Jerome Kelleher

Some options: - keep_coalescent_unary - keep_unary_that_coalesce - keep_locally_unary (as in, keep nodes that are locally unary, but with greater arity elsewhere)

See also discussion here: https://github.com/tskit-dev/tskit/discussions/2089#discussioncomment-2005450

> What about the other way around when TableCollection.tree_sequence is called? Do we keep that as a copy operation? For the first pass I think yes. The main thing we're...

We should make sure this change fixes #2080 while we're in here doing this type of stuff.

Revisiting this in the light of #2441 and recent updates (and rebasing on them), and it looks like it basically works so we should probably try and push it through....

Moving this to C 1.1 for now. Not essential for 1.0 release.

Yes, that's the sort of thing we'd be able to support. It's probably better to think of it as a function pointer, so we'd have something like ```python with open("ibd_segs.txt",...

I've change my mind about the ``time`` argument actually --- in practise we'd probably need the time for ``sample_a`` and ``sample_b`` as well for this to be generally useful, and...

Yeah, that's a fair point. I'd probably aim to keep things minimal though in the interest of clarity/simplicity at the C level, and keeping a reference to ``tables.nodes.time`` isn't too...

PR welcome!