Re: [PATCH v2 10/10] detect-compiler: detect clang even if it found CUDA

[Eric Sunshine <sunshine@sunshineco.com>]

On Thu, Mar 27, 2025 at 1:14 AM Jeff King <peff@peff.net> wrote:
> On Wed, Mar 26, 2025 at 02:07:10PM -0400, Eric Sunshine wrote:
> > It's probably an indication that I've done too much `sed` programming,
> > but I find Dscho's version more obvious. That aside, your response
> > made me take a closer look at what Dscho wrote and I noticed that it
> > is syntactically flawed, at least for BSD-lineage `sed`. Testing on
> > macOS reveals that this is indeed so:
> >
> >     % LANG=C LC_ALL=C cc -v 2>&1 | sed -n '/ version /{p;q}'
> >     sed: 1: "/ version /{p;q}": extra characters at the end of q command
> >
> > The problem is that the `q` function takes no arguments, but
> > BSD-lineage `sed` thinks that the closing `}` is an argument rather
> > than a terminator. Fixing this requires inserting a terminator after
> > `q`, which will be either a newline character or a semicolon. So, the
> > correct form is:
> >
> >     sed -n '/ version /{p;q;}
>
> Heh, I think it was the braces and semicolons that made my spider-sense
> tingle, probably because I've been bitten by those subtleties in the
> past.
>
> I think just "/foo/p;q" works on GNU sed, but no idea if it does
> elsewhere. What you wrote seems the safest.

That's not quite the same, though. The patternless `q` will cause
`sed` to terminate upon reading the first line of input, not upon the
first line which contains " version ". This matters, for instance, if
the first line output by `$CC -v` is not the version string (i.e. it
might be a copyright notice).