BUG: PandasAutoDateLocator num_days calculation is wrong

[dadkins]

This line in PandasAutoDateLocator looks like it's missing some parentheses.

https://github.com/pandas-dev/pandas/blob/master/pandas/tseries/converter.py#L264
num_days = ((delta.years * 12.0) + delta.months * 31.0) + delta.days

I stepped through here and with a delta of 5 years, num_days was 60! It's easy enough to fix:

    num_days = (((delta.years * 12.0) + delta.months) * 31.0) + delta.days

[jorisvandenbossche]

That indeed seems to be the case.

Do you have a plot example that shows 'incorrect' behaviour because of this? That would make it possible to check that it actually changes the look in a positive way

[mroeschke]

Looks like PR 14716 addressed this but stalled.

[jorisvandenbossche]

Good catch.
Maybe we should just merge it, without a test / example of visual change, as it is clearly a mistake in the code. Although a confirmation with an actual example which improves due to this change would still be useful.

[jorisvandenbossche]

Actually, looking at that code, I don't think we are going to find an example:

num_days = ((delta.years * 12.0) + delta.months * 31.0) + delta.days
num_sec = (delta.hours * 60.0 + delta.minutes) * 60.0 + delta.seconds
tot_sec = num_days * 86400. + num_sec

if abs(tot_sec) < self.minticks:
    ....

tot_sec (and thus num_days) is only used to check whether it is smaller than minticks which is typically a number like 5. So once you have hours, let alone years, of data, the tot_sec will always be a huge number.
@mroeschke Or am I missing something?

[mroeschke]

Not too familiar with this section of the codebase, but agreed that this shouldn't have an obvious change. The if abs(tot_sec) < self.minticks: block just looks like it's adjusting the axis to plot data that spans maybe seconds(?), in which case num_days would be 0 if triggered (which occurs with the current code and the proposed change.)

[jorisvandenbossche]

I merged the PR, so this can be closed.
@dadkins Thanks for the report

[dadkins]

Thank you for the quick fix.