Re: [PATCH 8/8] t0012: test "-h" with builtins

From: Junio C Hamano <gitster@pobox.com>
Date: Jun 1, 2017, 04:17 UTC
Message-ID: <xmqqinkg734c.fsf@gitster.mtv.corp.google.com>
In-Reply-To: <20170530152756.vs777v6unaxg6otb@sigill.intra.peff.net>

Jeff King <peff@peff.net> writes:

Show 7 quoted lines

> Anyway, the problem is sk/dash-is-previous, specifically fc5684b47
> (revision.c: args starting with "-" might be a revision, 2017-02-25). It
> looks like the revision parser used to just bail on "-h", because
> revision.c would say "I don't recognize this" and then cmd_rev_list()
> would similarly say "I don't recognize this" and call usage(). But now
> we actually try to read it as a ref, which obviously requires being
> inside a repository.

Heh, I found another ;-)

95e98cd9 ("revision.c: use refs_for_each*() instead of for_each_*_submodule()", 2017-04-19), which is in the middle of Duy's nd/prune-in-worktree series, does this:

#0  die (err=0x6128f8 "BUG: setup_git_env called without repository") at usage.c:114
#1  0x00000000004f9467 in setup_git_env () at environment.c:172
#2  0x00000000004f966c in get_git_dir () at environment.c:214
#3  0x000000000055113b in get_main_ref_store () at refs.c:1544
#4  0x0000000000570ee0 in handle_revision_pseudo_opt (submodule=0x0, 
    revs=0x7fffffffd6a0, argc=1, argv=0x7fffffffe180, flags=0x7fffffffc59c)
    at revision.c:2110
#5  0x00000000005716f5 in setup_revisions (argc=2, argv=0x7fffffffe178, 
    revs=0x7fffffffd6a0, opt=0x0) at revision.c:2254
#6  0x000000000043074a in cmd_diff_files (argc=2, argv=0x7fffffffe178, prefix=0x0)
    at builtin/diff-files.c:29
#7  0x0000000000405907 in run_builtin (p=0x87ba00 <commands+672>, argc=2, 
    argv=0x7fffffffe178) at git.c:376
#8  0x0000000000405bb5 in handle_builtin (argc=2, argv=0x7fffffffe178) at git.c:584
#9  0x0000000000405e04 in cmd_main (argc=2, argv=0x7fffffffe178) at git.c:683
#10 0x00000000004a3364 in main (argc=2, argv=0x7fffffffe178) at common-main.c:43

when jk/consistent-h is merged into it and then "git diff-files -h" is run.

I guess anything that calls setup_revisions() from the "git cmd -h" bypass need to be prepared with that

  check_help_option(argc, argv, usage, options);

thing. Which is a bit sad, but I tend to agree with you that restructuring to make usage[] of everybody available to git.c is probably too noisy for the benefit it would give us.