I am currently upgrading pandas, and as part of this, attempted to test
its parquet I/O functionality.  This failed to even import fastparquet,
with the error

AttributeError: module 'cramjam' has no attribute 'gzip'.  Did you mean:
'bzip2'?

e.g. https://salsa.debian.org/science-team/pandas/-/jobs/9907658

(fastparquet's own autopkgtest does not actually test anything, because
plain pybuild-autopkgtest cannot find the tests.  This is probably a bug
in itself.)

I do not know whether this bug is actually in fastparquet or in cramjam;
it is not *obviously* known to either upstream.

Control: forwarded 1141635 https://github.com/dask/fastparquet/issues/984
Control: affects 1141635 python3-cramjam

This looks like it's due to this patch in cramjam:

https://salsa.debian.org/python-team/packages/python-cramjam/-/blob/master/debian/patches/drop-cramjam-unsupported?ref_type=heads

which removes gzip support entirely.

The fastparquet package could change compression.py to only
conditionally include gzip support based on whether cramjam.gzip is
present.  However, fastparquet appears to actually use gzip as the
default compression algorithm in a few different places, so perhaps
someone with a better understanding of the parquet ecosystem can think
through the implications of such a change better than i can.