zend_hrtime: use CLOCK_MONOTONIC instead of CLOCK_MONOTONIC_RAW

[nicolas-grekas]

This should fix the hrtime.phpt failure reported in GH-22508.

CLOCK_MONOTONIC_RAW is not disciplined by NTP, so it ticks with the raw frequency error of the underlying oscillator. Under WSL2 here, that is 4% fast, measured over a 1s sleep:

             microtime delta   hrtime delta
8.5          1.000083 s        1.000086 s
8.6          1.000072 s        1.038898 s
8.6 + this   1.000158 s        1.000163 s

The relative uncertainty that ext/standard/tests/hrtime/hrtime.phpt computes goes from 0.038 back to 0.0002 with this, against the 0.05 the test allows. @mbeccati measured 0.0501 on Ubuntu 26.04 while packaging 8.6.0alpha1 and suggested raising that limit - I think the limit is fine and it is the clock that moved, but a confirmation on that box would be welcome.

The adjtime/NTP slew that GH-19221 wanted to avoid is bounded to 500ppm by the kernel, and it is precisely what makes CLOCK_MONOTONIC track elapsed real time, which is the contract that test asserts. CLOCK_MONOTONIC_RAW is also served by the vDSO only since Linux 5.3, so on eg. RHEL 8 it costs a syscall per call.

I kept the startup check from GH-19221, so that zend_hrtime() doesn't need to check what clock_gettime() returns. zend_hrtime_posix_clock_id stays for ABI compatibility even if it's now always CLOCK_MONOTONIC; removing it is for master.

[TimWolla]

@marc-mabe as the original author of #19221.

[TimWolla]

I kept the check-the-clock-once part

zend_hrtime_posix_clock_id is never reassigned though.

[nicolas-grekas]

Right, PR updated: the variable is gone. What I meant to keep is the startup check, which lets zend_hrtime() ignore the return value of clock_gettime().

I also retargeted to PHP-8.6: that's where the _RAW switch and the hrtime.phpt failure are, and dropping zend_hrtime_posix_clock_id is free only as long as no release exports it.

[mbeccati]

Even if useless, please keep ZEND_API clockid_t zend_hrtime_posix_clock_id = CLOCK_MONOTONIC; as removing it is considerent an ABI break, which isn't allowed at this stage of 8.6.

It can then be removed in master instead.

[nicolas-grekas]

PR updated, the variable is back. I'll send its removal to master once this is merged up.

[marc-mabe]

Thanks for tracking this down, and you're right. I read the man page's note that "´CLOCK_MONOTONIC_RAWisn't affected by NTP or adjtime" as a reason to prefer it. I thought NTP could makeCLOCK_MONOTONICjump during a measurement but it only nudges its rate slightly to correct the hardware clock's error while_RAW` keeps that error, and in VMs it's much larger.

Keeping the startup check is enough. LGTM, and +1 to dropping zend_hrtime_posix_clock_id in master.

[TimWolla]

I just made a final adjustment to the error message, merged the latest PHP 8.6 and added NEWS. Will merge when CI is green.