[Python-Dev] os.path.getmtime on Windows

["Martin v. Löwis"]

Christian Tismer wrote:
> 1. create a file
> 2. get it's os.path.getmtime()
> 3. change your time zone
> 4. get os.path.getmtime again
> 
> compare - the time stamps are different.
> Change the time zone back, and they are identical, again.

Just to add an important detail here: I assume
you did not exit Python between step 2 and step 4, right?
(please confirm whether this is the case).

If so, it has nothing to do with daylight saving time.

If I start a new Python interpreter and fetch the mtime
again, it will report the value I got at step 2.

This is on NTFS, so the time stamps the system reports
(in FindFirstFile) are true UTC.

What happens is apparently this: msvcrt converts the UTC
time to local time, using FileTimeToLocalFileTime; this
gets the new time zone offset from the system. It then
converts it back through __loctotime_t. This invokes
__tzset, however, __tzset caches the _timezone offset,
so it is unaffected by the time zone change.

So *this* specific problem goes away as soon as we
start dropping msvcrt.

Regards,
Martin

P.S. The other problem (FindFirstFile gives bad UTC time
stamps on FAT, for files created in a different DST period)
has no real solution, AFAICT. There is no portable way to
determine whether the underlying file system stores time
stamps in local time or in UTC, and even if you know that
file stamps were in local time, you still couldn't reliably
convert that to UTC (because of the repeated hour, and
because of potential time zone changes).

So I would rather return to the user what the system gives
me, knowing that
a) the problem exists on FAT only, and people should use
   NTFS if they care about time stamps
b) it is better to document the limitation instead of
   trying to work around it: msvcrt shows that attempts
   to work around it make the problem worse, not better.
   So I would like to use FindFirstFile for stat(),
   and GetFileType+GetFileInformationByHandle for fstat().