[Mageia-dev] Library policy query: What do we do when SONAME includes both major and minor? (Re: [changelog] [RPM] cauldron core/release kdelibs4-4.9.0-2.mga3)

Thu Aug 2 11:31:04 CEST 2012

On Thu, Aug 2, 2012 at 10:17 AM, Colin Guthrie <mageia at colin.guthr.ie> wrote:
+> 'Twas brillig, and Balcaen John at 02/08/12 10:01 did gyre and gimble:
+>> Le jeudi 2 août 2012 09:28:28 Colin Guthrie a écrit :
+>>> 'Twas brillig, and Christiaan Welvaart at 01/08/12 23:09 did gyre and
+>>>
+>>> gimble:
+>>>> On Wed, 1 Aug 2012, Colin Guthrie wrote:
+>>>>> I have to agree here that something is "funny" in the libattica package
+>>>>> which ultimately helped to contribute to this issue.
+>>>>>
+>>>>> e.g. on my system before update (tho' with similar results after):
+>>>>>
+>>>>> [colin at jimmy ~]$ rpm -q --provides lib64attica0
+>>>>> libattica.so.0.3()(64bit)
+>>>>> lib64attica0 = 0.3.0-1.mga2
+>>>>> lib64attica0(x86-64) = 0.3.0-1.mga2
+>>>>> [colin at jimmy ~]$ rpm -ql lib64attica0
+>>>>> /usr/lib64/libattica.so.0.3
+>>>>> /usr/lib64/libattica.so.0.3.0
+>>>>>
+>>>>> So I can see how this mistake was made and TBH I could have made the
+>>>>> same mistake myself (with the caveat that I likely would not have bumped
+>>>>> the version of someone else's package with out confirming first and that
+>>>>> it should have been obvious from testing and installing the build)
+>>>>>
+>>>>> But either way this seems like an issue to fix properly (possibly with
+>>>>> an upstream fix or some modification to the library policy when the
+>>>>> minor version is "presented" like this).
+>>>>
+>>>> Good catch! Of course it's never the library policy that's wrong. The
+>>>> library major version is apparently 0.4 so the correct package name is
+>>>>
+>>>>    lib64attica0.3  for the previous one
+>>>>    lib64attica0.4  for the current one
+>>>>
+>>>> ... in the specfile:   %define attica_major 0.4
+>>>>
+>>>> Can the maintainer of this package please fix this?
+>>>>
+>>>> To find the version to use, look up the 'soname' of the library. I use:
+>>>>   readelf -a /usr/lib64/libattica.so.0.4|grep SONAME
+>>>>
+>>>> =>
+>>>> ...                    Library soname: [libattica.so.0.4]
+>>>>
+>>>> What follows ".so." is the major version of the library.
+>>>
+>>> Is that really the correct definition of what a "major" version is?
+>>>
+>>> I always thought the major was just the first number.
+>>>
+>>> The library policy certainly doesn't mention "double digit majors" or
+>>> similar.
+>>>
+>>> Is this something upstream is doing deliberately or is it just an oversight?
+>> https://projects.kde.org/projects/kdesupport/attica/repository/revisions/master/entry/CMakeLists.txt
+>
+> Actually it's this file/line:
+>
+> https://projects.kde.org/projects/kdesupport/attica/repository/revisions/master/entry/lib/CMakeLists.txt#L120
+>
+> So it's seems like this was done deliberately due to a ABI breakage a
+> while ago:
+>
+> https://projects.kde.org/projects/kdesupport/attica/repository/revisions/ac2270b1f9c445fd39e48051b99d35d9b9693a34
+>
+> Now, this is an interesting point (regarding our lib policy) bumping the
+> major for an API change and the minor for an ABI change seems kinda
+> sensible to me. So how should we deal with that in our policy? Just use
+> 0.4 as the "major" value here as Christiaan suggested?
+
+A minor change is supposed to only add interfaces and be backwards
+compatible, that's why it is not in the soname.
+If there has been an ABI breakage, the major should be incremented.
+