Robert Haas <robertmhaas@gmail.com> writes:
> On Fri, Sep 26, 2025 at 11:33 AM Jacob Champion
> <jacob.champion@enterprisedb.com> wrote:
>> Agreed, it's pretty inefficient. (For comparison, it runs in about two
>> seconds on my machine, so I hadn't noticed how bad it had gotten.
>> Thanks for pointing it out.)
> Interesting. Are you on Linux? I'm testing directly on macOS, and I
> wonder now whether macOS is comparatively slower at starting new
> processes...
Something else going on, perhaps? With PROVE_FLAGS=--timer,
I see this on my RHEL8 workstation:
[11:55:32] t/001_test_json_parser_incremental.pl .. ok 475 ms ( 0.02 usr 0.00 sys + 0.26 cusr 0.25 csys = 0.53
CPU)
[11:55:32] t/002_inline.pl ........................ ok 2883 ms ( 0.11 usr 0.01 sys + 1.40 cusr 1.70 csys = 3.22
CPU)
[11:55:35] t/003_test_semantic.pl ................. ok 70 ms ( 0.00 usr 0.00 sys + 0.06 cusr 0.02 csys = 0.08
CPU)
[11:55:35] t/004_test_parser_perf.pl .............. ok 57 ms ( 0.00 usr 0.00 sys + 0.05 cusr 0.00 csys = 0.05
CPU)
[11:55:35]
All tests successful.
Files=4, Tests=3930, 3 wallclock secs ( 0.14 usr 0.01 sys + 1.77 cusr 1.97 csys = 3.89 CPU)
Result: PASS
and this on my MacBook Pro M4 (Sequoia 15.6.1):
[11:56:06] t/001_test_json_parser_incremental.pl .. ok 571 ms ( 0.01 usr 0.00 sys + 0.24 cusr 0.28 csys = 0.53
CPU)
[11:56:06] t/002_inline.pl ........................ ok 3544 ms ( 0.06 usr 0.02 sys + 1.43 cusr 1.86 csys = 3.37
CPU)
[11:56:10] t/003_test_semantic.pl ................. ok 54 ms ( 0.00 usr 0.00 sys + 0.03 cusr 0.02 csys = 0.05
CPU)
[11:56:10] t/004_test_parser_perf.pl .............. ok 39 ms ( 0.00 usr 0.00 sys + 0.03 cusr 0.01 csys = 0.04
CPU)
[11:56:10]
All tests successful.
Files=4, Tests=3930, 4 wallclock secs ( 0.08 usr 0.02 sys + 1.73 cusr 2.17 csys = 4.00 CPU)
Result: PASS
The M4 is about 2.3x the speed of the workstation according to
geekbench, so macOS is not covering itself with glory here,
but it's sure not 10x worse than Linux. Maybe you've got a
Perl installation with debugging turned on, or something?
Having said that, even three-ish seconds is an annoying
contribution to check-world, so let's see if there's not some
low-hanging fruit here.
regards, tom lane
