[_flux]

Yes, this would be very nice to have!

The workaround for a slightly different issue (similar as in using global state) I've used is to have an integer for identifier and then have a predicate that finds all the uses of those integers in the state and then picks the smallest integer not in the set of those ids. Now that I'm writing this I'm wondering how useful this was, though, even if it does avoid the GC step..

Maintaining that predicate is a bit cumbersome and error-prone in presence of multiple different kind of nodes in the system much like with GC, however, so a builtin magical id type would help a lot.

It seems that this approach of actually modifying the ids could be more effective in reducing the searh space, but will then introduce additional state transitions in the form of the GC step.

[colanderman]

Instead of integers you could use symmetric model values which you map to integers in a separate dedicated variable. Then it's easier to find the values in the other variables, and you don't have to mutate those other variables. (It's still hard to find them in e.g. the body of a function with infinite domain.)

(Of course with model values you must allocate them ahead of time. And with symmetry, the practical limit is 3-5 such values.)