Jerome Kelleher

It probably wouldn't matter the vast majority of the time, but yes, we'd end up doing this I'd say.

Another thing we need to check as well before going through with this - does the subclassed numpy array still work well with numba? Can it get direct memory access...

Taking this out of 0.4.0 for now, as I don't think we want to rush into it. The cat's out of the bag for the ``parent_array`` etc, so let's take...

It's not obvious that this will do what it looks like it should do, numba could be doing something odd with the references. But, it looks good and I think...

I'm not particularly motivated tbh, there's a lot of critical path stuff we need to get done for the tskit paper first.

> One question: if you add chromosome labels, and edges on chromosome i are in [0,i-1), then do you plan to simplify separately by chromosome or change how edges are...

> From a forward simulation perspective, an alternative that seems more natural is to keep an array of table collections, and simplify each independently. Further, I then know immediately how...

> Certainly I think that we want to be able to restrict recent events such that multiple discontiguous regions (on different chromosomes) pass through a single individual ancestor. Essentially this...

Hmm, last point was wrong. The alternative would be: ```python for tree in ts.trees(start=chrs[22].start, end=chrs[22].end): for site in tree.sites(): pos = site.pos - chrs[22].start # Pos should now be mapped...

> I do like the idea of simplifying in parallel, though! Nothing stopping you doing it within fwdpp! Whichever way we end up encoding things for the canonical representation, internally...