BUG: Why is resampling with rule='7d' different than resampling with rule='168h'?

[mkp-gebensleben]

Research

• I have searched the [pandas] tag on StackOverflow for similar questions.

• I have asked my usage related question on StackOverflow.

Link to question on StackOverflow

https://stackoverflow.com/questions/70436514/timestamp-binning-mechanics-when-resampling

Question about pandas

>>> df = pd.DataFrame(index=pd.date_range(start='2021-04-21 01:00:00', end='2021-04-28 01:00', freq='1d'), data=[1]*8)
>>> df.resample(rule='7d', origin='2021-04-29 00:00:00', closed='right', label='right').sum()
            0
2021-04-22  2
2021-04-29  6
>>> df.resample(rule='168h', origin='2021-04-29 00:00:00', closed='right', label='right').sum()
            0
2021-04-22  1
2021-04-29  7

1. Why does this happen?
2. Should this happen?

Using pandas 1.3.5

[ms7463]

Looks like the issue happens here:

pandas/pandas/core/resample.py

Lines 1656 to 1668 in 5a22750

	if self.freq != "D" and is_superperiod(self.freq, "D"):
	if self.closed == "right":
	# GH 21459, GH 9119: Adjust the bins relative to the wall time
	bin_edges = binner.tz_localize(None)
	bin_edges = bin_edges + timedelta(1) - Nano(1)
	bin_edges = bin_edges.tz_localize(binner.tz).asi8
	else:
	bin_edges = binner.asi8
	# intraday values on last day
	if bin_edges[-2] > ax_values.max():
	bin_edges = bin_edges[:-1]
	binner = binner[:-1]

The 7d closed=right resample hits this condition, and this code branch readjusts the bin edges (starting at line 1666). If you run in debug mode and skip over this adjustment, you get the same results in your example.

Looks like this logic or similar has been there for a long time (but potentially meant to deal with Monthly/Weekly frequencies, rather than N*Daily frequencies?), maybe @mroeschke or @jorisvandenbossche can comment on it, since it seems they touched this section of the code most recently.

[mroeschke]

This looks like a bug because resample has special logic for redefining 'D' in the presence of a DST transition #41943 which seems to negatively impact when there is no timezone defined in the example above.

This will hopefully be fixed in 2.0 where we want to remove this special casing: #44823

[mkp-gebensleben]

@mroeschke

which seems to negatively impact when there is no timezone defined in the example above

I'm getting the same results when using a timezone.

>>> df = pd.DataFrame(index=pd.date_range(start='2021-04-21 01:00:00', end='2021-04-28 01:00', freq='1d', tz=0), data=[1]*8)
>>> df.resample(rule='7d', origin='2021-04-29 00:00:00+00:00', closed='right', label='right').sum()
                           0
2021-04-22 00:00:00+00:00  2
2021-04-29 00:00:00+00:00  6
>>> df.resample(rule='168h', origin='2021-04-29 00:00:00+00:00', closed='right', label='right').sum()
                           0
2021-04-22 00:00:00+00:00  1
2021-04-29 00:00:00+00:00  7