dest.c: Use monotonic time for cups_enum_dest

[zdohnal]

Patch written by Adam Williamson.

gettimeofday() is vulnerable to clock jumps, so it is better to use clock_gettime() to avoid issues with libcups if system changes clock, f.e. in Live CDs.

[michaelrsweet]

I'm thinking we need a CUPS API that returns a monotonic value, probably a double for portability.

double cupsGetClock(void);

If we back-port this to CUPS 2.4.x, we'll want to leave this as private API (_cupsGetClock).

[michaelrsweet]

[master 0350eba] Add cupsGetClock API.

Also in libcups repository.

[zdohnal]

@michaelrsweet hmm, looks like Mac monotonic clock precision is buggy...

I've looked into Mac results on current master branch and there is something wrong with MacOS test suite - I see fail for cupsGetClock(), but the CI run shows as passed:

cupsGetClock(initial): PASS (0)
cupsGetClock(1 second): PASS (1.09793)
cupsGetClock(5 seconds): FAIL (got 5.14539, expected 5.0 +/- 0.1)
cupsGetClock(10 seconds): FAIL (got 10.2297, expected 10.0 +/- 0.1)
cupsGetClock(30 seconds): FAIL (got 30.3457, expected 30.0 +/- 0.1)
cupsGetClock(60 seconds): FAIL (got 60.3833, expected 60.0 +/- 0.1)
cupsGetClock(120 seconds): FAIL (got 120.406, expected 120.0 +/- 0.1)

I will try the CLOCK_MONOTONIC_RAW, if it helps MacOS...

[zdohnal]

To no avail... here is the post I found about Mac and CLOCK_MONOTONIC https://discussions.apple.com/thread/253778121?page=2&sortBy=rank .

[michaelrsweet]

The CI failure might be cause by load on the system or virtualization speed - the testclock program is just seeing that the monotonic clock is increasing and roughly close to the expected (wall) clock time. Having jitter that large isn't ideal but as long as the values are increasing it won't cause a problem.

[zdohnal]

@michaelrsweet so I would ifdef the test binary and raise acceptable delta for Mac? Or introduce a new function which will allow us to set acceptable delta and use that with a different value on Mac?

[michaelrsweet]

@michaelrsweet so I would ifdef the test binary and raise acceptable delta for Mac? Or introduce a new function which will allow us to set acceptable delta and use that with a different value on Mac?

This is just in the unit test program, so we can increase the acceptable delta (say to 1 second) or just drop it to avoid future woe. The key is to flag values going backwards.