Aryeh Gregor schreef:
In that case I would suggest we change trunk to be an unstable testing ground, and not be so rigorous about requiring that it be guaranteed to work at all times (although any brokenness should still be reverted as soon as it's discovered). Commit access to unstable could be given out more liberally, and committers could be encouraged to be more generous in committing patches to it if they look okay but maybe haven't been thoroughly reviewed.
Sounds good.
Since stable shouldn't lag too far behind unstable, people should still be able to submit patches relative to stable, which would be reviewed and enabled on Wikipedia and so probably not horribly broken. People running trunk production wikis would switch to stable, while people running development-only wikis would use unstable.
This way we'd seem to get the best of all worlds. Of course, we'd probably need to use a DVCS for sanity.
My point was more or less that we can improve a lot while still using SVN. I'm not dead set against moving to another VCS, but I do believe we should try to get the most out of SVN before considering a merge.
We could also do that in SVN with a stableapi branch, into which I would merge reviewed commits. Brion and Tim would then merge the includes/api directory from my branch instead of trunk, after brief review.
Yes, but merges in SVN are *terrible*.
Tell me about it. I spent quite a lot of time being annoyed at merge conflicts in the apiedit branch (and later the ApiEdit_Vodafone branch), but that was mostly because I didn't run the merge often enough (only like once a month).
svn merge just applies a diff, with no authorship info preserved, all as one giant commit. DVCSes actually copy all the commits on merge, so all the authorship info (author name, date) and the full list of individual changes and commit messages is preserved. You can easily cherry-pick individual commits into your stable branch, or pull a whole bunch, without any merge commits, without svn log making it look like you wrote the commit (commit info still exists, but it's separate from authorship info), etc.
That would be a good argument for moving to another VCS; branches would also be less cumbersome.
Roan Kattouw (Catrope)