Hi all, I think the github mirror we set up for testing purposes has been more or less a success: https://github.com/yt-project/yt-test The mirroring seems to work well: the github repository is staying in sync with the bitbucket repo every time we merge pull requests on bitbucket. To make further work on transitioning to github easier, I'd like to settle on what the final converted repo should like. For the purposes of discussion, I'd like to propose simply renaming the yt-project/yt-test repo to be yt-project/yt and agreeing that the current conversion is satisfactory. This means we'll agree not to use an alternate conversion technique that will change git commit hashes. Our mirror bot uses hg-git, and the converted repo hg-git produces isn't necessarily perfect. In particular, two issues have come up: * Since git is more selective than mercurial about e-mail and username formatting, there are a number of commits associated with dummy e-mail addresses like none@none * Since we do most of our development on a branch named "yt" (not "default" as is typical of mercurial repos) almost every commit in our history has a comment in the git commit message recording that it hapenned on the "yt" or "stable" branch. This makes it possible to round-trip from hg to git and back to hg, but it's also some extra noise in every single commit message. We could in principle ameliorate these issues by using something besides hg-git to do the conversion or by modifying hg-git to avoid these issues. The former options will lose the nice property of being able to round-trip and for others to independently export work from their forks. The latter option will require us to modify hg-git and waiting for upstream hg-git to review our changes, which might take a while, or telling everyone to use a hacked version of hg-git. Personally, I don't think either of those options are worth the hassle for what to me are minor annoyances. I think I've also ameliorated the username/e-mail ambiguity in the converted history by adding a .mailmap file to the repo: https://bitbucket.org/yt_analysis/yt/pull-requests/2582 In addition, the mirroring process we have now seems to work well, so just renaming the yt-test repo and calling the conversion final avoids a bunch more work which will block further progress on the conversion. What do you all think? It would be great to get people's +/-1 on renaming yt-test to yt on github.com/yt-project. Once we have a final repo name we will be able to start porting tooling like the website generator without worrying about the name or content of the final repo changing. -Nathan
On 04/18/2017 09:34 PM, Nathan Goldbaum wrote:
I think we should recreate it from scratch. I force-pushed to it once or twice stripping some commits. There's a high chance that one of the forks still have those, and they can be accidentally reintroduced. Moreover, it only has master(yt) branch. It lacks maintenance(stable), yt2(yt-2.x) and all the tags. Cheers, Kacper
On Tue, Apr 18, 2017 at 9:42 PM, Kacper Kowalik <xarthisius.kk@gmail.com> wrote:
Good point! I didn't realize you force-pushed. Did you force-push in a way that would overwrite commits created in hg? In that case I agree with you that we need to redo the conversion from scratch. If not, I don't think that requires recreating the repo, just pushing. I think all you need to do is add an "hg push" after this line (i.e. push with no explicit bookmark, so all heads are pushed to the local git repo, we still need the equivalent of "hg push -B master" so the master branch gets updated correctly): https://bitbucket.org/yt_analysis/fido-prhandler/src/ e9a6e927d6fc4ce73e8f23908f463894c18ac430/handle_mirror_ webhook.py?at=default&fileviewer=file-view-default# handle_mirror_webhook.py-40 And then make it so this line does "git push --all": https://bitbucket.org/yt_analysis/fido-prhandler/src/ e9a6e927d6fc4ce73e8f23908f463894c18ac430/handle_mirror_ webhook.py?at=default&fileviewer=file-view-default# handle_mirror_webhook.py-47
On 04/18/2017 09:47 PM, Nathan Goldbaum wrote:
AFAIR I stripped commits resulting from https://github.com/yt-project/yt-test/pull/1
Both branches and all tags are now pushed. Cheers, Kacerp
On Tue, Apr 18, 2017 at 10:31 PM, Kacper Kowalik <xarthisius.kk@gmail.com> wrote:
I guess those commits are still in the reflog? I don't think in practice that will matter.
Thanks! Any chance we can keep the names "yt-2.x" and "stable" instead of "yt-2" and "maint"? I think that will help reduce confusion and ease porting scripts.
On 04/18/2017 10:35 PM, Nathan Goldbaum wrote:
I don't know how to do that :) What we are mapping to git branches are hg bookmarks, and you can't have bookmark matching named branch: $ hg bookmark yt-2.x abort: a bookmark cannot have the name of an existing branch Cheers, Kacper
On Wed, Apr 19, 2017 at 7:40 AM, Kacper Kowalik <xarthisius.kk@gmail.com> wrote:
I created a pull request that should fix this issue: https://github.com/yt-project/fido-prhandler/pull/1 In addition this pull request sets the mirror target to the final repo location, in expectation of getting approval in this thread.
I'm okay with renaming it, but honestly I would kind of like to get rid of our first PR being "test" and our second being "grid visitors". I *think*, but am not *sure*, that this might mess up a bunch of things with getting issue number redirection. On Wed, Apr 19, 2017 at 11:08 AM, Nathan Goldbaum <nathan12343@gmail.com> wrote:
participants (4)
-
John ZuHone -
Kacper Kowalik -
Matthew Turk -
Nathan Goldbaum