From what I can tell, incremental computation is starting to be used in practice:

Jane St has an OCaml library called Incremental:

"Data Driven UIs, Incrementally" by Yaron Minsky at Strange Loop 2018: https://www.youtube.com/watch?v=R3xX37RGJKE

This talk has a pretty impressive live JS table demo! It looks faster and nicer than most JS UIs I've seen.

-----

And the Skip language was developed Facebook. Apparently they started to use it for some production stuff, but somehow they decided against it and spun it out:

http://skiplang.com/

https://news.ycombinator.com/item?id=18077612

Julien Verlaguet - Reflex: Reactive Programming at Facebook -- https://www.youtube.com/watch?v=AGkSHE15BSs

Reflex was an earlier name or earlier incarnation of Skip.

Adapton's idea is based on self-adjusting computation [1], which is considered pretty mature [2].

In the field there are also approaches like incremental lambda calculus [3] and differential dataflow [4]. Both of them focuses on propagating differences (changes) instead of invalidating dirty nodes in a dependency graph.

[1] https://www.semanticscholar.org/paper/Self-adjusting-computa...

[2] https://web.archive.org/web/20170628103354/lambda-the-ultima...

[3] https://www.semanticscholar.org/paper/A-theory-of-changes-fo...

[4] https://www.semanticscholar.org/paper/Differential-Dataflow-...

I’m not sure how practical this kind of generalized approach is, but notice how incremental computation is already all around you: in your build system, in the way React recomputes its tree and your browser recomputes layout, in any pull-based protocol where the server keeps state.

Didn't yet go through the work posted, though at multiple investment banks the pricing/risk engines make use of _compute graphs_ which have a similar execution style. Only recompute nodes that get invalidated on change in input data and propagate till the output.