Re: [RTC/PATCH] Add 'update-branch' hook
Felipe Contreras felipe.contre...@gmail.com writes: I have a branch which should always be recompiled on update; post-update-branch would be a good place for that. And why would pre-update-branch not serve that purpose? Because the code that needs to be compiled is not yet in the workspace And it won't be in 'post-update-branch' either. Then you are using a very odd definition of post update % git checkout master % git branch feature-a stable - update-branch hook will be called here The hook will get 'feature-a' as the first argument, but the code in the workspace would correspond to 'master'; the checked out branch (pre or post). Then the hooks should be called 'pre-branch', 'post-branch'; there is no update involved. The hook I need is actually post-merge, since merge is the command that updates the workspace. Sorry for the noise. -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [RTC/PATCH] Add 'update-branch' hook
Felipe Contreras felipe.contre...@gmail.com writes: Stephen Leake wrote: Felipe Contreras felipe.contre...@gmail.com writes: Ilya Bobyr wrote: On 4/21/2014 2:17 PM, Felipe Contreras wrote: Ilya Bobyr wrote: Also, most have names that start with either pre- or post-. It seems reasonable for both pre-update-branch and post-update-branch to exist. I don't see what would be the point in that. Do you see the point in the other hooks doing that? Yes, there a reason for the existance of those hooks. Now tell me why would anybody use post-update-branch instead of pre-update-branch? I have a branch which should always be recompiled on update; post-update-branch would be a good place for that. And why would pre-update-branch not serve that purpose? Because the code that needs to be compiled is not yet in the workspace -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [RTC/PATCH] Add 'update-branch' hook
Felipe Contreras felipe.contre...@gmail.com writes: Ilya Bobyr wrote: On 4/21/2014 2:17 PM, Felipe Contreras wrote: Ilya Bobyr wrote: Also, most have names that start with either pre- or post-. It seems reasonable for both pre-update-branch and post-update-branch to exist. I don't see what would be the point in that. Do you see the point in the other hooks doing that? Yes, there a reason for the existance of those hooks. Now tell me why would anybody use post-update-branch instead of pre-update-branch? I have a branch which should always be recompiled on update; post-update-branch would be a good place for that. -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: emacs buffer names
Duy Nguyen pclo...@gmail.com writes: The --daemon part is probably not worth mentioning because I always have one emacs window open. The problem is switch-buffer based on file name can be confusing (git.c and git.c2, which belongs to which checkout?). I ended up modifying files in the wrong checkout all the time. (setq uniquify-buffer-name-style (quote post-forward-angle-brackets) nil (uniquify)) This puts enough of the directory name in the buffer name to make the buffer names unique; very helpful! -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: `git stash pop` UX Problem
Matthieu Moy matthieu@grenoble-inp.fr writes: Stephen Leake stephen_le...@stephe-leake.org writes: So a message merge complete; you can drop the stash would be the most git should do. From the user experience point of view, that would be good. It could bother some users, but we have advice.* to silent this kind of warnings. snip explanation of implementation issues So, I wouldn't object, but I don't think the implementation cost is worth the benefit. Ok, that makes sense. -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: `git stash pop` UX Problem
Junio C Hamano gits...@pobox.com writes: Stephen Leake stephen_le...@stephe-leake.org writes: Matthieu Moy matthieu@grenoble-inp.fr writes: li...@haller-berlin.de (Stefan Haller) writes: Your intention was clearly to drop the stash, it just wasn't dropped because of the conflict. Dropping it automatically once the conflict is resolved would be nice. Your intention when you ran git stash pop, yes. Your intention when you ran git add, I call that guessing. You might be adding other files for other reasons. But if you add a file that does resolve a conflict caused by 'git stash pop', it is not guessing. The only thing you know for sure is that the user has consumed _one_ part of the stashed change, no? What if the stash had changes for more than one path? Count the unmerged paths in the index; when the count is zero, all conflicts are resolved. paths in the stash that had no conflicts are already in the index. So _if_ there is nothing going on except finishing the stash pop, an unmerged path count of zero means you are done with the stash, and it can be dropped. At the time of git add $path, can you reliably tell if the conflict to the $path the user is resolving came from a previous git stash pop, not from any other mergy operations, e.g. git stash apply or git apply -3? This is the real problem. I can impose a rule on my team of don't do more than one merge at a time by implementing that in the front-end, but git can't assume that. -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [RFC 0/3] Make git more user-friendly during a merge conflict
David Kastrup d...@gnu.org writes: Stephen Leake stephen_le...@stephe-leake.org writes: David Kastrup d...@gnu.org writes: Stephen Leake stephen_le...@stephe-leake.org writes: David Kastrup d...@gnu.org writes: do the right thing commands also tend to do the wrong thing occasionally with potentially disastrous results when they are used in scripts where the followup actions rely on the actual result. That is bad, and should not be allowed. On the other hand, I have yet to see an actual use case of bad behavior in this discussion. Huh. http://permalink.gmane.org/gmane.comp.version-control.git/242744 That's about backward incompatibility, which is bad, but not what I was talking about above. No, it isn't. I quote: I sometimes run git reset during a merge to only reset the index and then examine the changes introduced by the merge. With your changes, someone doing so would abort the merge and discard the merge resolution. I very rarely do this, but even rarely, I wouldn't like Git to start droping data silently for me ;-). You should not make statements like I have yet to see an actual use case of bad behavior in this discussion when you actually mean I have not yet seen anything I would be interested in doing myself. Clearly I misunderstood your point. Merely repeating the same statement that I misunderstood, and adding a misunderstanding of what I said, is not helpful. So let me see if I can expand on your use case: - you do 'git merge', which results in conflicts - you edit some workspace files to resolve some of those conflicts (I added this step later, since it was implied but not explicit) - you do 'git reset', intending 'git reset --mixed' (because that is the current default) Actually, I can't find a precise definition of 'git reset'. Here is the synopsis from the man page for 'git-reset' (from git 1.7.9): git reset [-q] [commit] [--] paths... git reset (--patch | -p) [commit] [--] [paths...] git reset (--soft | --mixed | --hard | --merge | --keep) [-q] [commit] In 'git reset', there is no path, so it must be the second or third form. But those _require_ one of the -- options. So 'git reset' is illegal. Clearly something is wrong here; apparently the third line should be: git reset [--soft | --mixed | --hard | --merge | --keep] [-q] [commit] with '--mixed' as the default, as is stated later. (perhaps the original intent was to not have a default for the third form, but that got changed sometime?). This command resets the index but not the working tree. I'm not clear what reset the index means here; does it mean remove all entries from the index, or reset the index to some previous state? In other man pages, reset can have either meaning depending on context. - then you examine changes introduced by the merge. I don't know what this means in detail. Before resetting the index, you could diff a workspace file against either HEAD or index. Now you can only diff against HEAD, so I don't understand why you wanted to reset the index. That's not relevant to this use case; I'll just accept that resetting the index is a useful thing to do here. But I would like to understand why. - with the do the right thing patch, 'git reset' does 'git reset --merge' instead That Resets the index and updates the files in the working tree that are different between commit and HEAD. commit in this case defaults to HEAD, so the working tree is not changed. So as I understand it, this does _not_ lose your conflict resolutions. In fact, it now seems that 'git reset --mixed' is always the same as 'git reset --merge'. So I must be missing something! -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [RFC 0/3] Make git more user-friendly during a merge conflict
Matthieu Moy matthieu@grenoble-inp.fr writes: $ git status On branch master nothing to commit, working directory clean $ ok, you've lost your conflict resolutions. In fact, it now seems that 'git reset --mixed' is always the same as 'git reset --merge'. So I must be missing something! git reset --merge is an alias for git merge --abort (IIRC, it's actually the other way around). Essentially, it reverts, or tries to revert everything (worktree and index) as it was before the merge. That includes throwing away conflict resolution. Ok. Now, I do agree that the documentation of git reset is terrible, Ok, good. So is this a sufficient bug report to request that the documentation be fixed? (I obviously don't know enough to even think about submitting a patch). -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [RFC 0/3] Make git more user-friendly during a merge conflict
Jonathan Nieder jrnie...@gmail.com writes: Hi, Andrew Wong wrote: The first two patches are just about rewording a message, and adding messages to tell users to use git merge --abort to abort a merge. Sounds like a good idea. I look forward to reading the patches. We could stop here and hope that the users would read the messages, but I think git could be a bit more user-friendly. The last patch might be a bit more controversial. It changes the default behavior of git reset to default to git reset --merge during a merge conflict. I imagine that's what the user would want most of the time, and not git reset --mixed. I don't think that's a good idea. I'm not sure what new users would expect; As a newbie, I would like to know how to abort the merge, so I like this message. in any case, making the command context-dependent just makes the learning process harder imho. I like commands that do the right thing. So no, this would not be confusing. And for experienced users, this would be a bad regression. Backward incompatibility is a real concern. It might be best if git reset (with _no_ option) be made to error out, so all users have to specify what they want. The transition process Junio proposed sounds good to me. -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: Branch Name Case Sensitivity
Karsten Blees karsten.bl...@gmail.com writes: If I understand the issue correctly, the problem is that packed-refs are always case-sensitive, even if core.ignorecase=true. Perhaps that could be changed? if core.ignorecase=true, packed-refs should be compared with case-insensitive string compares. -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [RFC 0/3] Make git more user-friendly during a merge conflict
David Kastrup d...@gnu.org writes: Stephen Leake stephen_le...@stephe-leake.org writes: I like commands that do the right thing. So no, this would not be confusing. I _hate_ commands that think they know better than to do what they are told. In particular when doing destructive things. And just because _you_ like them does not mean they are not confusing. Ok, I should have said not confusing for me. People differ. In the long run, it is much more confusing if you come to rely on some commands doing the right thing while in other cases, the actually written thing is done. There should always be the option of telling git exactly what to do. In my emacs front end, the command that does the right thing is _called_ do-the-right-thing. All of the other commands do exactly as told. In this case, it is only git reset that would do the right thing, since you did _not_ tell it specifically what to do. Relying on a default is always problematic, in my experience; I much prefer no default to a default that people voted on 10 years ago, and now we are stuck with it. do the right thing commands also tend to do the wrong thing occasionally with potentially disastrous results when they are used in scripts where the followup actions rely on the actual result. That is bad, and should not be allowed. On the other hand, I have yet to see an actual use case of bad behavior in this discussion. -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: `git stash pop` UX Problem
Brandon McCaig bamcc...@gmail.com writes: On Thu, Feb 27, 2014 at 9:57 PM, Stephen Leake stephen_le...@stephe-leake.org wrote: You might be adding other files for other reasons. But if you add a file that does resolve a conflict caused by 'git stash pop', it is not guessing. Staging a file doesn't tell git that you resolved a conflict. Git will happily accept a blob full of conflict markers. Git doesn't know the difference. Git expects the user to know what is right. The user has the freedom to manipulate the index as they see fit, which means both adding and removing from it anytime they wish. But git has a notion of unresolved conflict. For example, when I have conflicts from a 'git stash pop', 'git status' shows: stephe@takver$ git status # On branch master # Unmerged paths: # (use git reset HEAD file... to unstage) # (use git add/rm file... as appropriate to mark resolution) # # both modified: CommandBasedAutonomous.java # both modified: DriveByInches.java # # ... How does it know those files are unmerged? I'm guessing it has recorded the fact that they had conflicts. Where does it record that? In fact, at this point, I have edited CommandBasedAutonomous.java to resolve the conflicts. But git apparently doesn't know that. So I do 'git add CommandBasedAutonomous.java', then 'git status': stephe@takver$ git status # On branch master # Changes to be committed: # (use git reset HEAD file... to unstage) # # modified: AerialAssist2014/src/org/usfirst/frc1939/AerialAssist2014/commands/CommandBasedAutonomous.java # # Unmerged paths: # (use git reset HEAD file... to unstage) # (use git add/rm file... as appropriate to mark resolution) # # both modified: AerialAssist2014/src/org/usfirst/frc1939/AerialAssist2014/commands/DriveByInches.java And git thinks that file is now merged. So it appears that adding a file _does_ tell git that the conflict is resolved. Or am I still missing something? So git add and git stash * are lower level tools; to get the effect we are asking for, we should use a front-end (which is why I'm writing one for Emacs :). You *can* use a front end, but I would argue that you shouldn't necessarily. Most third-party front ends only serve to confuse users. In general, they only cause problems and encourage ignorance. Won't happen here; I'm writing it. It may confuse other people, but not me :). Git is a very pure system. Hmm. We'll have to disagree on that. git gives the impression of having grown organically for quite a while, and therefore suffers from changing and competing design paradigms and conflicting requirements due to preserving backward compatiblity. monotone is much cleaner, since it has had very few design paradigm changes, and they were implemented cleanly, without preserving backward compatibility. monotone is not as flexible as git, but what I've seen so far could be added to monotone (I don't think it ever will be; monotone is dying as a project). We are probably using different definitions of pure here. It is up to the user to learn how to assemble those tools for good (and many front ends exist to help; sometimes arguably too many as it is, such as git-pull(1) for example). Yes. Which is why we are discussing how much help git should be while still learning the rules. This isn't a case of the API being wrong. This is a case of PEBKAC, IMO. (wikipedia to the rescue; PEBKAC = operator error) Yes, I'm not using it correctly, because I don't understand it yet. That's the definition of newbie. Dropping the stash after adding all changes to the index after a failed pop is not universal. Not universal, but it appears to be very common; it is certainly what I expect, as a newbie. So it could be the default as long as there is a configuration option to have it not do that. I _did_ RTFM (specifically the man page on 'git stash', and before that the git book at http://git-scm.com/documentation (which did not mention stash)); it did not explain the full cycle of how to resolve conflicts after stash pop. Perhaps there is a different manual that I could read instead? In particular, one that explains what unmerged paths means in the 'git status' output? The 'git-status' man page does not do that. -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
git reset path returns unwanted failure status
The use case: I'm doing a 'git stash pop'; it had conflicts. At this point, 'git status' shows: # On branch master # Changes to be committed: # (use git reset HEAD file... to unstage) # # modified: Target.java # # Unmerged paths: # (use git reset HEAD file... to unstage) # (use git add/rm file... as appropriate to mark resolution) # # both modified: DriveByInches.java # # Changes not staged for commit: # (use git add file... to update what will be committed) # (use git checkout -- file... to discard changes in working directory) # # modified: CommandBasedAutonomous.java # # Untracked files: # (use git add file... to include in what will be committed) # # AerialAssist2014/src/org/usfirst/frc1939/AerialAssist2014/Autonomous/ As part of the conflict resolution, I decide to unstage Target.java: stephe@takver$ git reset Target.java Unstaged changes after reset: M CommandBasedAutonomous.java U DriveByInches.java M Target.java stephe@takver$ echo $? 1 The issue is the error status and the messages about other files. If I had not specified a path to 'git reset', the error status would make sense; those files were not reset. However, since the file I specified was reset, there should be no error. Similarly, if I specify no path to a git command, I expect warning messages about files in the workspace that might need attention. However, if I do specify a path, I expect warning messages about files in that path only. This can be stated more concisely if the default path is considered to be * (and recursive); don't error if the operation succeeded for all files in the path; don't warn about files not in the path. -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [RFC 0/3] Make git more user-friendly during a merge conflict
David Kastrup d...@gnu.org writes: Stephen Leake stephen_le...@stephe-leake.org writes: David Kastrup d...@gnu.org writes: do the right thing commands also tend to do the wrong thing occasionally with potentially disastrous results when they are used in scripts where the followup actions rely on the actual result. That is bad, and should not be allowed. On the other hand, I have yet to see an actual use case of bad behavior in this discussion. Huh. http://permalink.gmane.org/gmane.comp.version-control.git/242744 That's about backward incompatibility, which is bad, but not what I was talking about above. Specifically, the proposed change is: 'git reset' will have different default actions depending on context: - if a merge is not in progress, it will do 'git reset --mixed' - if a merge is in progress, it will do 'git reset --merge' Is there a use case where this will do the wrong thing? Of course, I fully understand that not being able to come up with a wrong thing use case is not the same as proving it cannot happen, especially for a system as complex as git. So it would be ok to say we don't do that so we are not exposed to unintended consequences. But wrong thing use cases are more convincing :). -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: `git stash pop` UX Problem
Matthieu Moy matthieu@grenoble-inp.fr writes: Stephen Leake stephen_le...@stephe-leake.org writes: So it appears that adding a file _does_ tell git that the conflict is resolved. Yes it does. Git _knows_ that you consider the conflict to be resolved. It cannot know how happy you are with the result. Similarly, in a conflicted merge, the last git add does not trigger a commit silently. And a silent commit would be much less serious than a silent data drop. Ok, I see your point now. So a message merge complete; you can drop the stash would be the most git should do. -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: `git stash pop` UX Problem
David Kastrup d...@gnu.org writes: Stephen Leake stephen_le...@stephe-leake.org writes: Brandon McCaig bamcc...@gmail.com writes: On Thu, Feb 27, 2014 at 9:57 PM, Stephen Leake stephen_le...@stephe-leake.org wrote: You might be adding other files for other reasons. But if you add a file that does resolve a conflict caused by 'git stash pop', it is not guessing. Staging a file doesn't tell git that you resolved a conflict. Git will happily accept a blob full of conflict markers. Git doesn't know the difference. Git expects the user to know what is right. The user has the freedom to manipulate the index as they see fit, which means both adding and removing from it anytime they wish. But git has a notion of unresolved conflict. Not really. It has a notion of unmerged path. snip The index contains the unmerged versions of the file. Possibly also the version with conflict markers, but it's been too long since I last checked. Paraphrasing, is this correct? the index contains both versions of the unmerged file; any file with more than one version in the index is unmerged. So what 'git add' does in this case is replace both versions of the file in the index with a new version. I was not aware that the git system could support more than one version of a file in one branch. That makes it more like monotone :). If you apply a stash with unmerged paths to a worktree/index, possibly containing unmerged paths of its own, possibly getting new unmerged paths by failing to apply the stash, you get unmerged paths from several different unresolved conflicts. Yes; doing too many things at once is a bad idea. But that should never cause git to lose data or do something wrong. At the same time, it seems all unmerged paths result from unresolved merge conflicts, so the two notions are equivalent for git? Git has no idea about the history of unmerged paths. So having git add modify the operation of git reset whenever git add overwrites an unmerged path in the index could lead to quite funny results. Ok; I'll take that as describing a large class of bad thing use cases. -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: `git stash pop` UX Problem
Junio C Hamano gits...@pobox.com writes: Stephen Leake stephen_le...@stephe-leake.org writes: One _could_ argue that stashed changes are what could be reflected to the working tree and form the source of the latter, but my gut feeling is that it is a rather weak argument. At that point you are talking about what you could potentially change in the working tree, No, I saved things in the stash on purpose. For example, I had changes that were not ready to commit, but I wanted to do a merge from upstream. I often save things by running git diff P.diff on purpose. Ok. How is that better than 'git stash save'? Should git status read these patches and tell me what paths I could change in the working tree by applying it? No, 'git stash save' appears to be the method git provides to do this, so it is the only one that git needs to support. (The content of 'P.diff' already tells you what paths are modified, as does 'git stash show') But I am new to git, so I could just be missing the point. Where does it end? Where we agree it ends :). There are workflows where the stash is not important; provide an option to 'git status' that means ignore stash. How is that different to tell those who want to know what are in the stash to type git stash list when they want to learn that information? You are correct, this is a question of style. The question is: Which style is best for git, considering the needs of newbies and seasoned users? As a newbie, I find these things confusing: - the stash status is not displayed by 'git status' - 'git add' does not report that all pending merge conflicts are now resolved. I'm sure I will discover other confusing things in the future :). I am a seasoned user of CM systems in general; in all cases, I have customized an Emacs front-end to do _exactly_ what I want, rather than relying on the command line tools directly. So I have a rather extreme perspective on this :). I do rely on the command line tools while learning a new CM system. In general, I expect seasoned users to be more accepting of the need to provide additional options to customize the tools to their workflow. -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: `git stash pop` UX Problem
Simon Ruderich si...@ruderich.org writes: On Mon, Feb 24, 2014 at 05:21:40PM +0100, Matthieu Moy wrote: One easy thing to do OTOH would be to show a hint at the end of git stash pop's output, like I think that's a good idea. It makes it obvious that Git has kept the stash and that the user should drop it when he's done - if he wants to. +1 This does not mean I don't _also_ think 'git add' dropping the stash when the last conflict is resolved is a good idea. If that is implemented, 'stash pop' might have to mention that effect as well; that does make things more complicated. -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: `git stash pop` UX Problem
Matthieu Moy matthieu@grenoble-inp.fr writes: Omar Othman omar.oth...@booking.com writes: Though I don't know why you think this is important: Now, the real question is: when would Git stop showing this advice. I don't see a real way to answer this, and I'd rather avoid doing just a guess. If it is really annoying for the user, we can just have a configuration parameter to switch this message on/off. Just saying You have X stash is OK to me as long as there is an option to deactivate it. +1 Hinting You should now run git stash drop. OTOH is far more dangerous if guessed wrong. Keeping a stash active when you don't need it does no real harm, but droping one you actually needed is data loss. I agree giving possibly incorrect advice is bad. Can you construct a use case where git will give incorrect advice? I don't know git well enough to do that, nor to assert that it will never happen. I think we need a more concrete proposal to move this forward. -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: `git stash pop` UX Problem
Matthieu Moy matthieu@grenoble-inp.fr writes: li...@haller-berlin.de (Stefan Haller) writes: Your intention was clearly to drop the stash, it just wasn't dropped because of the conflict. Dropping it automatically once the conflict is resolved would be nice. Your intention when you ran git stash pop, yes. Your intention when you ran git add, I call that guessing. You might be adding other files for other reasons. But if you add a file that does resolve a conflict caused by 'git stash pop', it is not guessing. The condition for dropping the stash should be more conflits resolutions are done AND the user is happy with it. Otherwise, if you mess up your conflict resolution, and notice it after running git add, then you're screwed because Git just happily discarded your important data. The point of keeping the stash is to leave it up to the user to decide between I'm happy, I can drop or I'm not, I should re-apply, and Git cannot tell which is which. Yes, that makes sense. Hinting the user to run stash pop would be more acceptable, but talking about git stash in git add's code is somehow a dependency order violation (stash is normally implemented on top of Git's basic features, not the other way around). Does not seem serious from at first from the user point of view, but this pushes the codebase one step in the direction of an unmaintainable mess. Also makes sense. So git add and git stash * are lower level tools; to get the effect we are asking for, we should use a front-end (which is why I'm writing one for Emacs :). -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: `git stash pop` UX Problem
Junio C Hamano gits...@pobox.com writes: ... So resolve the conflicts is assuming the intention of the user who issued pop too much (let alone manually---it does not matter how the user resolves conflicts---the only thing we want to say is Git did all it would and no further automated help in resolving is availble, but manually is not quite the word). +1 The stash was not dropped is the most important thing in your additional text. How about rephrasing like this? $ git stash pop Auto-merging foo.txt CONFLICT (content): Merge conflict in foo.txt The stashed change could not be replayed cleanly, leaving conflicts in the working tree. The stash was not dropped in case you need it again. After you are done with the stash, you may want to git stash drop to discard it. +1 -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: `git stash pop` UX Problem
Matthieu Moy matthieu@grenoble-inp.fr writes: Omar Othman omar.oth...@booking.com writes: [omar_othman main (trunk|MERGING*)]$ git add path/to/file.txt [omar_othman main (trunk*)]$ Note how the status message has changed to show that git is now happy. It is at that moment that the stash reference should be dropped Dropping the stash on a git add operation would be really, really weird... Why? That is when the merge conflicts are resolved, which is what logically indicates that the stash is no longer needed, _if_ the merge conflicts are related to the stash, which is true in this use case. There are other uses for 'git add' that don't indicate that; we'd have to be very careful to not throw away the stash at the wrong time. (or the user (somehow) is notified to do that herself if desired), because this means that the popping operation has succeeded. But how would you expect to be notified? When 'git add' checks to see if all merge conflicts are now resolved, and those merge conflicts were related to the stash, it can either pop the stash, or issue a message telling the user it is now safe to do so. We would need a config setting to indicate which to do. Maybe that check is hard to do in general? -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: `git stash pop` UX Problem
Junio C Hamano gits...@pobox.com writes: status is about reminding the user what changes are already in the index (i.e. what you would commit) and what changes are in the working tree, from which you could further update the index with (i.e. what you could commit). I believe status should tell me everything git knows about the current workspace in a resonably concise way. That includes the stash. One _could_ argue that stashed changes are what could be reflected to the working tree and form the source of the latter, but my gut feeling is that it is a rather weak argument. At that point you are talking about what you could potentially change in the working tree, No, I saved things in the stash on purpose. For example, I had changes that were not ready to commit, but I wanted to do a merge from upstream. There are workflows where the stash is not important; provide an option to 'git status' that means ignore stash. So, I tend to agree with you, while I do understand where I want to know about what is in stash is coming from (and that is why we do have git stash list command). My Emacs front end currently checks both 'git status' and 'git stash list' to build the status of the current workspace. -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Determining update/merge/current state of a workspace
I'm working on the DVC Emacs front-end for git (http://www.emacswiki.org/emacs/DistributedVersionControl), adding features similar to the ones I added for monotone (http://www.monotone.ca). I'm used to monotone and new to git, so this may seem like an odd workflow. I always do 'fetch' and 'merge' separately, never 'pull'. So after a 'fetch', the DVC Emacs front end must determine what needs to happen next. I think there are three cases: 1) 'fetch' did not retrieve any revisions from remote; the last local commit is the head of the branch. The workspace is up to date (it may need to be comitted). 2) 'fetch' retrieved revisions, and there were no local commits since the previous fetch. The last fetch is the head of the branch; if not equal to HEAD, the workspace needs to be updated (via 'merge'). 3) fetch retrieved revisions, and there were local commits since the previous fetch. There are two heads for the branch (the two described above), they need to be merged, then the workspace updated. I'm not sure how 'git fetch' handles case 3); I have not tested that case yet. The question I have is: What git queries can I run to determine which of the three states the current workspace is in? 'rev-parse HEAD' gives the last workspace commit. 'rev-parse refs/remotes/remote/branch' gives the head of the branch in the remote repository as of the most recent fetch. But to distinguish among the cases, I need to determine if one of these two revs is a child of the other or not. I don't see a git query to determine that directly. I could try parsing a 'log' output; I have not investigated that. This is easy in monotone; there is a command 'mtn heads' that gives this result directly (it returns either one or two revs), and another command 'mtn automate toposort' that orders revs topologically (by parent/child relationships). -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: aborted 'git fetch' leaves workspace unusable
Junio C Hamano gits...@pobox.com writes: stephen_le...@stephe-leake.org writes: However, in this case, even running the fetch was a mistake; I would have prefered that it leave FETCH_HEAD in its previous state. I think the clearing of leftover FETCH_HEAD is one of the early things git fetch does, unless --append is in effect. I haven't looked at the code for a long time, but it may be possible to move the logic of doing so around so that this clearing is done as lazily as possible. I however suspect that such a change may have fallouts on other people who are writing tools like yours; they may be depending on seeing FETCH_HEAD cleared after a failed fetch, and be surprised to see a stale contents after they (attempt to) run git fetch in it. So it is not so clear if it is a good thing to change the behaviour of git fetch not to touch FETCH_HEAD upon a failure. Ok; backwards compatibility is important. Perhaps FETCH_HEAD could be copied to FETCH_HEAD_prev or some such, to allow recovering in an error case? -- -- Stephe -- To unsubscribe from this list: send the line unsubscribe git in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html