Race conditions in Tests_Date_I18n

[pento]

All of the tests that generate the expected and actual values from date() calls are subject to a race condition, where the time can tick to the next second between calls.

See ​this test job for an example of an inadvertent failure.

[swissspidy]

@pento Thanks for opening the issue! I just saw it today and realized how bad the tests were. Mind a review?

[pento]

To ensure consistent behaviour, the best option would be to get the output of time() at the start of each test, then pass it to the date() and date_i18n() calls, to ensure they're both generating from the same timestamp.

[swissspidy]

Replying to pento:

To ensure consistent behaviour, the best option would be to get the output of time() at the start of each test, then pass it to the date() and date_i18n() calls, to ensure they're both generating from the same timestamp.

But then I can't really test for the case where no timestamp is passed to the function. I mean, ideally we'd mock dates, but the delta comparison should work just as well.

[pento]

I think this'll be the first time we've ever used delta comparisons in our unit tests. :-)

Yah, let's go with it. Can you confirm that the $message parameter isn't required before you commit? The ​PHPUnit Docs imply that it is.

[swissspidy]

Ah yeah you're right, the $message comes before the $delta. It's an empty string by default but I added a message in 38381.2.diff​.

[swissspidy]

In 38871:

Date/Time: Fix unit tests after [38804].

The tests for date_i18n() need to use a delta for comparing timestamps.

Fixes #38381. See #37910.

[netweb]

Referencing here [39012] and [39013] follow up commits that were added to fix an issue with Daylight Savings Time ending in Europe, see ​https://wordpress.slack.com/archives/core/p1477789573011015

[SergeyBiryukov]

In 43562:

Tests: Avoid a race condition in test_date_i18n_handles_shorthand_formats() by using a delta for comparing timestamps.

See #20973, #38381.