Subrepos and diff command

[Matt Mackall]

On Fri, 2011-10-21 at 09:47 -0700, Eric ROSHAN EISNER wrote:
> On Fri, Oct 21, 2011 at 09:07, Patrick Mézard <pmezard at gmail.com> wrote:
> 
> > Le 21/10/11 17:56, Eric ROSHAN EISNER a écrit :
> > > On Fri, Oct 21, 2011 at 07:30, Patrick Mézard <pmezard at gmail.com<mailto:
> > pmezard at gmail.com>> wrote:
> > >
> > >     I have some questions after looking at "Issue3056 - hg diff/status -
> > subrepo files not listed on revision where subrepo was added". With the
> > following setup (one modified subrepo being added):
> > >
> > >      $ hg init sub
> > >      $ echo b > sub/b
> > >      $ hg -R sub add sub/b
> > >      $ hg -R sub ci -m "addb"
> > >      $ hg init repo
> > >      $ cd repo
> > >      $ hg clone ../sub sub
> > >      $ echo sub = sub > .hgsub
> > >      $ hg add .hgsub
> > >      $ echo b >> sub/b
> > >
> > >     1- Should "hg diff -S" display the changes to sub/b ?
> > >
> > >     I guess it should, it is the whole point of -S.
> > >
> > >     2- Say we commit this change as revision 0. Should "hg diff -S
> > --change 0" display the changes to sub/b ?
> > >
> > >     Yes for consistency with [1].
> > >
> > >     No because it is undefined : we know the subrepo was introduced at
> > revision 1, we do not know the use start adding it while being at revision 0
> > and committed it recursively.
> > >
> > >     3- Say we modify "sub" 3 times with as many recursive commits. Should
> > "hg diff -S -r null:3" display the changes to sub/b and which ones ?
> > >
> > >
> > > The transparent thing to do is show the file diffs as if these files were
> > directly tracked by the main repo. Since these files were unknown in rev 0
> > and known in rev 3, diff -S should show them all adding their full contents
> > from /dev/null. Before you wrote this I had been under the impression that
> > diff -S and status -S already did this.
> > >
> > > Either way the subdiffs are useless for patching purposes (as any new
> > commit made with them will conflict with .hgsubstate).
> >
> > Then how do you address [1]? Display a diff of sub/b against /dev/null
> > instead of the actual change in the subrepo?
> >
> > If you do that, you do not know what you are committing in the subrepo
> > unless you ask the subrepo yourself.
> 
> 
> Well you're doing two different things with the recursive commit, so I think
> it's reasonable to ask the two repos separately to get the two different
> answers.
> 
>   $ hg diff -S # effect on root repo
>   ---/dev/null
>   +++sub/b
>   +b
>   +b
>   $ cd sub
>   $ hg diff # effect on sub repo
>   ---sub/b
>   +++sub/b
>   b
>   +b
>   $ hg commit -m b2 # manual recursive commit
>   $ cd ..
>   $ hg diff -S
>   ---/dev/null
>   +++sub/b
>   +b
>   +b
> 
> Importantly, it is consistent that a diff -S in the root repo gives the same
> result before and after the inner commit, because they end up with the same
> contents. I agree this is a bit confusing, but it reflects the confusing
> nature of recursive commit.

I agree.

-- 
Mathematics is the supreme nostalgia of our time.