More featureful define-inline

[usaoc]

Recent experiment of deforestation in Qi (see https://racket.discourse.group/t/todays-qi-meeting-notes/2323 for the context) points out that define-inline can be quite useful in a higher-order combinatory context, where the functions (serving as combinators) are always expected to be inlined and further optimized. There are things that it can do better:

1. The support for case-lambda is lacking;
2. Variable-reference case cannot be inlined.

I’m interested in looking into this at some point, so this issue serves as a note.

[samth]

For references as a variable, are you thinking of having it inline the lambda with the body? An alternative would be to have an expansion-time API that could get the function definition, that Qi could use as needed.

[usaoc]

Yes, that’s the straightforward way (“inline the lambda with the body”). Ideally, the change should not change the behavior of existing code, say by requiring an #:aggressive directive (and the programmer’s care!). Other approaches are welcome, though.