I believe Aaron Hsu's work on tree processing for Co-dfns [0] is a novel example of tree handling - a data parallel compiler for efficient compilation on a GPU using extensive tree-oriented "nanopass" transformations [1].

[0] https://github.com/Co-dfns/Co-dfns

[1] https://dyalog.tv/Dyalog18/?v=hzPd3umu78g

Holy crap, that video is mind-blowing. Thank you! I was vaguely aware of Co-dfns before, but now it's squarely on my radar.

If you have cycles in your graph, you can build a tree out of a vector of indices, or as a table of parent/child node keys. The former is what John Scholes does[1], and the latter has plenty of common SQL examples.

[1]: https://youtu.be/DsZdfnlh_d0

If you have a non-cyclic tree (e.g. a 256-trie) you can nest it directly: @/ will traverse to nodes, and COW keeps updates from trashing memory too much.

However trees are rarely the best data structure: Binary search on a sorted list can beat b-trees simply because the end of the search will already be in cache, so for a routing table of IP address ranges, I use something like:

    ipStart where ipStart bin x

John Earnest also details how trees are handled in K (an array based language invented by Arthur Whitney). https://github.com/JohnEarnest/ok/blob/gh-pages/docs/Trees.m...

I also wrote a paper about recursive combinators which touches on these ideas a bit: http://archive.vector.org.uk/art10501740

And I translated some of Aaron Hsu's APL ideas into k5. With a little massaging they work just as well: https://github.com/JohnEarnest/ok/blob/gh-pages/examples/key...