Disambiguate use of `trie` in variable names?

[SamuelBrand1]

The Trie struct looks pretty fundamental in the Gen codebase https://github.com/probcomp/Gen.jl/blob/master/src/trie.jl .

From reading the sources its a tree-structure, but maybe it should be documented?

[ztangent]

The Trie struct is internal and is only used in the implementation of Gen's dynamic modeling language, which is why it's not documented. ChoiceMaps are the main public interface for tree-shaped data, which are documented.

Confusingly, some of the code that calls choice_gradients seems to return a Trie, but actually returns a ChoiceMap (e.g. values_trie and gradient_trie are all actually choicemaps). So perhaps we should just the names of the variables that lead to that impression .

[SamuelBrand1]

Confusingly, some of the code that calls choice_gradients seems to return a Trie, but actually returns a ChoiceMap (e.g. values_trie and gradient_trie are all actually choicemaps). So perhaps we should just the names of the variables that lead to that impression .

Right! This is what was foxing me.

[SamuelBrand1]

I've changed the title of this issue to point at that. This feels like something an LLM or even just a regex could have a shot at doing with moderate high probability of success... do you have a favoured variable name style e.g. just drop _trie postfix, change to _map postfix?

[ztangent]

I think we should probably follow the documentation for choice_gradients and change values_trie to choice_values and gradient_trie to choice_grads: https://www.gen.dev/docs/stable/ref/core/gfi/#Gen.choice_gradients

This doesn't appear very many times in the codebase, so I think a regular find and replace will be enough.