{"thread":{"id":"22388","subject":"Re: mesa_7_7_branch -> master merges","startedAt":"2010-01-25T19:04:00Z","lastAt":"2010-01-26T10:59:55Z","messageCount":2,"participants":["tom fogal","Keith Whitwell"],"isPatch":false,"patchVersion":null,"patchTotal":null},"messages":[{"id":"132620","messageId":"auto-000021766217@sci.utah.edu","threadId":"22388","inReplyTo":"1264443264.3029.255.camel@jfonseca-laptop","subject":"Re: mesa_7_7_branch -> master merges","fromName":"tom fogal","fromEmail":"tfogal@alumni.unh.edu","sentAt":"2010-01-25T19:04:00Z","receivedAt":"2010-01-25T19:04:00Z","isPatch":false,"sender":{"key":"tfogal@alumni.unh.edu","avatar":"https://gravatar.com/avatar/a2f71bfbe12b2b73cad0be2386d554aa403fff7a89df2f79e812c97c7eab0498?d=mp&s=160"},"body":"I think we've touched on a core git workflow issue here, and its likely\nothers have hit this && have a solution, so I've added the git ML to\nthe CC list.\n\nGit: the situation in this repo is a fast-moving master that is\nincluding many changes to internal interfaces.  Stable branches just\nget bugfixes, and are periodically merged to master.  However, the more\nthe heads diverge, the more difficult it is for a bugfix to merge into\nthe head.  The major issue is that more experienced developers should\nreally weigh in on these merges, because they tend to automagically\nundo some of the interface changes.  Yet during such a delay, master\ninevitably moves, and the bugfixer has to do even more work to \"redo\"\nthe merge (and potentially get more review!).\n\nOf course, if there are two bugfixers trying to make separate changes\nin the same time period, this gets worse.\n\nIs there a workflow that can solve this issue?\n\n writes:\n> On Mon, 2010-01-25 at 09:52 -0800, tom fogal wrote:\n> > writes:\n> > [snip]\n> > > The ideal would be to peer-review the merges before committing,\n> > > but it seems difficult to do that with git, while preserving merge\n> > > history and not redoing merges.\n> > \n> > Google has developed an infrastructure to do peer review using git.\n> > `Gerrit':\n[snip]\n> Review infrastructures are nice. I'd have some bias towards\n> http://www.reviewboard.org/  by the similar reasons ;)\n\nHeh, yeah I can understand the bias ;)\n\nPersonally, I'm not keen on a review tool I can't use from the command\nline, or at least not-the-web.  Then again, my reviews wouldn't really\nbe important in Mesa, so my opinion is irrelevant here ;)\n\n> But automated infrastructures aside, my worry with reviewing merges is\n> the actual constraints that git has. For example, let's suppose the\n> following scenario:\n> \n> 1) Developer A merges a stable branch into master.\n> 2) After spending a bunch of time fixing conflicts the best he can, he\n> emails the patch to mesa3d-dev for peer review.\n> 3) Developer B checks in a change into master.\n> 4) Developer A takes feedback from list, updates the code, and commits.\n> 5) Developer A cannot push because remote head has moved.\n> \n> So what can Developer A do now?\n>\n> a) Redo the merge, using the new master head.\n> b) Rebase the merge on top of the new head (I'm not sure it works, or\n> that it preserves branch history)\n> c) Double merge, i.e., merge its local head with the new master head.\n\nHrm, I was thinking of some sort of staging branch, but I can't think\nof a good way to make it work.  The crux of the issue seems to be that\na developer needs to somehow give a version control promise that they\nwill do the merge, even if the merge isn't done yet, because otherwise\nanyone else coming afterwards will duplicate the work (potentially\nincorrectly).  That would mean some kind of lock though, which sounds\nlike a terrible idea...\n\n-tom\n\n------------------------------------------------------------------------------\nThe Planet: dedicated and managed hosting, cloud storage, colocation\nStay online with enterprise data centers and the best network in the business\nChoose flexible plans and management services without long-term contracts\nPersonal 24x7 support from experience hosting pros just a phone call away.\nhttp://p.sf.net/sfu/theplanet-com\n"},{"id":"132666","messageId":"1264503595.18994.5.camel@toffee","threadId":"22388","inReplyTo":"auto-000021766217@sci.utah.edu","subject":"Re: [Mesa3d-dev] mesa_7_7_branch -> master merges","fromName":"Keith Whitwell","fromEmail":"keithw@vmware.com","sentAt":"2010-01-26T10:59:55Z","receivedAt":"2010-01-26T10:59:55Z","isPatch":false,"sender":{"key":"keithw@vmware.com","avatar":null},"body":"On Mon, 2010-01-25 at 11:04 -0800, tom fogal wrote:\n> I think we've touched on a core git workflow issue here, and its likely\n> others have hit this && have a solution, so I've added the git ML to\n> the CC list.\n> \n> Git: the situation in this repo is a fast-moving master that is\n> including many changes to internal interfaces.  Stable branches just\n> get bugfixes, and are periodically merged to master.  However, the more\n> the heads diverge, the more difficult it is for a bugfix to merge into\n> the head.  The major issue is that more experienced developers should\n> really weigh in on these merges, because they tend to automagically\n> undo some of the interface changes.  Yet during such a delay, master\n> inevitably moves, and the bugfixer has to do even more work to \"redo\"\n> the merge (and potentially get more review!).\n> \n> Of course, if there are two bugfixers trying to make separate changes\n> in the same time period, this gets worse.\n> \n> Is there a workflow that can solve this issue?\n> \n\nSpeaking from the Mesa side, I think part of our problem is that it's\nnot easy to build the entire mesa tree, which means that the developer\ndoing the merge cannot even compile-test the result, meaning that many\ntrivial failures go unnoticed.\n\nI'd argue that if we had a maximal mesa build target that compiled\n*everything*, regardless of whether it produced drivers or not, we'd\nhave a much better chance of catching bogus merge droppings.\n\nDespite Jose's valid concerns, I'd still argue that the situation we\nhave now is superior to what came before - where people were supposed to\nbe cherry-picking bugfixes but more likely they were forgotten or it\nfell on Brian's shoulders to do manually.\n\nKeith\n"}]}