Controllable labels, or a way to fold children into nodes

[gussmith23]

Here is my situation: my language uses "indirect" ops, e.g. instead of having (Add x y), my language instead uses (Op2 (Add) x y). When I visualize the egraph, it's not super useful to see a node labeled Op2 -- specifically, when the arrow for its first child (the Add) is lost in a jumble of other arrows.

I have my own local hack of egraph_serialize that allows users to specify a custom label_fn to generate labels. The function takes a node ID and the egraph, and produces a string. I can use the function to produce node labels like "Op2 Add", which is much more useful.

Is something like this already possible somehow? Would you be amenable to adding something like this?

[saulshanabrook]

I added a way to "inline leaves", which is used for this kind of visualization… It basically turns any pointers to unique children with no other parents and folds them into the class itself.

egraph-serialize/src/algorithms.rs

Lines 8 to 11 in f6aa7e7

	/// Inline all leaves (e-classes with a single node that has no children) into their parents, so that they
	/// are added to the function name like f(10, ·).
	/// Returns the number of leaves inlined.
	pub fn inline_leaves(&mut self) -> usize {

This combined with split classes (which is first used to make seperate classes say for each (add) instead of them all being as one) is how I implement this for egglog:

egraph-serialize/src/algorithms.rs

Line 134 in f6aa7e7

pub fn split_classes(&mut self, should_split: impl Fn(&NodeId, &Node) -> bool) {

Would this serve your purposes?