[arch-general] [arch-dev-public] Strange behaviour of pacman

[Pierre Chapuis]

Le Thu, 22 Oct 2009 14:02:53 -0500,
Aaron Griffin <aaronmgriffin at gmail.com> a écrit :

> On Thu, Oct 22, 2009 at 1:45 PM, Pierre Chapuis <catwell at archlinux.us> wrote:
> > On Tue, 20 Oct 2009 09:48:58 +1000,
> > Allan McRae <allan at archlinux.org> wrote:
> >
> >> Jan de Groot wrote:
> >> > On Mon, 2009-10-19 at 15:18 -0500, Aaron Griffin wrote:
> >> >
> >> >> Are you saying that the .pyo files are no longer architecture
> >> >> independent? I was under the assumption they were.
> >> >>
> >> >
> >> > Actually, they're even python-version specific. Updating python could
> >> > break the precompiled .pyo files.
> >> >
> >>
> >> And this whole issue was a fairly major source of headaches during the
> >> python-2.6 transition...  which is why I started making the python
> >> packaging policy to deal with them, although that obviously was never
> >> finished with  (in fact, I had never seen the comment with --optimize=1
> >> in it).
> >>
> >> Now my main concern about all of this is that .pyc and .pyo files used
> >> to contain full paths to where they were created.  That meant they need
> >> to be created on the users system and not during the packaging stage.
> >> I have not confirmed if this is still the case.
> >>
> >> So the best way to deal with them seems to be:
> >> 1) touch them during packaging
> >> 2) generate them during post_install()
> >
> > I have found a way to automate that which is, I believe, not PKGBUILD-dependant.
> >
> > Here's what I do in the PKGBUILD:
> >
> > [...]
> > install="pyo_remover.install"
> > [...]
> > build() {
> >  [...]
> >  # Take care of .pyo files
> >  cd $pkgdir
> >  echo "post_install() {" > $startdir/$install
> >  for _i in  $(find . -name '*.pyo'); do
> >    echo "rm -f "$(echo "$_i" | cut -c2-) >> $startdir/$install
> >    echo > "$_i"
> >  done
> >  echo -e '}\npost_upgrade() {\npost_install $1\n}\n' >>  $startdir/$install
> > }
> >
> > pyo_remover.install can be anything, even an empty file. For packages that need a .install file this has to be adapted.
> >
> > Does this look like a good way to solve the problem?  I know the way I do it for now is kind of ugly, but I think it could be much cleaner if the same kind of thing was done directly by makepkg.
> 
> 
> Did you mean for this to be post_install? This should be done on
> remove, if I'm not mistaken, as the pyo files are actually a good
> thing

No, I meant that to be post_install, because that way:

 - Pacman will track the .pyo files because they are in the package (as empty text files), so they will be deleted on removal.

 - The .pyo files will be deleted after install so I think they will be re-generated at runtime. This means there will always be generated for the right architecture and Python version.

By the way I think you can do the same for .pyc files.

-- 
catwell