fix upstream dev issues #9953
fix upstream dev issues #9953
Conversation
| @@ -579,20 +579,23 @@ def _numbers_to_timedelta( | |||
| ) -> np.ndarray: | |||
| """Transform numbers to np.timedelta64.""" | |||
| # keep NaT/nan mask | |||
| nan = np.isnan(flat_num) | (flat_num == np.iinfo(np.int64).min) | |||
| if flat_num.dtype.kind == "f": | |||
There was a problem hiding this comment.
That's a bit more explicit. The asarray-wrapping is needed for upstream-dev to keep 0d-arrays an array.
|
|
||
| # in case we need to change the unit, we fix the numbers here | ||
| # this should be safe, as errors would have been raised above | ||
| ns_time_unit = _NS_PER_TIME_DELTA[time_unit] | ||
| ns_ref_date_unit = _NS_PER_TIME_DELTA[ref_unit] | ||
| if ns_time_unit > ns_ref_date_unit: | ||
| flat_num *= np.int64(ns_time_unit / ns_ref_date_unit) | ||
| flat_num = np.asarray(flat_num * np.int64(ns_time_unit / ns_ref_date_unit)) |
There was a problem hiding this comment.
For pandas 3.0 CoW (copy on write) this array might be flagged readonly. So creating a copy here. The asarray-wrapping is needed for upstream-dev to keep 0d-arrays an array.
There was a problem hiding this comment.
Very nice @kmuehlbauer, thanks—great that pandas 3 will convert datetime.datetime objects to microsecond-resolution np.datetime64 values.
The upstream build failure with the latest commit looks to be unrelated—a hopefully transient issue installing NumPy—and it passed with the commit prior.
|
Thanks for checking @spencerkclark. I'll merge later today, if no one beats me to it. |
|
@pydata/xarray There is something strange going on wrt to the failing test (linux/macos py 3.12): ERROR xarray/tests/test_distributed.py::test_dask_distributed_zarr_integration_test[True-True]
- Failed: 5 thread(s) were leaked from testThis surfaced some days ago, first only on linux then also on macos. Since many things have changed in the recent days (ubuntu-latest 22.04-> 24.04, zarr 2 -> zarr 3 with dependency change) this is very difficult to debug. As everything works well for our upstream-dev, it really looks like an upstream issue which was already fixed in the dev versions. Another reason might be some inconsistent conda dependencies which would equally painful to debug. |
|
Going to merge now, to unblock the upstream-dev runs again (and nightly run). |
whats-new.rst