thalassemia

I've been noticing sporadic failures on the Windows `cp311 (build sdist + wheel), full, no pythran` test for a while now, and I do not know what is causing it....

Yeah, I'm not sure how to do the rebase. Would be great if you could make that commit @h-vetinari!

Oh that makes sense! I basically copied the preprocessor flags at the top of my wrappers from `npy_common.h` and was confused that using `npy_complex64/128` directly still caused segfaults. I didn't...

Thanks for pointing out those warnings and investigating @rgommers! On my machine, casting the return type variable to the expected Numpy type gets rid of those warnings (3ccf67c).

> Note the use of scipy-openblas wheels requires a `scipy_` prefix, which in turn requires something like defining `BLAS_SYMBOL_PREFIX=scipy_` (done by the pkgconfig file in the wheel) and using the...

> Hmm.. Do we actually compress those v1 pages then? 🤔 Yup, here's the relevant code. https://github.com/pola-rs/polars/blob/864e7504fc8b38e82688b43f63681b6cda77fd61/crates/polars-parquet/src/parquet/write/compression.rs#L23-L38 > I don't think we should activate it for nested types then. That...

@ritchie46 So sorry for this and thank you @nameexhaustion for fixing it! I should've written a test with more random, realistic data. Since this feature still sounds worthwhile, I'll look...

This test was re-enabled in gh-19855 (also fixed the complex BLAS 32-bit segfaults) and made deterministic in gh-20221. Did not realize this issue existed when making those PRs, but I...