[Mageia-dev] Mirror layout, round two

Mon Nov 29 09:06:25 CET 2010

Olivier Thauvin skrev 29.11.2010 03:06:
+> * Thomas Backlund (tmb at iki.fi) wrote:
+>>
+>> So the mirror medias accordingly to all comments so far would be a simple:
+>>
+>> * core
+>>    - enabled by default
+>>    - mirrors must mirror this media to be listed as a mirror
+>>    - only GPL stuff
+>>    - must be selfcontained
+>>
+>> * nonfree
+>>    - disabled by default, installer will ask to enable it if
+>>      it detects hw that need driver/fw from here...
+>>    - mirrors must mirror this media to be listed as a mirror
+>>    - contains apps/drivers/firmware that are free to redistribute,
+>>      but we dont have GPL source for
+>>    - for example ati/nvidia drivers/firmware, Oracle Java, ...
+>>
+>> * tainted
+>>    - disabled by default
+>>    - mirrors are free to not mirror this media
+>>    - stuff we think we can redistribute, but that may have some
+>>      patent issues or other restrictions in oter countries.
+>>
+>
+> I can't agree with the "mirrors are free to not mirror this media",
+> three reasons:
+> 1) I don't see an easy and safe way for mirrors to exclude a media (a
+> directory + hdlist in media/media_info) in each distribution,
+
+Thats easy, just use --exclude-from
+
+On my local mirror I exclude the debug stuff
+so I use --exclude-from=excclude.lst
+
+# cat exclude.lst
+i586/media/*debug*
+i586/media/media_info/*debug*
+x86_64/media/*debug*
+x86_64/media/media_info/*debug*
+
+And thats all needed.
+We can even provide this exclude file ourselves on the mirrors so we 
+know its correct and if we change the mirror layout, we update the file.
+
+> 2) I don't know how urpmi can manage this, all medias are described in
+> only one file, to detect something is missing
+> (media/media_info/media.cfg).
+
+It already works.
+For example if you only mirror x86_64, the media.cfg has references to 
+Main32 and Main32 Updates. but since the hdlists are symlinked to
+../../../i586/media/... the hdlists and synthesis will be empty,
+so no reference to any 32bit package gets into urpmi database.
+
+> 3) I tried to act as rules that a valid is a mirror containing
+> everything, to not have to deal with a lot of mirrors style.
+>
+
+Well,
+if te choice is between no mirror and a mirror without *tainted*
+I'd choose the mirror without *tainted*
+
+--
+Thomas
+