deconstruction order atm must correspond exactly to the order as defined data =. obviously this is cancerous behavior
Alternatives in algebraic sum types can either be inlined in the main struct/union or allocated seperately and pointed to. Currently subtypes are always allocated seperately, but this should only happen for subtypes much bigger than their alternatives. types <= pointer size should always be inline
Find the last use points of local bindings and free them just after (is this always possible ? : probably yes since let-in expressions are easy to reason with. we can't release top level bindings unless the program is in the form let topbinds in main)
Lazy evaluation + graph reduction
llvm-hs-pure module level fresh name supply (mainly for emitting string globals)
Improve parser:
- properly handle indentation
Benchmarks