Fix sporadic fails in timer unit-tests on Windows

[jschmidt-icinga]

Fix the test-cases base_timer/invoke and base_timer/scope still sporadically failing on Windows, even after increasing the leeway given to the timers in #10743. Essentially, this gives tests 10x the time on Windows, effectively restoring the timings to the state before #10550, but only on Windows.

[julianbrost]

Windows tests failing here, how ironic. However, that's not related to the change:

Failed to fetch results from V2 feed at 'https://community.chocolatey.org/api/v2/Packages(Id='winflexbison3',Version='2.5.24.20210105')' with following message : Response status code does not indicate success: 503 (Service Unavailable).

Let's try that again, I've restarted the pipeline.

[yhabteab]

Why the scientific notation though, instead of just using 1/10?

even after increasing the leeway given to the timers in #10743.

Wouldn't it make sense to revert that commit now then, since it didn't fix the issue it's supposed to fix?

[jschmidt-icinga]

Wouldn't it make sense to revert that commit now then, since it didn't fix the issue it's supposed to fix?

This PR essentially does that (values reverted to 0.55), just not in a separate commit. Do you want the separate revert commit for a clearer history or did you just not see that I changed the value back?

Why the scientific notation though, instead of just using 1/10?

I subjectively prefer it for literals of 1, 10, etc. since it seems pointless to write out a zero decimal. I can change it to 1.0 etc. but it really doesn't matter IMHO.

[yhabteab]

Do you want the separate revert commit for a clearer history

Yes, I expected a sperate commit generated with git revert 1430e63e727dd2dddadaa84aec31ea183e9cb82d.

Why the scientific notation though, instead of just using 1/10?

I subjectively prefer it for literals of 1, 10, etc. since it seems pointless to write out a zero decimal. I can change it to 1.0 etc. but it really doesn't matter IMHO.

It does matter, because it requires the reader to understand scientific notation to decipher that 1e0 is the same as 1, and so on. I would personally say it is pointless to use scientific notation for such trivial numbers and not the other way around, which even requires you to write more characters (10e0 instead of just 10).

[jschmidt-icinga]

• Added the revert as a separate commit.
• Changed the constexprs to be declared as doubles initialized with integers 1 and 10.

[yhabteab]

If this really fixes the issue, then I'm fine with it.

[julianbrost]

If this really fixes the issue, then I'm fine with it.

I'd expect it to due to the following:

effectively restoring the timings to the state before #10550, but only on Windows

And I think we only started noticing these somewhat regular test failures after that change.