Re: [PATCH] trailer: allow spaces in tokens

From: Jonathan Tan <jonathantanmy@google.com>
To: Junio C Hamano <gitster@pobox.com>
Cc: Jonathan Tan <jonathantanmy@google.com>,
	Max Bernstein via GitGitGadget <gitgitgadget@gmail.com>,
	git@vger.kernel.org,
	Max Bernstein <donotemailthisaddress@bernsteinbear.com>,
	Max Bernstein <max@bernsteinbear.com>,
	Christian Couder <christian.couder@gmail.com>
Subject: Re: [PATCH] trailer: allow spaces in tokens
Date: Thu, 18 Aug 2022 10:52:16 -0700	[thread overview]
Message-ID: <20220818175216.3530525-1-jonathantanmy@google.com> (raw)
In-Reply-To: <xmqqo7whqyzs.fsf@gitster.g>

Junio C Hamano <gitster@pobox.com> writes:
> Jonathan, do you remember if the "be stricter" was in a response to
> specific real world issues, or is this "we no longer allow whitespace"
> an unintended side effect of the change?

Rereading the relevant mailing list thread, I think it's a bit of both.
The sequence of events was as follows:
 - I made a patch set that changed sequencer.c (which, among other
   things, controls the behavior of "commit -s") to use
   interpret-trailers's functionality. This caused a change in the
   behavior of a "commit -s" unit test (the "I want to mention about
   Signed-off-by: here." one), because "commit -s" previously did not
   support spaces before the separator. [1]
 - You mentioned that perhaps we should tighten it back. [2]
 - I agreed and tightened it, but didn't change the documentation. [3]

[1] https://lore.kernel.org/git/a416ab9b-ff1f-9a71-3e58-60fd4f8a6b8e@google.com/
[2] https://lore.kernel.org/git/xmqqtwbroykt.fsf@gitster.mtv.corp.google.com/
[3] https://lore.kernel.org/git/cover.1478028700.git.jonathantanmy@google.com/

So the "real world issue" was a test in our test suite, and perhaps the
side effect was unintended in that we wanted to preserve the behavior of
"commit -s" but didn't think about other possible uses of
interpret-trailers.

> If the former, an equally valid "fix" to what Max reports here would
> be to add such a real world issue or two as new tests to demonostrate
> why allowing whitespaces was a mistake that hurts real-world workflow,
> and fix the documentation.
>
> I actually am suspecting it is the latter, only because we would have
> added a testcase to show why whitespaces in the trailer label is a
> bad idea in e4319562 if this was triggered by a real-world issue.

The "I want to mention about Signed-off-by: here." test might be a
sufficient real-world issue.

> I would say that it would be a disaster, if we took any random
> line with colon : in it in the middle of the commit message and
> mistook it as a trailer (like the line above), but since we do not
> allow paragraph breaks in the trailer block, as long as the message
> has a valid trailer block, it might not be a huge issue.  I dunno.

Yes, it would be fine if the paragraph was two or more lines long, but
not if it's a single line.