Jerome Kelleher

We also don't really handle missing data in the standard tree stats API, so I'd be happy to kick this (tricky) can down the road.

Tricky one. At a minimum we need to carefully explain and document this. Ideally I guess we'd support counting ordered coalescent trees also. From an empirical perspective, do you think...

I doubt there is a straightforward multiplier here because the symmetry breaking is recursive, and you'd need a separate rank/unrank method. I think the action we need to close this...

Feel free to open a "draft" PR with an early version for feedback @TymekPieszko

Hi @alxsimon :wave: First up, thanks a million for taking the time to do this packaging work! I don't think there's any particular reason for including/not including the tests directory...

One thing I spotted in your nix package: tskit doesn't use ``setuptools-scm`` (it doesn't work with the dual language structure of the repo)

> I do think this kind of iteration is intrinsically inefficient for most zarr arrays, because it doesn't take any account of the chunk structure. Even if __iter__ caches a...

Sorry, I don't get it (I must be missing something basic here!) ```python def first_dim_iter(z): for chunk in range(z.cdata_shape[0]): A = z.blocks[chunk] for a in A: yield a ``` Surely...

> That function will potentially load an entire array into memory (see the first case), which we do not want: Sorry I'm sure I'm being thick here, but I'm still...

Thanks @rabernat, that's why I'm confused then! But in this one chunk case, isn't the whole array getting loaded into memory anyway each time we access a single element, during...