num2date ignoring tz in v1.5.0

[brendene]

The behavior of matplotlib.dates.num2date changed between stable versions 1.4.3 and 1.5.

In version 1.5 the implementation of helper function, _from_ordinalf ignores the tz argument and always returns datetimes in UTC.

def _from_ordinalf(x, tz=None):
    """                                                                                                                                                         
    Convert Gregorian float of the date, preserving hours, minutes,                                                                                             
    seconds and microseconds.  Return value is a :class:`datetime`.                                                                                             

    The input date `x` is a float in ordinal days at UTC, and the output will                                                                                   
    be the specified :class:`datetime` object corresponding to that time in                                                                                     
    timezone `tz`, or if `tz` is `None`, in the timezone specified in                                                                                           
    `rcParams['timezone']`.                                                                                                                                     
    """
    if tz is None:
        tz = _get_rc_timezone()

    ix = int(x)
    dt = datetime.datetime.fromordinal(ix).replace(tzinfo=UTC)

    remainder = float(x) - ix

    # Round down to the nearest microsecond.                                                                                                                    
    dt += datetime.timedelta(microseconds=int(remainder * MUSECONDS_PER_DAY))

    # Compensate for rounding errors                                                                                                                            
    if dt.microsecond < 10:
        dt = dt.replace(microsecond=0)
    elif dt.microsecond > 999990:
        dt += datetime.timedelta(microseconds=1e6 - dt.microsecond)

    return dt

In version 1.4.3 this is handled correctly.

[tacaswell]

attn @pganssle

[tacaswell]

@brendene Can you provide a minimal example of the usage of this?

[brendene]

I discovered this when using Pandas (0.17.1) and matplotlib (1.5.0) while trying to plot time-series data where the x-axis is timezone aware datetime. No matter what I did I could not change the x-axis to use a timezone other than UTC.

Here is a simple example which I believe causes the plotting issue.

Version 1.5.0

In [1]: import matplotlib.dates, datetime, pytz

In [2]: x = datetime.datetime.utcfromtimestamp(1448154154)

In [3]: y = matplotlib.dates.date2num(x)

In [4]: print matplotlib.dates.num2date(y), matplotlib.dates.num2date(y, tz=pytz.timezone("US/Eastern"))
2015-11-22 01:02:34+00:00 2015-11-22 01:02:34+00:00

Version 1.4.3

In [1]: import matplotlib.dates, datetime, pytz

In [2]: x = datetime.datetime.utcfromtimestamp(1448154154)

In [3]: y = matplotlib.dates.date2num(x)

In [4]: print matplotlib.dates.num2date(y), matplotlib.dates.num2date(y, tz=pytz.timezone("US/Eastern"))
2015-11-22 01:02:34+00:00 2015-11-21 20:02:34-05:00

[mdboom]

git bisect points at one of these three commits:

The first bad commit could be any of:
5af753c
ee7931c
4b0d94b