sync of uncommitted code between 2 computers

[Fabien Tassin]

Binary package hint: bzr

It would be nice to improve usability of bzr between 2 computers or more to synchronize uncommitted code. This has been discussed on IRC when I first bring the topic there.

Use case:

1/ Max works from home (computer A) or from the office (computer B)
2/ Max is working on a project where the workflow is to commit to the central/shared branch (on computer C) only complete patches (the tree must always build)
3/ The patch Max is working on is taking several days so he cannot commit to C to pass code between A and B (because it's not complete). Fake commit/uncommits are also unacceptable because the branch is shared and it will force A or B to push in override mode which could easily lead to loss of data.
4/ shared filesystems between A and B are not an option, and transport has to be on ssh, preferably doable from the side only (the other side could be firewalled, NATed or simply use a dynamic IP)

It is possible to scp the whole tree from A to B, then diff the two working trees and patch B but it's laborious.
Another option would be to revert B and pass a bzr diff from A to B, but it could lead to loss of data on B if B had diverged (a merge would have prevented that).
A better proposal on IRC was to use merge --uncommitted but it doesn't currently work on bzr+ssh or sftp ("is not a local path").

a bzr sync would be nice to have, with its own remembered locations, and with the same merge/conflicts mechanism as regular commits.

[Wesley J. Landaker (wjl)]

I would also love to see this, as the single-person-multiple-computer uncommitted-code synchronization issue comes up a lot when developing code from multiple locations (.e.g home/work, or laptop/desktop).

I think having "bzr merge --uncomitted" work on remote paths would pretty much solve this.

Here is a workaround in the meantime:
cd $LOCAL_BRANCH
rsync -a $REMOTE_BRANCH ../tmp-remote-branch
bzr merge --uncomitted ../tmp-remote-branch
rm -rf ../tmp-remote-branch

[Fabien Tassin (fta)]

this bug is indeed lacking love.

The bzr guys are probably working with a single computer, surely a laptop. Otherwise, they would have a need for something like this.

[Jelmer Vernooij (jelmer)]

I generally just commit and push in situations like this.
--
Jelmer Vernooij <email address hidden> - http://samba.org/~jelmer/
Jabber: <email address hidden>

[Vincent Ladeuil (vila)]

> The bzr guys are probably working with a single computer, surely a laptop.

:-)

Not really, but anyway.

If you can't commit to *a* branch, why not commit to *another* *private* branch, enjoying the comfort of a real branch and, when you're ready, merge back in your integration branch and push to the public one from there ?

bzr provides many ways to work with dumb servers with *branches*. Handling working trees on such servers is not going to happen anytime soon if only because dumb *protocols* doesn't provide enough support for that.

And since *some* computer need to provide access to both sides (home/office) to allow sharing, you'd get more support from bzr by using branches on that computer.

[Wesley J. Landaker (wjl)]

For all the "just commit" type answers: consider why merge --uncommitted is useful locally. This feature was added in bzr 0.10. It's super useful.

It's really redundant, because you could just commit and then merge. Or commit to "*another* *private*" branch and then merge. But sometimes doing "merge --uncomitted" makes things a zillion times simpler. You don't have to bother with making another branch, or messing up your commit history, or *not* messing up your commit history by playing games with rebasing or uncommits or diff and patch ... or whatever.

I know the original reporter called for a "sync" operation, but really a "merge --uncomitted" that worked on remote trees would really address 99% of this use case.

Anyway, I know this is a wishlist bug, and you might want to reject this idea for any number of reasons, but please don't reject this idea just because you *could* do it a different (and much more painful) way.

[Jelmer Vernooij (jelmer)]

fwiw, I do think it's useful to support merge --uncommitted on remote
trees, but it is nontrivial to implement (remote working trees are not
supported atm).
--
Jelmer Vernooij <email address hidden> - http://samba.org/~jelmer/
Jabber: <email address hidden>