What if you had

purs compile --optimizer ./myoptimizer src/**/*.purs

where myoptimizer is a program, conceptually of type corefnJSON -> corefnJSON, that reads the original corefn from stdin and dumps the rewritten IR to stdout?

Some optimizations might be non local, requiring a signature [CoreFn.Module] -> [CoreFn.Module]. Using json, will open the additional possibility of writing plugins in purescript :).

Is this still something that people would be open to adding?

If so, I can write up a plan and start work on it.

You got two hearts, so it seems like it.

My initial plan is to implement purs codegen path/to/**/corefn.json path/to/**/more/corefn.json - which will resolve globs to get a list of files, decode the CoreFn.Module within each, and then run codegen on those modules. The outputted js files would go under output, namespaced under the appropriate module name.

I thought about not including the file globs argument, instead just assuming they will be in output/**/corefn.json, but I'd like to leave open the possibility of tools placing the corefn.json files wherever they see fit.

@colinwahl would having zephyr DCE in the compiler change the need for this?

Edit: My mistake. This would actually be a pre-requisite

I would say "not necessarily" 😉.

We'd still need to produce un-zephyr'd artifacts to have the standard purs compile work correctly for incremental compilation, so you'd only use it for a production build/pre-bundle step, and it would write them to a separate location than the standard output.