Including the pytest-xdist package automatically configures numprocesses
to $NIX_BUILD_CORES, so specifying these on a package level has become
redundant.
Using such an option within the package set this way will result in
different packages depending on different versions of dask and thereby
collisions.
Instead, we can simply declare in the package offering the extras the
packages the extra needs.
There were hydra failures (https://hydra.nixos.org/build/141870744/nixlog/1) for some tests requiring data on github (failing on mac and linux), and tests requiring local network access on mac.
Disable the former and enable the latter.
Importing dask.dataframe in a sandboxed build results in a TypeError like
this:
File "/nix/store/nv60iri29bia4szhhcvsdxgsci4wxvp6-python3.8-dask-2021.03.0/lib/python3.8/site-packages/dask/dataframe/io/csv.py", line 392, in <module>
AUTO_BLOCKSIZE = auto_blocksize(TOTAL_MEM, CPU_COUNT)
File "/nix/store/nv60iri29bia4szhhcvsdxgsci4wxvp6-python3.8-dask-2021.03.0/lib/python3.8/site-packages/dask/dataframe/io/csv.py", line 382, in auto_blocksize
blocksize = int(total_memory // cpu_count / memory_factor)
TypeError: unsupported operand type(s) for //: 'int' and 'NoneType'
This occurs because dask.dataframe has a non-deterministic component which
generates an automatic chunk-size based on system information.
This went unnoticed because the dask tests were disabled.
Changes:
- add a patch making the chunk-size inference more robust
- re-enable the tests
Resolves#120307
In 87a19e9048 I merged staging-next into master using the GitHub gui as intended.
In ac241fb7a5 I merged master into staging-next for the next staging cycle, however, I accidentally pushed it to master.
Thinking this may cause trouble, I reverted it in 0be87c7979. This was however wrong, as it "removed" master.
This reverts commit 0be87c7979.
I merged master into staging-next but accidentally pushed it to master.
This should get us back to 87a19e9048.
This reverts commit ac241fb7a5, reversing
changes made to 76a439239e.