Re: [PATCH v3] rev-list: skip --grep argument in early scan
- From
- Tanga Joris <tangajoris@gmail.com>
- Date
- Nov 5, 2026, 07:53 UTC
- Message-ID
- <CAKrRMNrnd2+HxTxadKaha+K9ASWi_oN4h7gdjVYAGmyODwmi=w@mail.gmail.com>
- In-Reply-To
- <20261026083055.18873-1-azuama69@gmail.com>
Hello Azuama,
Thanks for working on this.
I applied v3 on top of master (0ffa3e35c7) with "git am -3", because builtin/rev-list.c has changed since you sent it, so running "git am" fails. Rebasing on master would help. I built it with NO_RUST=1. t6022 passes, and both new tests fail when I revert only the change in builtin/rev-list.c, so they do catch the problem.
On Mon, 26 Oct 2026 at 11:30, Azuama Odinaka Maximus <azuama69@gmail.com> wrote:
> +test_expect_success "--grep accepts a pattern beginning with --" ' > + test_commit --no-tag --missing-only && > + commit=$(git rev-parse HEAD) &&
These lines are indented with spaces instead of a tab. "git diff --check" reports "indent with spaces" for all the new test lines.
> + } else if (!strcmp(arg, "--grep")) {
> + i++;This fixes --grep, but other options that take a separate value still have the same problem. In a repository whose only commit has "--missing-only" as both its subject and its author name, with this patch:
$ git rev-list --grep --missing-only HEAD <the commit> $ git rev-list --author --missing-only HEAD fatal: --missing-only requires --missing=print or --missing=print-info
--committer and --grep-reflog fail the same way. In revision.c, 19 options are parsed with parse_long_opt(), which accepts the value as a separate argument (--author, --committer, --since, --until, --skip, --exclude and others). Should the early scan skip the value of all of them, or is there a reason to handle only --grep? If this only fixes --grep for now, the commit message could say so, because its first paragraph describes the general problem.
Also, the NEEDSWORK comment just above this loop uses "--grep --exclude-promisor-objects" as its example of what the loop gets wrong. With this patch that example works, so the comment needs a new example (--author would do), or it simply becomes misleading.
Thanks, Tanga