> Le 22 oct. 2025 à 04:37, Ruoyu Zhong via GitGitGadget <[email protected]> a écrit :
> 
> From: Ruoyu Zhong <[email protected]>
> 
> As documented in git-bisect(1), `git bisect help` should display usage
> information. However, since the migration of `git bisect` to a full
> builtin command in 73fce29427 (Turn `git bisect` into a full built-in,
> 2022-11-10), this behavior was broken. Running `git bisect help` would,
> instead of showing usage, either fail silently if already in a bisect
> session, or otherwise trigger an interactive autostart prompt asking "Do
> you want me to do it for you [Y/n]?".

Good catch!

FWIW, in this project we describe the buggy behavior in the present tense (« is broken », « Running git bisect shows », etc.)

> 
> Similarly, since df63421be9 (bisect--helper: handle states directly,
> 2022-11-10), running invalid subcommands like `git bisect foobar` also
> led to the same behavior.
> 
> This occurred because `help` and other unrecognized subcommands were
> being unconditionally passed to `bisect_state`, which then called
> `bisect_autostart`, triggering the interactive prompt.
> 
> Fix this by:
> 1. Adding explicit handling for the `help` subcommand to show usage;
> 2. Validating that unrecognized commands are actually valid state
>   commands before calling `bisect_state`;
> 3. Showing an error with usage for truly invalid commands.
> 
> This ensures that `git bisect help` displays the usage as documented,
> and invalid commands fail cleanly without entering interactive mode.
> Alternate terms are still handled correctly through
> `check_and_set_terms`.
> 
> Signed-off-by: Ruoyu Zhong <[email protected]>
> ---
>    bisect: fix handling of help and invalid subcommands
> 
> Published-As: https://github.com/gitgitgadget/git/releases/tag/pr-git-2078%2FZhongRuoyu%2Fgit-bisect-subcommands-v1
> Fetch-It-Via: git fetch https://github.com/gitgitgadget/git pr-git-2078/ZhongRuoyu/git-bisect-subcommands-v1
> Pull-Request: https://github.com/git/git/pull/2078
> 
> builtin/bisect.c | 6 +++++-
> 1 file changed, 5 insertions(+), 1 deletion(-)
> 
> diff --git a/builtin/bisect.c b/builtin/bisect.c
> index 8b8d870cd1..993caf545d 100644
> --- a/builtin/bisect.c
> +++ b/builtin/bisect.c
> @@ -1453,9 +1453,13 @@ int cmd_bisect(int argc,
>        if (!argc)
>            usage_msg_opt(_("need a command"), git_bisect_usage, options);
> 
> +        if (!strcmp(argv[0], "help"))
> +            usage_with_options(git_bisect_usage, options);
> +

From an extremely quick look at the code, this might be better handled with a new OPT_SUBCOMMAND, though that might mean making the options array statically scoped to this file rather than the function.

It would also be nice to update the usage to match the manual while we’re here, which presumably in turn affects the test between command usage and manuals.

>        set_terms(&terms, "bad", "good");
>        get_terms(&terms);
> -        if (check_and_set_terms(&terms, argv[0]))
> +        if (check_and_set_terms(&terms, argv[0]) ||
> +            !one_of(argv[0], terms.term_good, terms.term_bad, NULL))
>            usage_msg_optf(_("unknown command: '%s'"), git_bisect_usage,
>                       options, argv[0]);
>        res = bisect_state(&terms, argc, argv);
> 
> base-commit: 81f86aacc4eb74cdb9c2c8082d36d2070c666045
> --
> gitgitgadget

I think this part is OK, since we only intend to check this when using the « git bisect <term> » form. 

"Ruoyu Zhong via GitGitGadget" <[email protected]> writes:

> From: Ruoyu Zhong <[email protected]>
>
> As documented in git-bisect(1), `git bisect help` should display usage
> information. However, since the migration of `git bisect` to a full
> builtin command in 73fce29427 (Turn `git bisect` into a full built-in,
> 2022-11-10), this behavior was broken. Running `git bisect help` would,
> instead of showing usage, either fail silently if already in a bisect
> session, or otherwise trigger an interactive autostart prompt asking "Do
> you want me to do it for you [Y/n]?".
>
> Similarly, since df63421be9 (bisect--helper: handle states directly,
> 2022-11-10), running invalid subcommands like `git bisect foobar` also
> led to the same behavior.
>
> This occurred because `help` and other unrecognized subcommands were
> being unconditionally passed to `bisect_state`, which then called
> `bisect_autostart`, triggering the interactive prompt.

Very good observations.

> diff --git a/builtin/bisect.c b/builtin/bisect.c
> index 8b8d870cd1..993caf545d 100644
> --- a/builtin/bisect.c
> +++ b/builtin/bisect.c
> @@ -1453,9 +1453,13 @@ int cmd_bisect(int argc,
>  		if (!argc)
>  			usage_msg_opt(_("need a command"), git_bisect_usage, options);
>  
> +		if (!strcmp(argv[0], "help"))
> +			usage_with_options(git_bisect_usage, options);

I briefly wondered why

    $ git grep 'strcmp.*"help"'

gives a single hit in parse-options.c but that is simply because
"git bisect" is an oddball.  Everybody, including "git bisect"
itself, takes "git <cmd> --help", but in addition, "git bisect" also
takes "git bisect help".  So this addition is unfortunate but cannot
be helped (no pun intended).  In an ideal world, as we support the
bog standard "--help", if we could remove "help" as a subcommand,
then the other part of this patch would take care of "bisect help"
without this part by saying "unknown command" ;-) but we cannot
retroactively do so now.

I also wonder if it would be cleaner to add "help" as a genuine
subcommand to the options[] table just like all the other
subcommands.  The above would not be needed if we did so.  But I do
not see huge upside for doing so (i.e., a single strcmp() with a
call like we see above, vs. a new helper function to make the same
call, to usage_with_options()), so what is written in this patch is
perfectly fine.

>  		set_terms(&terms, "bad", "good");
>  		get_terms(&terms);
> -		if (check_and_set_terms(&terms, argv[0]))
> +		if (check_and_set_terms(&terms, argv[0]) ||
> +		    !one_of(argv[0], terms.term_good, terms.term_bad, NULL))
>  			usage_msg_optf(_("unknown command: '%s'"), git_bisect_usage,
>  				       options, argv[0]);

This change is a bit hard to reason about, so let me think aloud.

If we were saying "git bisect olde" after somehow changing bad/good
to newe/olde, then we do not want to say "unknown command", and the
way it avoids that is to see if the given command is one of the
terms check_and_set_terms() have updated.  

In other words, if argv[0] caused check_and_set_terms() to return
non-zero, we do not have to do "unknown command", because the helper
would have issued a warning already.  When it returns 0, it may be
because it saw a valid command "skip", etc. that cannot be a custom
good/bad/new/old (in which case our one_of() would be false and we
end up saying "unknown command"---have I just spot a bug???), or we
haven't set custom terms and argv[0] is one of bad/good/new/old (in
which case our one_of() would be true and we avoid saying "unknown
command").

And it turns out that my finding about the 'skip' etc. is not a bug,
as these valid commands that cannot be good/bad/new/old aliases are
already caught by parse_options() call and we know argv[0] is not
such a valid subcommand.

So after all this looks good.

Thanks.  Will queue.

Hi Ben,


Thanks for the review!

> On Oct 23, 2025, at 1:52 AM, Ben Knoble <[email protected]> wrote:
> 
> Good catch!
> 
> FWIW, in this project we describe the buggy behavior in the present tense (« is broken », « Running git bisect shows », etc.)

Thanks! Will keep this in mind.

>> diff --git a/builtin/bisect.c b/builtin/bisect.c
>> index 8b8d870cd1..993caf545d 100644
>> --- a/builtin/bisect.c
>> +++ b/builtin/bisect.c
>> @@ -1453,9 +1453,13 @@ int cmd_bisect(int argc,
>>       if (!argc)
>>           usage_msg_opt(_("need a command"), git_bisect_usage, options);
>> 
>> +        if (!strcmp(argv[0], "help"))
>> +            usage_with_options(git_bisect_usage, options);
>> +
> 
> From an extremely quick look at the code, this might be better handled with a new OPT_SUBCOMMAND, though that might mean making the options array statically scoped to this file rather than the function.

I intended to keep it simple so I did not make it an OPT_SUBCOMMAND at the first
place. Given that Junio is okay with it, I'm going to keep it as is for now.
Still happy to turn this into an OPT_SUBCOMMAND if you would like.

> It would also be nice to update the usage to match the manual while we’re here, which presumably in turn affects the test between command usage and manuals.


Thanks for pointing that out! Yes, I think so too. Will do in a separate patch,
if you agree, in order not to digress too much.


Regards,
Ruoyu

Hi Junio,


I appreciate your detailed and thoughtful analysis and I resonate with your
points. Specifically:

> I also wonder if it would be cleaner to add "help" as a genuine
> subcommand to the options[] table just like all the other
> subcommands.  The above would not be needed if we did so.  But I do
> not see huge upside for doing so (i.e., a single strcmp() with a
> call like we see above, vs. a new helper function to make the same
> call, to usage_with_options()), so what is written in this patch is
> perfectly fine.

Agreed. Will keep this as is, as I also mentioned in my response to Ben.

>> set_terms(&terms, "bad", "good");
>> get_terms(&terms);
>> - if (check_and_set_terms(&terms, argv[0]))
>> + if (check_and_set_terms(&terms, argv[0]) ||
>> +    !one_of(argv[0], terms.term_good, terms.term_bad, NULL))
>> usage_msg_optf(_("unknown command: '%s'"), git_bisect_usage,
>>       options, argv[0]);
> 
> This change is a bit hard to reason about, so let me think aloud.
> 
> If we were saying "git bisect olde" after somehow changing bad/good
> to newe/olde, then we do not want to say "unknown command", and the
> way it avoids that is to see if the given command is one of the
> terms check_and_set_terms() have updated.  
> 
> In other words, if argv[0] caused check_and_set_terms() to return
> non-zero, we do not have to do "unknown command", because the helper
> would have issued a warning already.  When it returns 0, it may be
> because it saw a valid command "skip", etc. that cannot be a custom
> good/bad/new/old (in which case our one_of() would be false and we
> end up saying "unknown command"---have I just spot a bug???), or we
> haven't set custom terms and argv[0] is one of bad/good/new/old (in
> which case our one_of() would be true and we avoid saying "unknown
> command").
> 
> And it turns out that my finding about the 'skip' etc. is not a bug,
> as these valid commands that cannot be good/bad/new/old aliases are
> already caught by parse_options() call and we know argv[0] is not
> such a valid subcommand.
> 
> So after all this looks good.

Thank you for walking through this. This is very helpful and also matches my own
thought process when I first looked at it. The key is that this part of the code
is only reached if fn is still NULL after parse_options, i.e., argv[0] is not
one of the known subcommands.


Regards,
Ruoyu