Tracking differences between melc and ocamlc, for cmi generation

[jchavarri]

This issue tracks the differences between melc internals and ocamlc. One of the goals is to be able to compile Melange cmi files directly with ocamlc —and maybe cmt—.

Note: strict removals of functionality in Melange / ReScript (like removal of Printf) do not affect this goal, as the idea is to build Melange code with OCaml compiler (not the other way around), so for now they are excluded from the list.

List (ongoing, non-exhaustive)

1. Melange externals can be defined with non-function types

Melange allows defining externals like external foo: string = "foo", which is invalid in ocamlc, e.g.:

File "node_modules/reason-react/src/legacy/ReactEventRe.rei", line 32, characters 27-31:
32 |   external preventDefault: unit = "preventDefault";
                                ^^^^
Error: External identifiers must be functions

Fixing this issue exclusively on Melange side (i.e. without changing the way ocamlc works) would involve breaking the ecosystem of bindings and libraries in some way.

2. Melange injects some marshalled-encoded info in externals

Melange injects some marshalled data into externals when processing the ast:

melange/jscomp/frontend/ast_external_process.ml

Line 1013 in d979086

pval_prim = [prim_name; External_ffi_types.to_string ffi];

This breaks when using ocamlc to generate cmis:

Values do not match:
    external bubbles : 'a synthetic -> bool = "bubbles" "BS:external"
is not included in
    external bubbles : 'a synthetic -> bool = "bubbles"
The native names of the primitives are not the same