Disable the local keyword in typing

[stedolan]

If the local_ keyword somehow sneaks into the parsetree even with the extension disabled (e.g. because the parsing occured in a ppx), then the typechecker should nonetheless reject uses of it.

Tested manually with a temporarily hacked lexer that lets the keyword through always, example output:

# let f () = local_ ref 42;;
Error: The local extension is disabled
       To enable it, pass the '-extension local' flag

[gretay-js]

I don't know anything about typing, but as far as this change goes, all it does is raise in a few places if Clflags.Extension.(is_enabled local) is false and a local attribute or extension is encountered, so it looks fine to me.

[xclerc]

I am certainly out of my depth here, but there are occurrences
of "extension.local" in "ocaml/typing/typecore.ml" (lines 2848,
2977, and 5778) where we do not check whether the extension
is enabled. I assume this will be caught by surrounding code, but
I am having a hard time proving it to myself.

It is also not clear to me whether something should be done
about "extension.escape".

[stedolan]

I am certainly out of my depth here, but there are occurrences of "extension.local" in "ocaml/typing/typecore.ml" (lines 2848, 2977, and 5778) where we do not check whether the extension is enabled. I assume this will be caught by surrounding code, but I am having a hard time proving it to myself.

It will. These three are in ancillary code that checks side-conditions (whether a function is annotated with local_ return, what the approximate type of a recursive value is, and whether an expression is a lambda, respectively). The code will always be typechecked by the main typechecking function after this, which has the check.

It is also not clear to me whether something should be done about "extension.escape".

Nothing! It's not a user-visible feature, it exists only temporarily during typechecking as a workaround for tricky compilation behaviour of optional arguments.

[xclerc]

Sorry, unrelated to this diff but maybe we can sneak it in.
I think in "parsing/lexer.mll", is_keyword should have a
condition to accept/reject local_. It may also be possible
to recycle Extension_not_enabled rather than add a new
constructor, but that is certainly not blocking.