BUG: Incorrect parsing of ISO 8601 durations

lmeyerov commented 4 years ago

[x ] I have checked that this issue has not already been reported.
[ x] I have confirmed this bug exists on the latest version of pandas.
[ ] (optional) I have confirmed this bug exists on the master branch of pandas.

Code Sample, a copy-pastable example

pd.Timedelta('P1Y')
pd.Timedelta('P1M')
pd.Timedelta('P1W')
pd.Timedelta('P1D')
pd.Timedelta('P1DT1H')
pd.Timedelta('PT1H')
pd.Timedelta('P1H')

Problem description

Trying most iso8601 duration formats fails.

For above examples:

Expected pd.Timedelta instances of 1yr, 1mo, 1wk, 1d, 1d 1hr
... but got ValueError: Invalid ISO 8601 Duration format - ... on Y/M/W, PT1H
... 1D was '0'
... 1DT1H was '1d' (no hr)

Encountered while trying to figure out neo4j.time.Duration().isoformat() -> pandas ->arrow > rapids cudf timedelta[*] . Neo4j returns values like P1Y3M15DT2H3M and P1Y3M.

jreback commented 4 years ago

Y M are not valid Timedeltas (as they are not fixed intervals)

they the other might be a bug if they are in fact ISO 8601

pull requests welcome

lmeyerov commented 4 years ago

This sounds like pandas docs should not say they support ISO durations

Pandas date time guide:

vs wikipedia summary of iso8601 durations:

lmeyerov commented 4 years ago

Linking:

https://github.com/pandas-dev/pandas/issues/29773

https://github.com/pandas-dev/pandas/pull/15136

pandas-dev / pandas

BUG: Incorrect parsing of ISO 8601 durations #36204

Code Sample, a copy-pastable example

Problem description