Icon caching is unusable [MY TAKE]

[Loïc Minier]

        Hi,

 First some context.  This discussion is quite old and started in
 february on IRC with various Debian people when we scratched our butt
 and wondered what on hell was wrong with icon caching.  We didn't
 notice first, and then we suffered, eg. <http://bugs.debian.org/296029>
 where a random application borke.
   We decided there was something wrong with the implementation, or even
 with the spec, and we opened a discussion with upstream at:
    <http://bugzilla.gnome.org/show_bug.cgi?id=167941>

 We're now in october and, as I understand it, Fedora/Redhat went on
 with caching and changed the load of packages using icons.  Debian
 looks damn stupid because you can point at Redhat and say something in
 the lines of "they did it, just do like they did".  Terribly true.

 The icon caching code is now cold and accepted, I personally doubt
 anyone would bother changing 1/ the spec, 2/ the implementation in gtk
 and 3/ fedora today.


 *But*, please look at the volume of packages that one needs to update
 because of they the implementation is done.  Please imagine what it
 takes to convince numerous upstream packages, numerous package
 maintainers, and in your case numerous distros to change the way they
 install files / packages: it's an awful lot of work, at all levels.

 As far as I can't tell, neither the rpm not the dpkg world have a big
 hook to catch what happens below /usr/share/icons and fix things when
 the installation is over.  Nor do they use a single
 "freedesktop-postinst" script cleaning things up, and calling the
 relevant utilities.


 So please, think of transition periods for future specifications,
 something permitting applications to continue working, while the
 updated applications benefit of the new features.  On the top of my
 head, the ugly solution of a /usr/share/icons/cache-able would have
 offered that.


 You might not be sensible to the amount of work it takes for
 applications in a distro, but you might be more sensible to the legacy
 support of older applications, not cache-aware.

 People are usually careful of not breaking the ABI compatibility at the
 ELF level, I suppose that the compatibility at the system level
 interest them too.


   Cheers,

-- 
Loïc Minier <lool at dooz.org>