Need to change the method for determining the number of CPUs on Linux

[mcatanzaro]

48c72c6

Need to change the method for determining the number of CPUs on Linux
https://bugs.webkit.org/show_bug.cgi?id=273675

Reviewed by NOBODY (OOPS!).

nproc is unreliable, so remove it and rely on the fallback that greps
/proc/cpuinfo instead. This looks a little bit fragile (e.g. it would
fail if my AMD Ryzen "Processor" were instead an AMD Ryzen "processor")
but surely an improvement.

* Tools/Scripts/webkitdirs.pm:
(determineNumberOfCPUs):

48c72c6

Misc	iOS, tvOS & watchOS	macOS	Linux	Windows
✅ 🧪 style	✅ 🛠 ios	✅ 🛠 mac	✅ 🛠 wpe	✅ 🛠 wincairo
✅ 🧪 bindings	✅ 🛠 ios-sim	✅ 🛠 mac-AS-debug	✅ 🧪 wpe-wk2
✅ 🧪 webkitperl	✅ 🧪 ios-wk2	✅ 🧪 api-mac	✅ 🧪 api-wpe
	✅ 🧪 ios-wk2-wpt	✅ 🧪 mac-wk1	✅ 🛠 wpe-skia
✅ 🛠 🧪 jsc	✅ 🧪 api-ios	✅ 🧪 mac-wk2	✅ 🛠 gtk
✅ 🛠 🧪 jsc-arm64	✅ 🛠 tv	✅ 🧪 mac-AS-debug-wk2	✅ 🧪 gtk-wk2
	✅ 🛠 tv-sim	✅ 🧪 mac-wk2-stress	✅ 🧪 api-gtk
	✅ 🛠 watch		❌ 🛠 jsc-armv7
	✅ 🛠 watch-sim		❌ 🧪 jsc-armv7-tests

[webkit-early-warning-system]

EWS run on current version of this PR (hash 48c72c6)

Misc	iOS, tvOS & watchOS	macOS	Linux	Windows
✅ 🧪 style	✅ 🛠 ios	✅ 🛠 mac	✅ 🛠 wpe	✅ 🛠 wincairo
✅ 🧪 bindings	✅ 🛠 ios-sim	✅ 🛠 mac-AS-debug	✅ 🧪 wpe-wk2
✅ 🧪 webkitperl	✅ 🧪 ios-wk2	✅ 🧪 api-mac	✅ 🧪 api-wpe
	✅ 🧪 ios-wk2-wpt	✅ 🧪 mac-wk1	✅ 🛠 wpe-skia
✅ 🛠 🧪 jsc	✅ 🧪 api-ios	✅ 🧪 mac-wk2	✅ 🛠 gtk
✅ 🛠 🧪 jsc-arm64	✅ 🛠 tv	✅ 🧪 mac-AS-debug-wk2	✅ 🧪 gtk-wk2
	✅ 🛠 tv-sim	✅ 🧪 mac-wk2-stress	✅ 🧪 api-gtk
	✅ 🛠 watch		❌ 🛠 jsc-armv7
	✅ 🛠 watch-sim		❌ 🧪 jsc-armv7-tests

[aperezdc]

How is nproc unreliable? I felt tempted to suggest getconf _NPROCESSORS_ONLN but the getconf utility might not be present (it comes with glibc, and not all Linux systems are on glibc)... How about lscpu, which comes with util-linux (expected to be installed in all Linux systems) and has a parseable output mode? The best I can come up using that is this:

lscpu -b --parse=cpu | awk '/^[^#]/{ ncpus += 1 } END { print ncpus }

Surely counting lines that do not start with a # can be done in Perl instead of also shelling out to Awk.

[mcatanzaro]

How is nproc unreliable?

Guess: it's counting deactivated core complexes? I don't know, but I don't see any reason to second guess the bug reporter.

lscpu -b --parse=cpu | awk '/^[^#]/{ ncpus += 1 } END { print ncpus }

We could do this, but it's nice to depend on as few external programs as possible, and this is already running in a really niche case: it's only used if you don't have ninja installed.

[aperezdc]

I still don't like this. And it is not really needed to shell out to nproc, or even to grep... remember that Perl is supposed to be a glorified grep. So I would instead suggest:

Suggested change

	# Don't use nproc because it returns unreliable results, https://webkit.org/b/273675
	$numberOfCPUs = (grep /processor/, `cat /proc/cpuinfo`);
	use POSIX;
	$numberOfCPUs = POSIX::sysconf(83); # _SC_NPROCESSORS_ONLN = 83
	if ($numberOfCPUs == "") {
	open CPUINFO, "/proc/cpuinfo";
	$numberOfCPUs = 0;
	while (<CPUINFO>) {
	if (/^[Pp]rocessor\s/) { $numberOfCPUs++; }
	}
	close CPUINFO;
	}

(Yes, I tested both the sysconf code path and the fallback to parsing /proc/cpuinfo here.)

[mcatanzaro]

OK, but now you've written 100% of the code in this pull request, so we might as well close this one and let you open a new one for me to review?

[aperezdc]

Done: #28645

[aperezdc]

Closing in favor of #28645