fix(material/datepicker): fix handling of short years

[mmalerba]

This should hopefully fix test flakiness introduced by dd49fa6. The previous fix seems to sometimes report an incorrect date based on the time zone the test runs in

[crisbeto]

I think that this is basically the same as the old fix, except that the previous approach had the small optimization that we only mutated the new date if we detect the edge case for years 0 to 99, whereas this approach does it every time. My concern is that since we're setting the hours/minutes/seconds to 0, we might break some people that depended on the old behavior.

If we only wanted to fix the tests, we could create the date as February 1st, instead of January 1st.

[mmalerba]

I think the old behavior set the hour/m/s/ms to 0 anyways? If you construct a date without specifying those 0 is assumed. I think this does differ from the old way. The old way you overwrote only the year, and my theory is that for locales where the timezone offset has changed over time this was leading to dates that were slightly off. At the very least this seems to pass on CI where the old approach was regularly failing

[crisbeto]

My hesitation comes from the fact this is used anywhere we interact with a date, not just when it's formatted which could break people in subtle ways, whereas the previous fix only affected formatting. Let's see how it goes during the presubmit and we can adjust the test, instead of changing the logic, if it ends up breaking too many targets.

[mmalerba]

I still don't quite understand your concern. I modified the same 2 methods your change did: _createDateWithOverflow and _format. Yes, _createDateWithOverflow zeros out the h, m, s, ms, but it always did that, so that's nothing new

[crisbeto]

I think that the changes are fine, but at a glance it isn't obvious why they would be fixing the test, considering that they're setting the same values as the defaults.

Looking through it again, I have a theory about what might've been throwing it off. Note how _createDateWithOverflow creates a regular date and _format creates a UTC one, however when both of them go through _correctYear, we call setFullYear no matter whether it's UTC or not. Maybe the problem is that _format should call setUTCFullYear and _createDateWithOverflow should use setFullYear (which is basically what your changes do after you remove _correctYear)?

[angular-automatic-lock-bot]

This issue has been automatically locked due to inactivity.
Please file a new issue if you are encountering a similar or related problem.

Read more about our automatic conversation locking policy.

_{This action has been performed automatically by a bot.}