Wrong parsing of microseconds in !BaseTemporalField.to_python

[Aymeric Augustin]

Lines 345-348 of django/forms/filelds.py read:

                            datetime_str, usecs_str = value.rsplit('.', 1)
                            usecs = int(usecs_str)
                            dt = datetime.datetime.strptime(datetime_str, format[:-3])
                            return dt.replace(microsecond=usecs)

This only works if there are exactly 6 digits after the dot.

I suggest something along the lines of usecs = int(usecs_str[:6].ljust(6, '0')).

[Calvin Spealman]

Can you explain or demonstrate how it fails currently? I'm not seeing the problem. maybe an example of a value that won't parse correctly and why?

[Cliff Dyer]

I just figured it out. .%f was added to datetime in python2.6 It should fail in the way this bug lists in python2.5 and lower. I'll check it out.

[Cliff Dyer]

Another bug in there: It checks to see that there is one and only one '.' in the input before applying the workaround. However, a format like '%Y.%m.%d %H:%M:%S.%f' would require three dots, and would break this. It should instead test that the number of dots in the input match the number of dots in the format, and then apply the fix. I'll make sure that case gets tested as well.

[Cliff Dyer]

I'm having trouble getting a 2.5 environment set up right now, but the following regression test patch should exercise the bug, and the other patch should fix it.

[Cliff Dyer]

Tests that show the bug on python < 2.6

[Cliff Dyer]

Test to exercise the bug in python < 2.6

[Cliff Dyer]

Tests that show the bug on python < 2.6

[Cliff Dyer]

Fix for the bug.

[Cliff Dyer]

Use 17134_datetime_bug_tests.3.diff and 17134_datetime_bug_fix.diff. Thanks to Jim Allman, I managed to verify that the tests and fix both do what they're supposed to do in a python2.5 environment.

[Karen Tracey]

It's not fixed until it's checked into the repo...

[Karen Tracey]

In [17092]:

Fixed #17134: Corrected Python 2.5 fallback code for parsing microseconds in time values. Thanks aaugustin and jcd.

[Cliff Dyer]

Thanks kmtracey, and sorry for the process faux pas.