hschreiber

Neither the C++ documentation, nor the Python documentation mentions the complexity of `num_simplices`, so yes, we should definitely add it. To avoid the extra cost for people not using this...

> I didn't check this particular case, but note that the dummy could return by value and the non-dummy return by reference, that's not forbidden. And then it may be...

I agree that it is not clear how feasible it really is. Perhaps it is possible to generalize those methods with callback function, I don't know yet.

> > only simplicial/cubical ones (i.e. with a predictable boundary size) > > I think more precisely: cells of dimension 0 must have an empty boundary, and cells of dimension...

> > Also it is used to determine the start sign of the alternated sum for the boundary, so it should be bot completely random. > > (in Z/2Z the...

So, I changed it to a non-documented method temporarily. > pcoh was already supposed to be generic, but I guess there is still space for a lower level, which would...

> Timings with `alpha_complex_persistence` utility are almost the same between the 2 versions: | dataset | # vertices | # simplices | compute_persistent_cohomology | _without_optimizations | | --- | ---...

We should investigate if keeping the C++ versions of the sphere/torus samplings in addition to the python versions is really worth it. If having everything in Python makes it easier...

> It would have been so much easier to read if you didn't include unrelated changes in the PR (reformatting, reordering includes, etc)... It is too late to go back...

Thanks for the unitary tests! But I think you forgot to include them?