Re: [PATCH v2 1/3] chainlint: sidestep impoverished macOS "terminfo"

[Eric Sunshine <sunshine@sunshineco.com>]

On Fri, Nov 11, 2022 at 10:02 AM Ævar Arnfjörð Bjarmason
<avarab@gmail.com> wrote:
> On Fri, Nov 11 2022, Eric Sunshine via GitGitGadget wrote:
> > Sidestep this Apple problem by imbuing get_colors() with specific
> > knowledge of capabilities common to "xterm" and "nsterm", rather than
> > trusting "terminfo" to report them correctly. Although hard-coding such
> > knowledge is ugly, "xterm" support is nearly ubiquitous these days, and
> > Git itself sets precedence by assuming support for ANSI color codes. For
> > other terminal types, fall back to querying "terminfo" via `tput` as
> > usual.
>
> Doesn't test-lib.sh have the same problem then?

Generally speaking, yes, but in practice, no, not for this particular
case. I specifically wanted to use "dim" here in chainlint.pl, which
is (oddly) missing in Apple's terminfo. test-lib.sh just uses some
colors, bold, reverse, and "reset", all of which are present in
Apple's terminfo.

> This is somewhat of an aside, as we're hardcoding thees colors in
> t/chainlint.pl now, but I wondered when that was added (but I don't
> think I commented then) why it needed to be re-hardcoding the coloring
> we've got in test-lib.sh.
>
> I.e. if test-lib.sh is running it could we handle these cases, and just
> export a variable with the color info for "bold" or whatever in
> GIT_TEST_COLOR_BOLD, then pick that up?

When adding colorizing to chainlint.pl, one of my very first thoughts
was to somehow reuse the color information from test-lib.sh. That's
relatively easy in the case when the test script is being run
standalone because it has already "sourced" test-lib.sh before it runs
chainlint.pl, thus could pass the color information along to
chainlint.pl somehow. But the other case, when "make test" (or "make
test-chainlint", etc.) is used is harder because that's just the
Makefile running chainlint.pl directly, so test-lib.sh isn't involved
in the equation. It would probably be possible to make it work, but
the solutions seemed ugly and too invasive, especially for an initial
implementation of colorizing in chainlint.pl.

> I have a local semi-related patch which made much the same change to
> test-lib.sh itself, to support --color without going through whether
> tput thinks we support colors:
> https://github.com/avar/git/commit/c4914db758b
>
> I think this is fine for now if you don't want to poke more at it, but
> maybe this should all be eventually combined?

Peff also expressed such a sentiment[1][2][3]. I'm still somewhat
hesitant to make chainlint.pl dependent upon so much outside machinery
since it's still nicely standalone and _might_ be useful elsewhere.

> I also wonder to what extent this needs to be re-inventing
> Term::ANSIColor, which has shipped with Perl since 5.6, so we can use it
> without worrying about version compat, but that's another topic...

Gah, why didn't I know about this sooner?! (Note to self: hit self
over head.) Back when I was adding colorizing to chainlint.pl, I
searched around to determine if Perl had any built-in colorizing
support, but I completely overlooked this. I must have missed it
because I was focusing on "curses" since I thought I would need to
pull color codes directly from "curses", and Perl doesn't have a
standard (shipped) "curses" module. I even said as much to Peff[4].

Since it's been shipping with Perl for quite some time,
Term::ANSIColor would be a much nicer solution; worth looking into.

[1]: https://lore.kernel.org/git/Yx%2FLpUglpjY5ZNas@coredump.intra.peff.net/
[2]: https://lore.kernel.org/git/Yx%2FeG5xJonNh7Dsz@coredump.intra.peff.net/
[3]: https://lore.kernel.org/git/Yx%2FPnWnkYAuWToiz@coredump.intra.peff.net/
[4]: https://lore.kernel.org/git/CAPig+cRJVn-mbA6-jOmNfDJtK_nX4ZTw+OcNShvvz8zcQYbCHQ@mail.gmail.com/