On Thu, Mar 03 2022, Junio C Hamano wrote:
> Emily Shaffer <emilyshaffer@xxxxxxxxxx> writes:
>
>>> +The goal of your log message is to convey the _why_ behind your
>>> +change to help future developers.
>>> +
>>
>> This is pretty compelling. Is it clear enough why we care about this in
>> the commit message, as opposed to in the patch description (cover letter
>> or post-"---" blurb)? Is it too obvious to explicitly mention that the
>> commit message is the first thing we try to make sense of during a 'git
>> blame' or 'git bisect'?
>
> Again, patches welcome ;-)
I think for something that's a stylistic preference I'd see why Emily
would try to see how you feel about it first.
I.e. in this project it's ultimately up to you to decide on those
things, so for coding guidelines you've just updated I'd probably do the
same.
Don't take that as a complaint on the end result b.t.w., I think the
overall excellent state of the codebase in this project speaks for
itself on that front.
I just see why other contributors would be gun shy about pulling the
trigger on a patch submission on this particular front :)
> Having to say "this may be better in the in-code comment rather than
> the log message" to some patch recently (I do not remember), I tend
> to agree that some guidance would help people decide between the two
> (or writing both).
[A somewhat sneaky $subject change :)]
I think you're referring to this comment on (some admittedly tricky)
code I wrote[1].
First, given the above I'll adjust that to your preferences on a
re-submission. So don't take this as some argument on *that* specific
point. I'll add a comment in a re-roll.
But on reflection I still wouldn't put a comment on that code if it were
purely up to me. Why?
First, I think all programmers go through a phase of learning where they
feel more compelled to comment on the "how" v.s. "why" early on.
At the most extremes beginner programmers explaining how say a common
standard library function works in code that's relatively
straightforward.
But that's really not the case in [1], that code really is doing
something odd and worth explaining. So why not add a comment?
Because access to ubiquitous and *local* source control from git changed
a lot of people's habits on this front, including mine. For this code I
*would* definitely add a comment there if it was the pre-DVCS[2] days.
But I'd say that today my criteria for adding a comment is closer to:
Is this so essential to note for the understanding of the rest of
this code that nobody who's skimming past this line or reading this
part *wouldn't* want this information?
Or rather, they might not absolutely want to know, but it might be
useful, *and* there's nothing odd about the pattern itself that
makes you go "hrm?" enough to run "git blame/log" on it.
In this case it's clearly pretty weird that we run the exact "test-tool
regex" command twice. But I think that "hrm?" applies. To elaborate:
A. It's setting up a self-contained prereq, so it's not essential
to understand that implementation detail to read the rest of the
code.
B. Anyone who does want to see why that odd case is the way it is can
run some version "git blame" or say:
git log -p -L14,21:t/t7812-grep-icase-non-ascii.sh
C. Each comment you add, even within a function or other scope dilutes
the value of other more important comments. It trains people not to
read them, as they're probably not that important.
D. Comments that are "frozen in time" by adding them to the code
itself are almost always in danger of drifting in accuracy from the
rest of the implementation and its assumptions.
Even in well-curated codebases like git.git they're *much more*
likely to drift away from the "ground truth" than code is.
E. Even if "D" isn't true, commit messages (in this case my [3]) are
almost always at an advantage over comments in that they accompany
a change to the pre-image.
So e.g. that commit message doesn't need to waste time explaining
what pattern we'd prefer not to have here instead of the post-image,
you get that context for free.
Due to a combination of "D" and "E" I almost never read comments in
their current context, unless it's painfully obvious that they *must*
still apply to the current code. I'll usually run some variant of "git
blame" or say:
git log -p -L:relevant_function:file
And then see how the code looked when that comment was added, and page
through what's changed since then.
1. https://lore.kernel.org/git/xmqqsfryah42.fsf@gitster.g/
2. To those who'd nitpick DVCS v.s. VCS: Yes nothing changed in theory from CVS/SVN
etc. on this front, but in practice it did.
Access to version control didn't tend to be ubiquitous, and even if it was
accessible many people browsing or contributing to your code would probably
do so via a downloaded tarball than trying to get CVS or whatever to work.
So the "D" in DVCS really did change this.
3. https://lore.kernel.org/git/patch-12.15-f3cc5bc7eb9-20220302T171755Z-avarab@xxxxxxxxx/
