Ilhan Polat
Ilhan Polat
As a CI job I think compiling the entire stack with ASAN enabled might be valuable but is there a pixi solution that only enables ASAN on scipy and not...
> The issue with only compiling part of your stack with sanitizers is you will have both false positives and false negatives at library boundaries. IMO it's worth the internal...
> and both have ASan CI to keep things clean. Depends on your `ASAN_OPTIONS` in terms of how strict your flags are. There are always a few bytes leaking here...
Structure of an array or ndarray with banded slices is not the structure `assume_a` accepting functions expecting. The `assume_a` works on the common properties of all the slices. The point...
Solve banded and solve can call the same C++ code. It only needs to skip the slice loading which is just a flag, solve_banded is not batched yet. and if...
I mean it doesn't accept different kl ku arrays. Without it batching does not make too much sense. At c++ level it is fast and can be done per slice...
> Some preliminary tests show that this can lead to segfaults. This is a no-no. It has to come back to Python layer and die there if needed no matter...
On MSVC flags are slash format `/WL`, `/Wall` not dash format like `-wl`, `-Wall`.
This one, from a 2 minute digging seems like choking on checking the dimensions of B (2nd arg) in `idd_reconid` and `idz_reconid`. We might throw some protection at the entrance...
Hi @charlotte12l, in #13938 another attempt has been started hence would you like to join forces or review or any other way be involved if you can still spare some...