Re: [PATCH] diff-tree: do not show the sha1 of the given head with --quiet

From: Junio C Hamano <gitster@pobox.com>
Date: Jul 23, 2015, 17:06 UTC
Message-ID: <xmqq615aregb.fsf@gitster.dls.corp.google.com>
In-Reply-To: <xmqqegjyrfrq.fsf@gitster.dls.corp.google.com>

Junio C Hamano <gitster@pobox.com> writes:

Show 16 quoted lines

> Sebastian Schuberth <sschuberth@gmail.com> writes:
>
>> Well, from a user's perspective it does not matter which part of the
>> internal implementation of diff-tree is responsible for printing that
>> single line,...
>
> That is not "internal implementation", but "logically separate
> parts".  View it more like "'git show -s' does squelch the diff part
> but does not squelch the log output".  After all, a single commit form
> of 'diff-tree' is a degenerate use case of feeding a single commit
> to 'diff-tree --stdin' from its standard input, which is a rough
> plumbing-level equivalent of 'show'.
>
> Documenting the behaviour correctly is the best thing you could do
> at this point, as this is one of the oldest part of the system that
> existing scripts would rely on.

Having said that.

Existing scripts by definition would not be using a new option you will invent that used not to be a valid one. So that would be one way that you can shorten your script without breaking other people.

If we were living in an ideal world equipped with a time machine, I would redesign "git diff-tree $commit" so that it does not show the commit object name in its output at all, with or without "--quiet".

In "git rev-list ... | git diff-tree --stdin" output, the commit object name is absolutely necessary, with or without --quiet, as it serves as the sign that the output switched to talk about a different commit. But the case that feeds a single commit to the command, used as a poor-man's "git show $commit", does not need one---the caller knows exactly which commit the output is about. It is an unfortunate historical accident that a single commit usage is defined to be a degenerate case of feeding a sequence of commits to the command and the length of the sequence happens to be one.

But we do not live in an ideal world.