Re: [RFC PATCH] rev-list: clarify --abbrev and --abbrev-commit usage

From: Jeff King <peff@peff.net>
Date: Jun 14, 2019, 16:18 UTC
Message-ID: <20190614161841.GB30083@sigill.intra.peff.net>
In-Reply-To: <20190613221541.10007-1-emilyshaffer@google.com>

On Thu, Jun 13, 2019 at 03:15:41PM -0700, Emily Shaffer wrote:

Show 15 quoted lines

> I thought this was odd when I was working on the other rev-list changes
> - --abbrev doesn't do anything on its own. It looks like it does work by
> itself in other commands, but apparently not in rev-list.
> 
> Listed this patch as RFC because maybe instead it's better to fix
> something so --abbrev can be used alone, or teach --abbrev-commit=<n>.
> It looks like `git log --abbrev=5` also doesn't work the way one might
> expect, which makes sense to me, as they use the same internals for
> option parsing (parse_revisions()).
> 
> The manpages for log and rev-list both correctly indicate that
> --abbrev=<n> is an optional addition to --abbrev-commit. `git log -h` is
> generated by parse-options tooling and doesn't cover --abbrev-commit at
> all, but `git rev-list` doesn't use an option parser on its own and the
> usage is hardcoded.

Yeah, "--abbrev" is a bit tricky here. It is really "when you abbrev, do it to this level". In "log", that means that "git log --abbrev=5 --raw" _does_ do something useful (it abbreviates the blob hashes). And then you may add "--abbrev-commit" on top to ask to abbreviate the commit ids.

And rev-list follows that same pattern, except that rev-list _never_ shows diff output. You'd traditionally do (and this is how log was implemented once upon a time):

  git rev-list HEAD | git diff-tree --stdin --abbrev=5 --raw

But even there, we are not seeing the commit id output by rev-list. It goes to diff-tree, which then formats it separately. So if you wanted abbreviated commits there, you'd add "--abbrev-commit" to the diff-tree invocation, not rev-list!

So no, I cannot see a way in which "rev-list --abbrev" is useful without "--abbrev-commit". Which means that perhaps the former should imply the latter.

And as you noticed in your other patch, there is no way to abbreviate "--objects" output at all. I am not sure I have ever seen a good use for that. Though to be honest, I am not sure that "--abbrev" is really all that useful in the first place. Machine-readable output should never abbreviate, and human-readable ones generally already do.

But at any rate, before making any behavior changes it may make sense to think about how they'd interact with "rev-list --objects" abbreviation, if it were to be added.

As for the patch itself:

Show 12 quoted lines

> diff --git a/builtin/rev-list.c b/builtin/rev-list.c
> index 9f31837d30..6ae0087b01 100644
> --- a/builtin/rev-list.c
> +++ b/builtin/rev-list.c
> @@ -49,8 +49,8 @@ static const char rev_list_usage[] =
>  "    --objects | --objects-edge\n"
>  "    --unpacked\n"
>  "    --header | --pretty\n"
> -"    --abbrev=<n> | --no-abbrev\n"
> -"    --abbrev-commit\n"
> +"    --abbrev-commit [--abbrev=<n>]\n"
> +"    --no-abbrev\n"

So --no-abbrev clears both --abbrev and --abbrev-commit. That sort of makes sense, though I did not expect it. But it also means that:

  --abbrev-commit [--abbrev=<n> | --no-abbrev]

is not right. Possibly:

  --abbrev-commit [--abbrev=<n>] | --no-abbrev

would show the interaction more clearly, but I don't have a strong opinion.

-Peff