Fix parsing AM/PM hours

[straight-shoota]

Fixes #9333

[jhass]

Not quite there yet.

bin/crystal eval 'p Time.parse("12", "%H", Time::Location::UTC)'
Using compiled compiler at .build/crystal
0001-01-01 00:00:00.0 UTC

I think we need to turn @pm into a tri-state of "AM", "PM", "24 hour clock"

[straight-shoota]

So we're missing even more specs... =)

[jhass]

Might also want to add some to verify %l and %I on its own without %P parses to am.

[straight-shoota]

I found another issue and fixed it in 9a90d33: On the 24-hour clock, 24:00 is a valid time indicating midnight and equivalent to 00:00 of the next day (see ISO 8601 §5.3). The validation in Time#initializer incorrectly rejected this.

[jhass]

bin/crystal eval 'p Time.parse("12", "%l", Time::Location::UTC)'
Using compiled compiler at .build/crystal
0001-01-01 12:00:00.0 UTC

Should this be 00:00?

If yes, make sure %P%l still parses PM12 to 12:00.

[straight-shoota]

Hm, I guess either could be right?

Ruby parses 12 with %I as 12. libc's strptime parses it as 0.

Maybe Ruby's implementation didn't think about this case. And it's really rare.
I have not preference which one it should be.

[jhass]

Well, for 0-11 it parses as AM, so why not parse the full range as AM?

[straight-shoota]

Okay, I'll use that. That means Time.parse("00", "%I") should be invalid, at least it's in both Python and Ruby (which seems to indicate that Ruby's behaviour might not be fully intentional).

[straight-shoota]

Also added a spec for when 24-hour clock hour is used with am/pm (which is ignored).

[jhass]

Okay, I'll use that. That means Time.parse("00", "%I") should be invalid, at least it's in both Python and Ruby (which seems to indicate that Ruby's behaviour might not be fully intentional).

So right now this rejects 0 for %l but accepts 0am for %l%P, but the rationale was that if no %P is given for %l/%I, then we assume AM. And both Ruby and Python seem to follow that rationale (except for Ruby's 12AM bug) and reject 0AM generally.

I don't care too much which way we go, looking at https://en.wikipedia.org/wiki/12-hour_clock#Confusion_at_noon_and_midnight 0AM seems to be used by the Japanese at least sometimes, so I'd slightly favor allowing it. But then even Ruby does not parse it (being of Japanese origin). In any case 0 and 0AM should parse the same in all cases for %l and %l%P respectively.

[straight-shoota]

Oh yeah 0 am should be invalid, too. I'm not sure why I added it...

[jhass]

This seems to implement the most common interpretation according to https://en.wikipedia.org/wiki/12-hour_clock#Confusion_at_noon_and_midnight now