matplotlib.dates.DayLocator cannot handle invalid input

[chrisjcameron]

This is a relatively minor bug in the way DayLocator tries to report an error message when running the tick_values() method. It is unable to report the invalid input because the invalid input causes a type error while formatting the error message. To be clear, the issue is not that the class breaks with invlaid input but how it breaks with invalid input.

The method could check for Nan or Inf numpy.float64 values before trying to use the %d format, switch to the %f format or catch zero-width intervals at initialization time.

Matplotlib 1.5.2, Anaconda Python 2.7 on Mac OSX

    import datetime
    import matplotlib
    import matplotlib.pyplot as plt
    from matplotlib.dates import DateFormatter, DayLocator

    vals = range(22)
    dates = [datetime.datetime(2016, 8, 1, x, 20, 20) for x in range(22)]

    plt.plot(dates, vals)
    ax = plt.subplot(111)
    ax.xaxis.set_major_locator(DayLocator(interval=0))  #causes exception

/Users/cjc73/anaconda/lib/python2.7/site-packages/matplotlib/axis.pyc in iter_ticks(self)
    892         Iterate through all of the major and minor ticks.
    893         """
--> 894         majorLocs = self.major.locator()
    895         majorTicks = self.get_major_ticks(len(majorLocs))
    896         self.major.formatter.set_locs(majorLocs)

/Users/cjc73/anaconda/lib/python2.7/site-packages/matplotlib/dates.pyc in __call__(self)
    812             return []
    813 
--> 814         return self.tick_values(dmin, dmax)
    815 
    816     def tick_values(self, vmin, vmax):

/Users/cjc73/anaconda/lib/python2.7/site-packages/matplotlib/dates.pyc in tick_values(self, vmin, vmax)
    842                 'RRuleLocator estimated to generate %d ticks from %s to %s: '
    843                 'exceeds Locator.MAXTICKS * 2 (%d) ' % (estimate, vmin, vmax,
--> 844                                                         self.MAXTICKS * 2))
    845 
    846         dates = self.rule.between(vmin, vmax, True)

TypeError: %d format: a number is required, not numpy.float64

[anntzer]

Looks like we're having a premature optimization:

        # estimate the number of ticks very approximately so we don't
        # have to do a very expensive (and potentially near infinite)
        # 'between' calculation, only to find out it will fail.
        nmax, nmin = date2num((vmax, vmin))
        estimate = (nmax - nmin) / (self._get_unit() * self._get_interval())
        # This estimate is only an estimate, so be really conservative
        # about bailing...
        if estimate > self.MAXTICKS * 2:
            raise RuntimeError(
                'RRuleLocator estimated to generate %d ticks from %s to %s: '
                'exceeds Locator.MAXTICKS * 2 (%d) ' % (estimate, vmin, vmax,
                                                        self.MAXTICKS * 2))

        dates = self.rule.between(vmin, vmax, True)

I'd just strip out everything but the last line and call Locator.raise_if_exceeds with the actual tick count.