dev/null
@@ -1,125 +0,0 @@
-exception TooMuchRParen
-
-type raw_pos = int * int
-
-type pos = string * int * int
-
-type spaces =
- | Space_0
- | Space_1
- | Space_n
- | Space_cr
- | Space_none
-
-type context = I_scalar | I_hash | I_array | I_func | I_raw | I_star
-
-type maybe_context =
- | M_none
-
- (* scalars *)
- | M_bool | M_int | M_float
- | M_revision
- | M_string
- | M_ref of maybe_context
- | M_undef
- | M_unknown_scalar
-
- | M_tuple of maybe_context list
- | M_list
- | M_array
- | M_hash
- | M_sub
-
- | M_special
- | M_unknown
- | M_mixed of maybe_context list
-
-type sub_declaration_kind = Real_sub_declaration | Glob_assign
-
-type fromparser =
- | Undef
- | Ident of string option * string * pos
- | Num of string * pos
- | Raw_string of string * pos
- | String of (string * fromparser) list * pos
-
- | Ref of context * fromparser
- | Deref of context * fromparser
- | Deref_with of context * context * fromparser * fromparser (* from_context, to_context, ref, para *)
-
- | Diamond of fromparser option
-
- | List of fromparser list
- | Block of fromparser list
-
- | Call_op of string * fromparser list * pos
- | Call of fromparser * fromparser list
- | Method_call of fromparser * fromparser * fromparser list
-
- | Anonymous_sub of string option * fromparser * pos (* prototype, expr, pos *)
- | My_our of string * (context * string) list * pos
- | Use of fromparser * fromparser list
- | Sub_declaration of fromparser * string option * fromparser * sub_declaration_kind (* name, prototype, body, kind *)
- | Package of fromparser
- | Label of string
- | Perl_checker_comment of string * pos
-
- | Too_complex
- | Semi_colon
-
-type priority =
-| P_tok
-| P_tight
-| P_mul
-| P_add
-| P_uniop
-| P_cmp
-| P_eq
-| P_expr
-| P_bit
-| P_tight_and
-| P_tight_or
-| P_ternary
-| P_assign
-| P_comma
-| P_call_no_paren
-| P_and
-| P_or
-| P_loose
-
-| P_paren_wanted of priority
-| P_paren of priority
-
-| P_none
-
-type 'a any_spaces_pos = {
- any : 'a ;
- spaces : spaces ;
- pos : int * int ;
- mcontext : maybe_context ;
- }
-
-type 'a prio_anyexpr = {
- priority : priority ;
- expr : 'a
- }
-
-type prio_expr_spaces_pos = fromparser prio_anyexpr any_spaces_pos
-type prio_lexpr_spaces_pos = fromparser list prio_anyexpr any_spaces_pos
-
-type warning =
- | Warn_white_space
- | Warn_suggest_simpler
- | Warn_unused_global_vars
- | Warn_void
- | Warn_context
- | Warn_strange
- | Warn_traps
- | Warn_complex_expressions
- | Warn_normalized_expressions
- | Warn_suggest_functional
- | Warn_prototypes
- | Warn_import_export
- | Warn_names
- | Warn_MDK_Common
- | Warn_help_perl_checker