From: Neal Kreitzinger Date: Fri, 02 Mar 2012 04:03:37 GMT Subject: Re: Why Is There No Bug Tracker And Why Are Patches Sent Instead Of Pull Requests Message-ID: <4F504699.3070406@gmail.com> In-Reply-To: <7vmx80nt68.fsf@alter.siamese.dyndns.org> On 3/1/2012 11:10 AM, Junio C Hamano wrote: > > Something based on the idea mentioned in Ævar's message (downstream in > this thread) to seamlessly integrate with the e-mail traffic might have a > chance to succeed. I also think the integration must be two-way for it to > be useful. A summary of "new issues untouched for N weeks" and another > "older issues unclosed for N weeks" periodically sent here, or something. > > Perhaps collecting messages based on a handful of simple heuristics like > "A message mentioned the keyword 'bug', but no In-Reply-To for it from any > list regulars came in two weeks" might be a good place to start. > Why don't you just use git for your bug-tracking? With commit-messages, annotated-tags, signed-tags, branches, branch-descriptions, git-notes, submodules, sha-1's, send-email, patches, hooks, etc., not to mention all-kinds-of-files you can commit (screen-shots, bug-reports, etc.) it seems you have more than enough pieces to build a bug-tracker with git. For example: Some Ideas: (Setup) - The bug tracker repo could be a one-way mirror of git.git. That way the bug-reports are literally tied to the source but don't clutter git.git. - Create a bug tracking branch parallel to each release branch. - Create a bug-branch off-of the buggy-commit and do the fix and make the bug-tracking-admin updates around the bug branch. (Submitting Bugs) - Post-receive hook to assess new commits and send formatted email to mailing list, ie. new-bug push sends bug email to mailing list. - The mailing list creates "bug-report-commits" kind-of-like Junio mentioned above so the mailing list also feeding into it. - Bugs can be submitted directly to the bug repo and replicated to the mailing list via post-receive hook, and bugs can be submitted via the mailing list and automatically "forwarded" to the bug repo as "bug report" refs. - Screenshots could be a submodule for "large file" optimization. (Issue-Numbers) - The sha-1 of the originating bug-report could be its issue-id. - Tag the commit with a "bug tag". The sha-1 of the bug-tag could be the issue-id. - Calculate issue number from commit-counting to assign unique numbers. - The sha-1 of the buggy-commit could be the issue-id. (Status Updates) - Use git-notes to update issue status. - Use annotated-tags to update issue status. - Use commit-messages to update issue status (or a file in the commit). (Reporting) - Interrogate bug-formatted refnames with wildcards to generate queries of whats going on. - Use gitk for ad hoc or pre-defined gui queries of what-you-want-to-see. - Use git time-machine commands to see how old stuff is or whats new. - have a bug-report-id.txt file that people just modify and commit and use git-annotate to see the history of that bug-report. (Signoff) - Signed tags for authenticity. (Standardization) - Git-commands or git-aliases or git-contrib-scripts to make sure people do-it-right (format names/messages/files correctly), ie. templates. - Commit hook snippet that detects bug branch format and brings up bug commit template. - Rebase hook snippet that doesn't allow bug branches to be rebased. (Integration) - The bug tracking system becomes part of the release workflow, ie., bugs are only considered fixed when they graduate to an official release or something close. - The "bug" tracking system also tracks "enhancements" and they are integrated into new releases using that workflow. (Concurrency control) - Concurrent bug-report updates merge-conflict. - If someone beat-you-to-it then you cherry-pick or rebase on top of them. (Interfaces) - works from commandline - works with gitk, git-gui - works with gitweb (Quality) - Do you really need the greatest commercial-bug-tracker "x" or opensource-bug-tracker "y" to meet your needs? If you've done well with nothing then you probably don't need all the bells an whistles. Management-by-report is mainly for clueless managers who want to feel like they know whats going on, and to supervise people who don't-really-care. Since git people are on-the-ground and know whats going on, and are highly motivated individuals you probably don't really need all that fluff. I suspect you would already be using opensource-bug-tracker "y" if there was one you actually liked using. - There's probably someone on this list who thinks they can create a bug-tracker with git that's better-than-anything-else-out-there, and they're probably right. (Intuitive) - If you know how to use git then you know how to use git bug-tracker. - If you don't know how to use git then you probably shouldn't be submitting bug reports. (Open Source Reusability) - The git bug repo system can then be used by others internally as they apply that system to their own project's canonical repository to track their bugs internally. Some-Stuff-You-Already-Know: (Distributed) - clone, push, pull -- your using git. (Reliability) - you're using git so you know its uncorrupted, and if it does go down just clone from someone else and verify the checksum. (Support) - git@vger.kernel.org - If git.git is down there is not much point in bug-tracking it. I've been wanting to do this internally for a whle, but I'm still learning the git scm part. Some of the git technologies I listed above I haven't used yet so they may sound a bit off, but I think you can get the gist of it. I think git people in general would be enthusiastic about this, and if so, that would be the one thing that truly distinguishes this proposal from the others ;-) You could call the repo or system bug-git, git-err-done, or something like that. v/r, neal