Run only one of mini_mktime(), mktime() #21885
Conversation
|
Nice analysis! |
When you use the link creation tool in the github editor it creates a link using the highlighted text as the text of the link, not the url of the link. You need to replace the "url" in the produced markdown to make it a useful link. eg. from your markdown If you want the link text like https://perl.develop-help.com/?b=smoke-me%2Fkhw-mktime to just be both the text and url of a link don't bother with the link creation tool, github will linkify it for you and I can stop visiting |
ext/POSIX/t/posix.t
Outdated
| # This is an actual time a leap second occurred. | ||
| # https://en.wikipedia.org/wiki/Leap_second */ | ||
| try_strftime("Mon Jun 30 23:59:60 1997 181", 60,59,23, 30,5,97); |
There was a problem hiding this comment.
strftime() works in local time, but leap seconds are added at midnight UTC, not localtime.
So this change only matters when the current time zone is UTC (Britain in winter, some parts of Africa I assume)
There was a problem hiding this comment.
So my analysis was flawed.
Now updated to detect mktime not handling leap seconds like mini_mktime does and to override with the latter
mini_mktime() has less overhead than libc mkttime(), but it doesn't get all the desired fields on all platforms. Prior to this commit, both got called on such platforms, roughly doubling the amount of time spent. So just use the slightly more expensive function on those platforms and avoid that redundant overhead. The only glitch I've seen is that mktime() likely doesn't handle leap seconds the same way mini_mktime() does; so this calls both functions for those very rare instances.
mini_mktime() is a faster, stripped down version of mktime() that ignores locale-dependent information. But it doesn't work when the structure it fills in contains fields that are locale-related. A patch 20 some years ago called mktime() to fill in the missing fields.
But that ends up executing similar algorithms twice. If we need the locale-related info, why not just call mktime() once and be done with it. I figured I must be missing something, so started a smoke-me branch with this change. https://perl.develop-help.com/?b=smoke-me%2Fkhw-mktime. So far everything is passing.
There is one glitch on my Linux glibc. mktime() did not accept 60 for the second number (a leap second) , and mini_mktime() does. And there is a test that includes a leap second.
The right thing to do to calculate the number of seconds since the epoch is to have a table of when actual leap seconds happened, and take that into account. I guess there haven't been enough of them to make a difference. https://en.wikipedia.org/wiki/Leap_second has an extended discussion on this.
This commit uses plain mktime() when the fields exist that are not populated by min_mktime