Conversation
pandas 3.x changed the unit resolution (ns. vs. us) for strings. This
test does a `DatetimeIndex + Timedelta`. Previously both those used
ns resolution. Now `pd.date_range(...)` get `us` resoultion, while
`Timedelta(days=1)` gets `ns` resolution. Then we tried to append a
us-resolution index to a ns-resolution index, and got an error.
Fixed by changing to `Timdelta("1D")` to get `us` for both.
Fixes errors like `dask/dataframe/io/tests/test_parquet.py::test_append_overlapping_divisions`.
Fixes errors like `dask/dataframe/dask_expr/tests/test_collection.py::test_shift_with_freq_TimedeltaIndex`.
Fixes errors like `dask/dataframe/tests/test_arithmetics_reduction.py::test_reductions_non_numeric_dtypes`.
Unit Test ResultsSee test report for an extended history of previous test failures. This is useful for diagnosing flaky tests. 9 files ±0 9 suites ±0 3h 12m 25s ⏱️ +5s Results for commit a2b9ef2. ± Comparison against base commit 2f2d099. ♻️ This comment has been updated with latest results. |
# with '#' will be ignored, and an empty message aborts the commit. # # On branch tom/upstream-compat-3 # Your branch is up to date with 'origin/tom/upstream-compat-3'. # # Changes to be committed: # modified: dask/dataframe/tests/test_utils_dataframe.py # # Untracked files: # a.parquet # data.parquet/ # debug.py # ds.parquet/ # failures.md # nation.impala.parquet # out.log # out.parquet/ # string.parquet/ # test.parquet # test_bug.py #
|
This should be ready for review @jacobtomlinson or @jrbourbeau if either of you have a chance. The errors from my last run are here. a2b9ef2 fixes the first error and the remainder are resample errors and will hopefully be resolved by pandas-dev/pandas#63262. The outstanding item is mentioned in #12178 (comment) for how we ought to handle unit inference in |
Part of #12178. I still need to handle the resample failures.
See the commit messages for details.