Extending open

[trefis]

This is a refresh of #1506 which includes the restrictions that were discussed during the last developer meeting and since then on github (cf. #1506 (comment)).

More precisely:

• open in signatures is changed to accept extended module paths
• open in structures, as well as let open in expressions, are extended to accept any module expression
• let open in classes and class types is unchanged: before extending it we would need to add support for let module, which is complicated. Cf. the discussion on MPR#6271
• the short syntaxes for local module opens (e.g. M.(expr), M.(pat), M.[expr], etc.) are still restricted to module paths.

Things to note:

• I had to update Rec_check, but I didn't spend to much time on it, and it is untested. So it's likely to be wrong. It'd be nice if @gasche or @yallop could have a look.
• I imported the tests from Extending open to accept arbitrary module expression #1506 as they were, and updated the output in a separate commit, so it's easier to spot the differences between the two PRs
• the manual still needs to be updated

[yallop]

Thanks, @trefis. I plan to take a look at this soon (but not before next week).

[lpw25]

I've looked over the translation parts so far. Looks good but there are a couple of bugs and some places where we could avoid allocating in the open struct ... end case.

[lpw25]

There are a couple of bugs, and a couple of bits that should be tidied up. Once they're fixed this should be good to go.

[Octachron]

On the documentation front, I have a subpr here that adds some basic explanations and examples for the new features, if people wish to discuss the documentation separately.

[trefis]

I have rebased, answered all of @lpw25's comments, and included @Octachron's changes to the manual.

[trefis]

I pushed one last commit which updates the manual as discussed.
I think this is now ready to go.

[yallop]

I had to update Rec_check, but I didn't spend to much time on it, and it is untested. So it's likely to be wrong. It'd be nice if @gasche or @yallop could have a look.

Is the following intentional?

# let rec x = let module M = struct let y = lazy x end in ();;
val x : unit = ()
# let rec x = let module M = struct include struct let y = lazy x end end in ();;
val x : unit = ()
# let rec x = let module M = struct module M = struct let y = lazy x end end in ();;
val x : unit = ()
# let rec x = let module M = struct open struct let y = lazy x end end in ();;
Line 1, characters 12-74:
1 | let rec x = let module M = struct open struct let y = lazy x end end in ();;
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Error: This kind of expression is not allowed as right-hand side of `let rec'

[trefis]

I guess it is:

# let rec x = let module M = struct include (struct let y = lazy x end : sig end) end in ();;                                                                       
Line 1, characters 12-89:
1 | let rec x = let module M = struct include (struct let y = lazy x end : sig end) end in ();;
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Error: This kind of expression is not allowed as right-hand side of `let rec'

I guess you might want to make modexp a bit cleverer when looking at Tmod_constraint.

[lpw25]

From the CI, it looks like ocamlnat needs updating.

[objmagic]

just curious, do we have an official code style for ocaml compiler source code now? are we using ocamlformat?

[trefis]

No.

[objmagic]

should we iter _od here? also we should iter Tstr_open?

[trefis]

I updated this and Tstr_open to do the same thing as with includes.

[objmagic]

should we map over module expression of od here? Also Tstr_open

[trefis]

Updated.

[objmagic]

can we list out all other cases here explicitly?

[trefis]

I just moved this code from Translmod to Translcore. I didn't actually write it. So I'd rather not update it.

[objmagic]

I know this is copied over from another file, but I'd like to know what "manifest primitives" means. Thanks.

[trefis]

I think reading this as "externals" is acceptable.

[objmagic]

add one whitespace before let

[trefis]

Done.

[lpw25]

Good to merge once CI passes.

[yallop]

Thanks, @trefis!

[objmagic]

Thank you for the work @trefis!

[lpw25]

There's a bug in this PR. is_nonexpansive in typecore.ml needs to be extended to handle Texp_open. We should also add a test for this case to the testsuite, and maybe not use a fragile match for that function.