Git and tagging hook

Hello,

It seems that Git (at least v1.5.6) does not offer hooks on tag creation (a pre-tag and a post-tag hook). I need such a hook for integrating tag activities with an issue-tracker. Is it possible to add this hook ?

I had asked about this in the past, but did not receive a response.

http://bugzilla.mkgnu.net/show_bug.cgi?id=991

Thanks, Kristis

_______________________________________________ scmbug-users mailing list scmbug-users-G8y9j4K4DsPiwOUmbS1EgQ@public.gmane.org http://lists.mkgnu.net/cgi-bin/mailman/listinfo/scmbug-users

Kristis Makris wrote:

What you want is probably the post-receive or 'update' hooks on whatever repository you consider your public watering hole for your project. Integrating with an issue-tracker from the developers machine would be utterly stupid, as it would prevent tagging from happening while not connected.

But the post-receive is NOT executed when I apply a tag.

I want the integration when I apply the tag to a local repository, NOT only when I push/pull.

On Mon, 2008-10-06 at 09:17 +0200, Andreas Ericsson wrote:

Kristis Makris wrote:

If you are talking about taging locally, you can simply make an alias or do something after tagging. Search archives for description when it is worth to add a hook, and when it is not.

On Tue, 2008-10-07 at 19:28 +0200, Jakub Narebski wrote:

I am looking for a guarantee that is better than casually saying "simply". I will be providing the integration work to users that may not be as comfortable with making aliases.

I still don't see why a hook on local tagging is not available. Is it possible to add support in Git for such a hook ? Both pre-tag and post-tag.

_______________________________________________ scmbug-users mailing list scmbug-users-G8y9j4K4DsPiwOUmbS1EgQ@public.gmane.org http://lists.mkgnu.net/cgi-bin/mailman/listinfo/scmbug-users

Kristis Makris wrote:

Because noone's ever needed one before. If aliases can't do what you want, write a patch to support it and hope Junio accepts it. It's really quite straight-forward. Make sure you read Documentation/SubmittingPatches before you send it.

Note though that use of tags on the developer's side will still be up to the developer and not something you can force through other means than policy or convention.

On Wed, Oct 08, 2008 at 19:40:02 +0200, Andreas Ericsson wrote:

Care to explain why that would ever be useful? It's local, which means that:
 - the user can take it back without a trace it ever happened (git tag -d or
   even git update-ref -d) and
 - noone except the user will see it anyway, so it's not like they should
   care either.

Besides, you don't need git tag to create a tag in git, so the hook wouldn't really be guaranteed anyway (I mean, just like the commit hook is not -- you can still commit by calling write-tree, commit-tree and update-ref and avoid the hook).

Being possible was never a reason to add features to git and the less it was a reason to add hooks. And there does not seem to be a use-case that would clearly benefit from having such hooks, or at least none was shown on the list so far.

For integration with issue tracker, the local tag is neither final, nor useful to anybody except the user who did it until it hits the central repository. And working on the central repository directly does not seem like a good idea either.

Next time, don't include a members-only list in the CC, please. The usual action when replying to an email on git@vger is to "Reply All", which doesn't play nicely with such lists and will render one list archive incomplete (since people will inevitably cull that other list from their CC's after having sent to it once).

On Mon, 2008-10-06 at 09:20 +0200, Andreas Ericsson wrote:

Why would they cull that other list, if the usual action is to "Reply All" ?

Kristis Makris wrote:

Because it's a members-only list. Like me, I'm sure many will get annoyed enough to remember to remove it for further postings on the subject. Those nonsensical "awaiting moderator approval" emails are irritating noise, and since they're "To: me" and get a low spamrank they go straight to the high-priority folder. Many others on this list use similar filters as it's a neat way of dealing with several high-volume mailing lists sanely.

Jan, thanks for trying to clarify this for me.

I am working on adding integration support of Git with bug-trackers, using Scmbug. There may be an argument here towards/against distributed bug-trackers when it comes to Git.

Maybe there are things here I don't fully understand yet.

On Tue, 2008-10-14 at 19:22 +0200, Jan Hudec wrote:

I have two use cases:

(1) A developer maintains besides his local copy a local bug-tracking system in which he tracks his changes. We would like to apply various verification policies when he commits or tags. For example, for tagging we wants to ensure that he tags giving consistent labels to his intermediate builds. e.g. as in:

http://files.mkgnu.net/files/scmbug/SCMBUG_RELEASE_0-26-9/manual/html-single/manual.html#VERIFICATION-CHECKS-CONVENTION-BASED-LABELING

Or he may want to have Git force him to also supply a log message along with a tag, so that he can remember later more accurately why a tag was created and what it really captures. Even if Git (or other SCM systems) don't natively support log messages on tags. Scmbug plans to implement this.

http://bugzilla.mkgnu.net/show_bug.cgi?id=219

(2) I would like to apply various verification policies when work from a local repository is finally merged with the central repository. I assume there can/will be a central repository, and there is one "software product" that is being released somewhere among the many copies.

When its time to merge local changes to a central repository, the verification policies may deem that changes are not acceptable to be merged with the mainline. e.g. because log messages are too short, commits during the merge are issued against bugs in "a central" bugtracker that are either closed, assigned to someone else, or just plain wrong bug-numbers that belong to other products:

http://files.mkgnu.net/files/scmbug/SCMBUG_RELEASE_0-26-9/manual/html-single/manual.html#VERIFICATION-CHECKS-VALID-LOG-MESSAGE-SIZE http://files.mkgnu.net/files/scmbug/SCMBUG_RELEASE_0-26-9/manual/html-single/manual.html#VERIFICATION-CHECKS-OPEN-BUG-STATE http://files.mkgnu.net/files/scmbug/SCMBUG_RELEASE_0-26-9/manual/html-single/manual.html#VERIFICATION-CHECKS-VALID-BUG-OWNER http://files.mkgnu.net/files/scmbug/SCMBUG_RELEASE_0-26-9/manual/html-single/manual.html#VERIFICATION-CHECKS-VALID-PRODUCT-NAME

(I'm not very clear whether this is how Git works)

Does someone get to write-up a brand new log comment during the merge and the merge totally disregards older log comments ? My understanding is that log comments on the local copy are preserved (and will need to be mapped to bug-numbers in the central bug-tracker.

Thus the local verification policies may need to have already been configured to comply with future verification policies of the central repository. Else (perhaps considerable) mappings/adjustments will be needed during the push to the central copy.

I'm assuming someone who follows the recommended avenue of using Git wants the advantages of hooks. I certainly can't force people that bypass hooks to use them.

The local tag is useful to the local user and his local bug-tracker. He can have tag operations intercepted so that the tag names show up as versions in his bug-tracker. In this way he can keep track of which bugs still exist or have recently been introduced/discovered to his local copy, before he decides to publish his polished, final version:

http://files.mkgnu.net/files/scmbug/SCMBUG_RELEASE_0-26-9/manual/html-single/manual.html#TAGS

And his "local bug-tracker" may be reachable on the web and useful by others that take a peek at the users progress (even fetching it with Git).

_______________________________________________ scmbug-users mailing list scmbug-users-G8y9j4K4DsPiwOUmbS1EgQ@public.gmane.org http://lists.mkgnu.net/cgi-bin/mailman/listinfo/scmbug-users

Kristis Makris wrote:

I'm guessing releases will be cut from some sort of central or official repository anyway, so I fail to see why tags would need to be verified at create-time rather than at push-time. It's sometimes useful to create tags for ones own personal use, and commit messages that are no more than "wip", without signoff or anything. This needs to be implemented at the receiving end. Not at each developers end.

git supports optional log-messages on tags. There are two different kinds of tags in git; "annotated" (with logmessage) and "lightweight" (without logmessage). It's up to each user which sort of tag to create. Using the example update hook, lightweight tags are by default not allowed to be pushed to a repository.

Merges happen in the local repository. When a merge is pushed to the "release" repo, you can analyze the merges and all commits that the sender is trying to push.

That sort of work belongs in the update hook then. Cautious users, or release engineers, might want to enable pre-merge hooks and whatnot.

Try and find out. It would have been faster than writing that paragraph ;-)

Hooks can also be disabled, and they aren't enabled by default. They're also not cloned along with a repository (that would be stupid, as I most certainly don't want the email-on-commit hooks we're using at work), so installing said hooks would still be a manual job done by each developer that wishes to comply with the policy you're outlining. I have a hard time seeing how that can benefit the open community.

Relying on hooks on the developer side is fragile and *will* break. It will break often, and it will break badly. Any sort of "this-commit-is-releasable" verification simply *has* to be done in the release repository. Otherwise you'll be limiting how devs can use the scm while gaining absolutely nothing (since it has to be done in the release repo too for those times when the dev forgot to enable hook X in a newly cloned repository).

I still haven't figured out what the over-all plan is, so my advice and warnings may be counter-productive at worst. However, http://files.mkgnu.net/files/scmbug/doc/latest_manual/html-multi/x113.html just doesn't make sense to me. It's invalidated by clear and concise commit messages (which aren't always there, but education is nine times out of ten better than enforcement; It's better to understand *why* 6x7 = 42 rather than just knowing that it's so).

On Tue, 14 Oct 2008, Kristis Makris wrote:

Git supports using tags for a variety of other purposes in addition to the traditional ones (for example, you can tag a version as "the changes I'm sure of" or "the version I know is broken"; furthermore, your regression testing system could give you a tag that tags the commit you sent for testing with the test results). The bug tracker integration should only care about the traditional use (providing a persistant, human-recognizable name for a revision). Actually, it would probably be best, for integration with git, to skip tags entirely, and use the hash. With projects using git, it is routine to know that the bug was found in some particular flawed commit that didn't get tagged.

In general, you send to the central repository, and it rejects you if (a) a merge would be required; you have to fetch and merge in your local repository; or (b) there's something wrong with the changes you're making.

For (a), you make a merge commit and try again; the new commit has a log message, but the old commits retain their log messages. For (b), you make a different set of commits that does follow the required policy. "git rebase -i origin/master" will guide you through this process. In general, you do this before sharing your changes with anybody else, or at least before sharing them with anyone who cares about the end result, so that other people see nice commits. It's also possible to rearrange the changes you made in a bunch of local commits into a series that looks nice and makes sense and follows the project rules, when your initial work had you introducing a lot of silly mistakes and fixing them.

In general, there's a ton of adjustment needed between working on a project and pushing to the central location in any system. With git, however, version control may be used locally before these adjustments are made, and this provides a huge benefit in terms of being able to prepare commits just how they should be, and in terms of being able to avoid losing work during the adjustments.

In general, you want to have a local understanding of the central repository rules, so that you can do this mapping while you don't have network, but there's no reason to prevent saving your work before it conforms.

	-Daniel
*This .sig left intentionally blank*

On Tue, Oct 14, 2008 at 11:03:21 -0700, Kristis Makris wrote:

That already requires using additional interface to git alone. Using an alias for creating tags that are synchronized with that local bug-tracker is not that much more complex than IMHO. Besides you will sometimes want to use git tags, branches, commits and other things *without* synchronizing them to the local bug-tracker even if you usually synchronize -- they are often very useful for manipulating changes.

Git does support log messages on tags. It has unannotated tags (which are just refs) and annotated tags which have a special tag object with the log message (and optionally PGP signature).

Indeed, the locally made commits are transfered to the upstream repository as they are. In fact, because the commit id is a SHA1 checksum of all it's content, including the message and the parent ids (and therefore complete history), when the commit message is changed, it is no longer the same commit.

However, git provides many tools (commit --amend, rebase -i, filter-branch) and has additional extensions (stgit, topgit), that make it easy to create a new commit based on another one with some change. This is extremely useful and many people use it really often.

Such new commit will not replace the previous one -- it has different checksum -- so it's not good thing to do when other people already based further changes on your commit. But it's very useful for handling work in progress, quickly diverting to different tasks, making experiments and such.

Therefore it's the push that casts things in stone, but before that you can easily take back and redo both commits and tags. Additionally it's very useful to sometimes do commits and tags that you intend to replace later just to temporarily record some interesting state eg. to divert to other bug that suddenly got higher priority or to try out different approach.

Thus you only want to run the checks as warnings locally and probably want to have an option to avoid them in a particular case for performance reasons (all this stuff is so useful because it's fast). And since the tags that are ment for publication you will usually pull shortly after making them and because creating a tag is rather simple, I would say that running the check on push is sufficient there and the alias helps if you'd really want to run the check earlier. For commits you can of course use the pre-commit and post-commit hooks.

I would rather recommend having per-developer branches (you can actually have branch hierarchies, so it's rather per-developer branch directories) on the central repo, where the users would push their work they consider final enough to show to anybody. Than you can have a "pending" (or "for review" or both or whatever) state in the central bug-tracker for issues with fixes on such developer branches. That saves the hassle of installing per-developer trackers and gives developer more freedom to create temporary stuff locally.

That is not to say that your use case makes no sense. I am just trying to suggest a workflow, that might fit better with the existing practices used with git and maybe requiring less work to implement.

As for people replacing their local commits, this is common especially in Linux (and Git) development model. For Linux patches need to be sent split into logical steps to make it easier to review them, which is quite important for a critical piece of code like kernel. But everybody will inevitably make mistakes when implementing the changes; these mistakes must not appear in the final submission though, because they would interfere with review. When developing in Git, each commit will produce one patch in the submission, so people go and redo their commits multiple times to make them as much readable as possible and fix bugs they made earlier.

Jan Hudec wrote:

It's common everywhere where projects use peer review.

Results 1 - 50 of about 14,600 for '"PATCH v2" vger.kernel.org' Results 1 - 50 of about 360,000 for "PATCH v2"

So yes, it is a very common practice and anything interacting with a remote database based on each commit action will almost certainly be doing something wrong quite a lot of the time, especially since not all the patch-series end up being incorporated into a release anyway.

The only sane integration point is the public watering-hole repository used for the project, and especially the release branch (or the "for-linus" branches around the world for sub-projects).