Hi Taylor, On Wed, 13 Jan 2021 at 23:54, Taylor Blau <me@xxxxxxxxxxxx> wrote: > > Hi Chavri, > > On Wed, Jan 13, 2021 at 06:45:25AM +0530, Charvi Mendiratta wrote: > > Hi Everyone, > > > > Implementing "amend!" commit would be an alternative to the > > fixup!/squash! commit that addresses the issue as opened here[1]. Also > > the related patches[2], adds the options to `fixup` command in > > interactive rebase and supports the "amend!" commit upon > > `--autosquash`. Next, after discussing with Phillip and Christian, > > there could be 3 possibilities to implement the "amend!" commit (UI): > > To make sure that I'm understanding correctly: > > - Your series in [2] teaches the rebase directive 'fixup' two new > options: -C, and -c. Unlike of the option-less 'fixup' command, > these new options use the message from the fixup commit instead of > the original. > > If I'm understanding correctly, this old-style sequence: > > pick aaaaaaaaaa original > fixup bbbbbbbbbb fixup! new > exec git commit --amend -C bbbbbbbbbb --no-edit > > should be equivalent to this one: > > pick aaaaaaaaaa original > fixup -C bbbbbbbbbb fixup! new > Yes, `fixup -C ` works the same as above. > - Separate from that, you are asking about how to implement an > "amend!" commit which would behave exactly as the 'fixup -C' variant > that you are proposing. > > We should clarify what you mean by "implement". I take it from the > remainder of your message that you are really asking about how we should > _expose_ this new 'fixup -C' command to users. > I apologize for the confusion. But yes you got right, by using "implement" I mean to ask regarding the user interface for "amend!" commit. So, if we choose to go with adding additional arguments like , `git commit --fixup= --amend`, then it will create a new "amend!" commit and in sequencer it will insert fixup -C, upon `git rebase -i --autosquash`. > > Firstly, the `--fixup=<commit>` to have option like, > > `--fixup=reword/amend:<commit>` > > > > So, `git commit --fixup` can have 3 options: > > > > a) `--fixup=<commit>`, work as of now, make fixup! commit. > > b) `--fixup=amend:<commit>`, make "amend!" commit, takes changes and > > also opens the editor for a new message (Here it adds a new message to > > amend! commit's message body and upon autosquash it will fixup up the > > content and reword the original commit message i.e replaces the > > original commit message with the "amend!" commit's message). > > c) `--fixup=reword:<commit>`, makes (empty) "amend!" commit, do not > > take changes and open the editor for a new message(Here, upon > > autosquash it will just reword the original commit message). > > I think that this trio is a good path forward, but... > `--fixup` behaving as it always has, which is good. Then it makes > `--fixup=amend:<commit>` insert a 'fixup -C', and `--fixup=reword:<...>` > insert a 'fixup -c'. > > > Secondly, > > As an alternative to above, we can use `--fixup=<commit> --amend` and > > `--fixup=<commit> --reword`. > > ...I think that this option is even better. > > Here '--fixup=<commit>' with '--amend' inserts a 'fixup -C' sequencer > comamnd, and using the same instead with '--reword' inserts a 'fixup -c' > sequencer command. > I think this would be a bit different. Here both the `--fixup=<commit> --amend` and `--fixup=<commit> --reword` would create a new "amend!" commit and at the same time also opens the editor for a new commit message. And in the sequencer both insert "fixup -C". So, here for '--reword', 'fixup -c' is not used in sequencer upon --autosquash, as we need to open the editor only one time while `git commit --fixup=<commit> --reword` and works as an alternative to "squash!" commit that opens the editor both the time. And, the major difference is that " --fixup=<commit> --amend" would take both changes and also let the user change the commit message. On the other hand "--fixup=<commit> --reword" will not take any changes and only let the user to change the commit message. (So, we don't need to add additional --allow-empty as we do now). I also think that the above thing can be a bit confusing and I would be happy to do more improvements or suggestions to implement it in a more user friendly way. > This is clear to me because '--fixup' makes a commit that is interpreted > separately by the sequencer machinery, so any options that are given > with it seem to indicate that they will alter how the sequencer > interprets the resulting commit, which makes sense. > > Of course, it's all somewhat confusing because you have to keep track of > which are options to 'git commit', and which are sequencer commands, but > I like the direction that you're going in here. > > > Next, > > To use only, `--fixup=<commit> --edit` to make the "amend!" commit. > > I don't think that this is as good a direction forward. > I agree, thanks. Thanks and Regards, Charvi > > Also as discussed earlier[3] we are avoiding the use of additional > > options like `git commit --amend=<commit>` inorder to avoid confusion > > of doing similar things in different ways. So, I wonder which could be > > the best way to proceed with or if any other way to implement "amend!" > > commit ? > > > > Thanks and Regards, > > Charvi > > > > [1] https://github.com/gitgitgadget/git/issues/259 > > [2] https://lore.kernel.org/git/20210108092345.2178-1-charvi077@xxxxxxxxx/ > > [3] https://lore.kernel.org/git/95cc6fb2-d1bc-11de-febe-c2b5c78a6850@xxxxxxxxx/ > > Thanks, > Taylor
