Pyspark switched to pinning py4j==0.10.9 with v3.0.0 - see this commit:
https://github.com/apache/spark/\
commit/fc4e56a54c15e20baf085e6061d3d83f5ce1185d
This meant that since the bump to pyspark v3.0.0 - in this commit:
https://github.com/NixOS/nixpkgs/\
commit/5181547ae6624b462919a806c4d0888e6e4630f4 -
the patch was no longer matching on the 'py4j==0.10.7' string that was
working previously.
The failing patch went unnoticed previously because the version of py4j
pinned by pyspark>=3.0.0 was the same as the py4j provided by nixpkgs.
However, a recent PR (#101636) bumped the version of py4j to 0.10.9.1 in
this commit:
https://github.com/NixOS/nixpkgs/\
commit/43a91282d66223c5cb978d53fbe1033f56dd7f2b
which caused the version pinned by pyspark to no longer match the
version provided by nixpkgs. FWIW, @jonringer flagged this issue on
another PR that tried to bump py4j: #100623.
My solution here was to upgrade the patch's target string to match the
version found in pyspark's current setup.py.
cde89e3ecf