Time function improvements (fix #1912)

[nicowilliams]

• Fixes Wrong timezone names in strftime/strflocaltime for DST #1912, and fixes / documents mktime/0 returns GMT time instead of local time #2865.
• Makes mktime preserve any fractional seconds part.
• Adds timelocal and timegm.

[nicowilliams]

Gotta normalize tm_zone for tests, or maybe change the tests to normalize it. Probably the latter.

[nicowilliams]

For Windows we should try using tzname in f_localtime().

[nicowilliams]

We can also work around the lack of a tm_gmtoff field in some cases by checking the local and gm time and computing the difference, just as we can get the TZ name from tzname.

[nicowilliams]

We can also work around the lack of a tm_gmtoff field in some cases by checking the local and gm time and computing the difference, just as we can get the TZ name from tzname.

Er, this is fraught. For example, if I try to use localtime_r() and gmtime_r() then difftime() with a tm value that is in DST, I get the wrong result. I'm inclined not to do this.

[nicowilliams]

Now with some boundary tests (which found bugs).

[nicowilliams]

I've left resolved comments to explain some things, mainly the range validation checks for all the fields of broken down time (which were wrong in an earlier version).

[nicowilliams]

Well, I think we should merge this in time for 1.7, but I'm clearing the milestone anyways.

[nicowilliams]

Ok, we should allow datetimes before 1970, so I'm making the range 1900..9999, though perhaps on 64-bit systems we can go to -9999..9999?

EDIT: Yes, so let's go to 0001..9999 as the range for years. It works on Linux.

[trantor]

mmm was this not merged before 1.7 ? @nicowilliams

[nicowilliams]

mmm was this not merged before 1.7 ? @nicowilliams

@itchyny thought it was too risky for 1.7. I intend to do a new release soon with this and a few other things.