PERF: Remove docstrings from inline cython code

[wlach]

These functions aren't externally available, and removing them saves
a small amount of space (approximately 250k in the size of the generated
.so's on a linux64 machine).

[wlach]

More details:

base sizes: https://gist.github.com/wlach/d1a70c3246e51137e71d6f992ff34149
"no docstrings" sizes: https://gist.github.com/wlach/ac2151d94ba1c895f530ee1585c545c9

Calculating savings:

>>> import pandas
>>> df1 = pandas.read_csv('cython-sizes-base.csv')
>>> df1['bytes'].sum()
90027048
>>> df2 = pandas.read_csv('cython-sizes-no-docstrings.csv')
>>> df2['bytes'].sum()
89744976

I am especially interested in this for the case of pyodide, where the load times of pandas is a very large bottleneck: pyodide/pyodide#347

I don't expect this to help that much, but any savings (which doesn't affect end user functionality) is welcome.

[rth]

Thanks for the PR! I'm not sure they are really private, because this does break some of the tests checking docstrings.

[jbrockmendel]

I don't expect this to help that much, but any savings (which doesn't affect end user functionality) is welcome.

If you can get this passing the CI, this sounds good to me.

[jbrockmendel]

can you add a comment pointing back to this PR

[wlach]

Thanks for the PR! I'm not sure they are really private, because this does break some of the tests checking docstrings.

Yes, I think you're right on closer inspection (I had a hard time understanding the test runs) -- and it's quite likely it's the larger docstrings which are the ones that are really needed. This is probably not worth pursuing, but I'll leave this open for a bit in case I'm missing something obvious.

[lithomas1]

Perhaps you could try making this a compile time option that is disabled by default?

[wlach]

Perhaps you could try making this a compile time option that is disabled by default?

Not my call, but I'd personally be a bit reluctant to do this -- the savings aren't IMO great enough to justify the increased complexity.

[mroeschke]

As an aside, if you're looking to implement space savings, there has been consensus to remove tests from the distribution: #30741

[rth]

We are already unvendoring pandas tests as a separate package in Pyodide, so that is not an issue.

[github-actions]

This pull request is stale because it has been open for thirty days with no activity. Please update and respond to this comment if you're still interested in working on this.

[datapythonista]

Looks like we're not moving forward with this, and the discussion is stale, closing.