[PATCH] subrepo: avoids empty commit when .hgsubstate is dirty (issue2403)

[Erik Zielke]

fre, 17 12 2010 kl. 12:18 +0100, skrev Mads Kiilerich: 
> On 12/17/2010 11:53 AM, Erik Zielke wrote:
> > fre, 17 12 2010 kl. 11:18 +0100, skrev Mads Kiilerich:
> >> It seems to me like a proper fix requires some refactoring of the commit
> >> function. We basically don't know the state of the outer repo before we
> >> have done the recursive commit, so we have to do the recursive commit
> >> before we can create the commit context (which we however use for the
> >> recursive commit).
> >>
> >
> > So you think that it should just be backed out for now and then prevent
> > such empty commits somewhat differently, right?
> 
> I worked enough with this issue to know what didn't work for me, but I 
> gave up because I couldn't see any good solutions.
> 

Okay.

> But it seems like the annoyance fix introduced a more serious bug, so I 
> think it either should be fixed or backed out.
> 

Yes, I am sorry.

> > Although something like the following will fix the issue you pointed
> > out:
> >
> >                       for newstate in state:
> >                           if state[newstate][1] != pstate[newstate]:
> >                               changed = True
> > -                if changed:
> > -                    subrepo.writestate(self, state)
> > -                elif (changes[0] == ['.hgsubstate'] and changes[1] ==
> > [] and
> > +                subrepo.writestate(self, state)
> > +                if (not changed and changes[0] == ['.hgsubstate']
> > +                    and changes[1] == [] and
> >                        changes[2] == []):
> >                       return None
> 
> That would bring us back to still making empty substate commits when 
> there are other changes, right?
> 

When there are other changes they would go into the commit and therefore
not empty, or am I misunderstanding something?

> I guess a good fix in this direction should remove .hgsubstate from 
> changes[0] if it is unchanged, and _then_ we can try to abort the commit 
> if there is nothing left to commit ... unless we are in one of the 
> situations where we really want to make an empty commit.
> 

Okay.

> /Mads