hg push with mq applied

[Robin Farine]

On Fri May 25 2007 11:29, Bernd Schubert wrote:
> Robin Farine wrote:
> > Hi,
> >
> > On Thu May 24 2007 19:47, Bernd Schubert wrote:
> >> I'm adding patches to a project and I'm doing this on my local
> >> disk. In order store this to save (backuped) storage I would
> >> like to push everything to the backuped dir,  but this fails
> >> with
> >>
> >> abort: source has mq patches applied
> >>
> >> Well I know that it has mq patches applied and I would like to
> >> push these patches as well.
> >
> > But what happens when later you make changes to some of the
> > patches and try to push again ?
>
> Don't know what you mean, I can't push at all, so how should
> changing the patches help?

More explicitly, supposing Hg would accept to push mq managed change
sets (emulated by the "mv .hg/patches/status ..." hack below), what
should Hg do when you push with an applied updated patch:

$ hg init A
$ cd A
$ echo "blah" > blah
$ hg add blah
$ hg commit -m "create blah"
$ cd ..
$ hg clone A B
1 files updated, 0 files merged, 0 files removed, 0 files unresolved
$ cd B
$ hg qnew blah.patch
$ echo bli >> blah
$ hg qrefresh
$ mv .hg/patches/status .hg/patches/status.bak
$ hg push default
pushing to /tmp/mq-clone/A
searching for changes
adding changesets
adding manifests
adding file changes
added 1 changesets with 1 changes to 1 files
$ mv .hg/patches/status.bak .hg/patches/status
$ vi blah
$ hg qrefresh
$ mv .hg/patches/status .hg/patches/status.bak
$ hg push default
pushing to /tmp/mq-clone/A
searching for changes
abort: push creates new remote branches!
(did you forget to merge? use push -f to force)
$ cd ..
$ hg log -R A
changeset:   1:faa0bb8c87f9
tag:         tip
user:        Robin Farine <robin.farine at terminus.org>
date:        Fri May 25 14:46:52 2007 +0200
summary:     patch queue: blah.patch

changeset:   0:382f96d90a9f
user:        Robin Farine <robin.farine at terminus.org>
date:        Fri May 25 14:46:15 2007 +0200
summary:     create blah

$ hg log -R B
changeset:   1:4d4810ee6122
tag:         tip
user:        Robin Farine <robin.farine at terminus.org>
date:        Fri May 25 14:47:36 2007 +0200
summary:     patch queue: blah.patch

changeset:   0:382f96d90a9f
user:        Robin Farine <robin.farine at terminus.org>
date:        Fri May 25 14:46:15 2007 +0200
summary:     create blah


Should Hg replace change set 1 in A with the one in B or create a new
head ? Or should mq redefine 'push' and 'pull' to cope with this
situation ? What are the implications of each case ?

To me, it seems easier and safer to never publish a repo with mq patches
applied, but publish the patch series separately, perhaps as Gert suggests,
by having the patch queue managed with Hg as well.

Robin