Re: Use of "stg sync"

From: Catalin Marinas <catalin.marinas@gmail.com>
Date: Feb 6, 2007, 10:07 UTC
Message-ID: <b0943d9e0702060207v3a6ed16fs61feb34f03211a6@mail.gmail.com>
In-Reply-To: <20070204144233.GP5362@nan92-1-81-57-214-146.fbx.proxad.net>

On 04/02/07, Yann Dirson <ydirson@altern.org> wrote: [...]

Show 8 quoted lines

> Then I wanted to integrate the bugfix from "etch" into "lenny".  So I
> just tried to use "stg sync" on this patch.  It should be noted that
> this particular patch is the one that adds the Debian packaging files
> (as opposed to the other ones, which modify the upstream source
> files).
>
> Here is what I get.  I suppose that could have been expected, but that
> looks nevertheless annoying...

Have you modified the last try/except block in stgit.main.main? You should only get an error message, not the full backtrace.

The interesting messages are:

Show 5 quoted lines

> Synchronising "debian"...Error: File "debian/tau.files" added in branches but different
> Error: File "debian/tau-racy.files" added in branches but different
> Error: File "debian/rules" added in branches but different
> Error: File "debian/changelog" added in branches but different
> stg sync: GIT index merging failed (possible conflicts)

Which are what I would expect the 'sync' command to do - it tries a three-way merge between the debian//top, debian@etch//top and debian@etch//bottom (the order is probably different). Since the debian/rules etc. files aren't present in debian@etch//bottom and you modified them in debian@etch//top, you get a merge conflict - files added in both branches but different.

'stg status' should show the conflict files and you have to manually fix them. If you know that you only modified the patch in etch, you can do 'stg resolved -a -r patched'. You could also add 'recursive = True' argument to the git.merge() call in stgit.commands.sync.__branch_merge_patch(). This would perform a merge with 'git-merge-recursive' rather than 'git-read-tree -m' and it is supposed to do a smarter two-way merging. You might not get a conflict at all or, at least, the conflicted file would have some markers.

The reason I didn't add the recursive merge for 'sync' was the slowness of this operation (git-read-tree -m was faster but maybe git-merge-recursive was improved in the last weeks). Please try this and let me know whether you get better results so I can modify the 'sync' command.

Anyway, this is a limitation of 'sync' three-way merging and I couldn't find an easy solution. For me, the best thing is that if the patch was synchronised, there won't be any conflicts reported and I can be sure they are the same. You mentioned patches of patches some time ago but these would get lost on the first rebase (unless you want to have a patch made of sub-stacks but you get to a fractal structure with sub-sub-...-sub-stacks :-))

-- 
Catalin