[Yum-devel] [yum-commits] Branch 'yum-3_2_X' - 2 commits - yum/__init__.py yum/rpmsack.py

[James Antill]

On Sat, 2009-05-02 at 08:15 +0200, Tim Lauridsen wrote:
> On 05/01/2009 10:44 PM, James Antill wrote:
> > New commits:
> > commit 88a89c661c3856b38f7f01c79622a2d5411c9ca0
> > Author: James Antill<james at and.org>
> > Date:   Fri May 1 16:41:43 2009 -0400
> >
> >      Do skip-broken before we run postresolve callback, etc.
> >
> > diff --git a/yum/__init__.py b/yum/__init__.py
> > index dc80da2..4ab674d 100644
> > --- a/yum/__init__.py
> > +++ b/yum/__init__.py
> > @@ -716,14 +716,7 @@ class YumBase(depsolve.Depsolve):
> >
> >           (rescode, restring) = self.resolveDeps()
> >           self._limit_installonly_pkgs()
> > -        self.plugins.run('postresolve', rescode=rescode, restring=restring)
> >
> > -        if self.tsInfo.changed:
> > -            (rescode, restring) = self.resolveDeps(rescode == 1)
> > -        if self.tsInfo.pkgSack is not None: # rm Transactions don't have pkgSack
> > -            self.tsInfo.pkgSack.dropCachedData()
> > -        self.rpmdb.dropCachedData()
> > -
> >           #  We_must_  get rid of all the used tses before we go on, so that C-c
> >           # works for downloads / mirror failover etc.
> >           self.rpmdb.ts = None
> > @@ -738,6 +731,14 @@ class YumBase(depsolve.Depsolve):
> >               self._printTransaction()
> >               self.verbose_logger.debug('Skip-Broken time: %0.3f' % (time.time() - sb_st))
> >
> > +        self.plugins.run('postresolve', rescode=rescode, restring=restring)
> > +
> > +        if self.tsInfo.changed:
> > +            (rescode, restring) = self.resolveDeps(rescode == 1)
> > +        if self.tsInfo.pkgSack is not None: # rm Transactions don't have pkgSack
> > +            self.tsInfo.pkgSack.dropCachedData()
> > +        self.rpmdb.dropCachedData()
> > +
> >           self.verbose_logger.debug('Depsolve time: %0.3f' % (time.time() - ds_st))
> >           return rescode, restring
> >    
> Just curious, why do skip-broken before postresolve, post resolve can 
> change the transaction and the idea was to
> run skip-broken, when the transaction is no longer changing.

 So the big problem here is that some plugins which run in postresolve
expect that the transaction to be read-only (ie. done) so for example if
"rescode == 1" then the transaction is going to fail, and skip-broken
breaks that assumption.
 It also seemed like a bad idea to call it "post resolve" when we still
have stuff to do to resolve.

 Doing a quick "grep postresolve_hook" in yum-utils, I see:

allowdowngrade == dead, would have been better in preresolve
changelog == wants a read-only transaction (the reason I did the change)
fedorakmod == dead, probably wants to be "last" writable
protect-packages == wants a read-only transaction
remove-with-leaves == wants to be "last" writable

...so it looks like the new behaviour is closer to expected ... I can
change it back, for compatibility, except that openssl and flash have
conflicted for about 3 months on rawhide so it sucks to never get
changelogs there.
 Another way to go would be to add more hooks to true and fine tune the
behaviour everyone "wants".

-- 
James Antill <james at fedoraproject.org>
Fedora