On Mon, Jul 28, 2025 at 11:33 AM Junio C Hamano [off-list ref] wrote:

"D. Ben Knoble" [off-list ref] writes:

quoted

    I originally considered leaving out the changes to
    show_usage_with_options_if_asked and relying on the parse-options API to
    do the right thing. Unfortunately, most commands can't make it all the
    way to parse-options when setting up gently, and trying to parse options
    without a repo creates myriad dependency problems (like: we might read
    config after parsing CLI options, so we have to make sure the parsed
    options overrride config).

    Some usage.c callers, like check-ref-format, probably deserve to be
    ported to parse-options at this point.

It is unclear what you mean by all of the above, but hopefully it
would become clear as we read the code changes.

[snip]

I don't understand your "I originally considered leaving out ..." at
all.  We are special casing a lone "-h" here already because we know
this is where we should stop without exercising unnecessary code
because we may not even have repo!=NULL and that is the reason why
this function exists in the first place.  It is obvious to me that
we need the same special casing for "--help-all".

In other words, I think all the above changes are good.

Thanks.

I'm glad it was obvious to you! I spent a lot of time trying to get
commands to call parse_options early, before touching any repository
variables, since that was the only place where I knew "--help-all" was
handled. This turned out to be a terrible mess, so I looked around for
a different path forward (the one here now), and that turned out to
work much better.

In that old world, parse-options doesn't change and the builtins need
a lot of surgery (and probably still aren't completely right)—I can
see how that note is confusing without that context, though, and I'll
drop it in future rounds (esp. now we've discussed this version).