Christian Couder <christian.couder@xxxxxxxxx> 于2021年7月22日周四 下午3:50写道： > > On Mon, Jul 19, 2021 at 8:55 AM ZheNing Hu <adlternative@xxxxxxxxx> wrote: > > > > My ninth week blog finished: > > The web version is here: > > https://adlternative.github.io/GSOC-Git-Blog-9/ > > Thanks! > > > ## Week9 BUG REPORT > > > > ### BUG REPORT 1 > > First, it would be better if these bugs were submitted to the mailing > list separately, one in each own email with a relevant subject, so > that people can more easily find and discuss them separately. It's ok > then if you link to these bug reports from, or copy parts of them to, > a blog post. > You're right. Maybe the content of the blog did not attract everyone's attention. > > * What did you do before the bug happened? (Steps to reproduce your issue) > > > > Because someone told me that `git cherry-pick` can't gave useful > > prompt information like `git rebase -i` does: > > > > ``` > > You can amend the commit now, with > > > > git commit --amend > > > > Once you are satisfied with your changes, run > > > > git rebase --continue > > ``` > > > > I found that I can take use of "GIT_CHERRY_PICK_HELP" environment variable, > > > > ``` > > $ GIT_CHERRY_PICK_HELP="git cherry-pick --continue" ggg cherry-pick v1 > > Not sure what `ggg` is. Is it GitGitGadget, `git`, an alias with some > features turned on, or something else? > Oh, it's just git/bin-wrappers/git. > Not sure what is `v1` also. Is it a tag to a random commit? Yeah. > > > ``` > > > > which will output prompt information "git cherry-pick --continue", good! > > > > * What did you expect to happen? (Expected behavior) > > > > I could use `git cherry-pick --abort` to exit cherry-pick normally. > > > > * What happened instead? (Actual behavior) > > > > Then I couldn't use `git cherry-pick --abort` to exit cherry-pick normally. > > What happened when you tried to use it? Or what prevented you from using it? > > > * Anything else you want to add: > > > > See the print_advice() in sequencer.c, `CHERRY_PICK_HEAD` will be removed > > if we use the env "GIT_CHERRY_PICK_HELP". It is used by `git rebase > > -i` and somewhere else. > > So you got an error because `CHERRY_PICK_HEAD` had been removed? > Yeah, this is the situation I encountered. > > Here may have two solutions: > > 1. Prevent users from using the environment variable "GIT_CHERRY_PICK_HELP". > > This means that we should remove it, right? Otherwise what's the > purpose of keeping it if users are prevented from using it? Or maybe > there is something I don't understand? > Yeah, this may require rebase -i to use some new methods to achieve the original functions. > > 2. check if we are truly cherry-pick. > > > > ```c > > diff --git a/sequencer.c b/sequencer.c > > index 0bec01cf38..c01b0b9e9c 100644 > > --- a/sequencer.c > > +++ b/sequencer.c > > @@ -409,8 +409,9 @@ static void print_advice(struct repository *r, int > > show_hint, > > * (typically rebase --interactive) wants to take care > > * of the commit itself so remove CHERRY_PICK_HEAD > > */ > > - refs_delete_ref(get_main_ref_store(r), "", "CHERRY_PICK_HEAD", > > - NULL, 0); > > + if (opts->action != REPLAY_PICK) > > + refs_delete_ref(get_main_ref_store(r), "", "CHERRY_PICK_HEAD", > > + NULL, 0); > > return; > > } > > ``` > > So this solution means not removing CHERRY_PICK_HEAD if we are indeed > cherry-picking instead of rebasing. Yeah, this looks like a logical > solution to the issue to me. Could you send this in a separate patch > with a proper commit message? > I haven't carefully verified its correctness. but okay, I will send it separately. > > ### BUG REPORT 2 > > Thanks Atharva for replying to this bug report! > > > ### project progress > > > > I am still thinking about how to improve the performance of `git > > cat-file --batch`. > > This cannot be solved quickly, keep patient. > > Yeah, sure! In cases like this when the issue is not easy to > understand, it could be a good idea to keep some kind of research > journal where you describe the issue, the possible ways to solve it > and how your research to solve it goes. It could be a text or markdown > file in a branch, or a shared Google doc, for example. > Great, I happen to have the same idea as you. Some private attempts do not seem to really improve the performance of ref-filters, maybe I should show them to you. > > My mentors told me to split my main patch series into a few smaller > > patch series, but how? > > In other words, there is a certain correlation between these patches, > > If they are really split into > > multiple patches, how can I send them to the mailing list without repeating? > > Yeah, I am not sure if it's possible and easy to send a patch series > based upon another one when using GitGitGadget. But you could perhaps > still send the first patch series, and then focus on getting it merged > and on performance issues. Discussing performance issues can probably > be done by only sending diffs in regular emails or RFC patches that > cannot be applied to the mailing list until a proper solution is > found. Now I have an idea: The latter patch will not be sent to the mailing list first, and we can wait for the previous patch merge to master first. In the meantime, let us discuss performance solutions. > > I just received half of the GSoC bonus, and felt a burden of responsibility... > > No need to feel too much responsibility, we all share responsibility > in this. Also feel free to privately email me or Hariom and me (maybe > Kaartic too) if you want to discuss this further privately. I just think it seems that the "performance" problem seems to have not been solved well by Olga at the time, so I feel a certain amount of pressure. Thanks!!! -- ZheNing Hu
