Timestamp support has floating-point roundoff

[edemaine@…]

Example:

>>> import yaml, datetime
>>> yaml.dump(datetime.datetime(2005, 7, 8, 17, 35, 4, 517600))
'2005-07-08 17:35:04.517600\n'
>>> yaml.load(_)
datetime.datetime(2005, 7, 8, 17, 35, 4, 517599)

This breaks the desired rule that yaml.load(yaml.dump(x)) == x in a case where there should be no roundoff. (datetime.datetime uses integers everywhere to avoid any error.)

The offending code seems to be line 321 in yaml/constructor.py:

fraction = int(float(values['fraction'])*1000000)

This seems to be an "easy" way to convert the trailing '.517600' into an integer, but it can go beyond floating-point precision. Wouldn't the following work?

fraction = int(values['fraction'][:6].ljust(6, '0'))

[xi]

Fixed in [234].

Thanks for the patch, it works perfectly.