Pattern type constraint in let expression parsed inconsistently

[vicuna]

Original bug ID: 7758
Reporter: antron
Status: new
Resolution: open
Priority: normal
Severity: minor
Version: 4.06.1
Category: lexing and parsing
Monitored by: @nojb @Drup @ygrek

Bug description

The expression

(1) let foo : int = 42 in ...

seems to get parsed as

 `let foo : int = 42 : int in ...`

while the expression

(2) let (foo : int) = 42 in ...

which differs only in having parentheses, seems to get parsed as

  `let foo : int = 42 in ...`

I would have expected the second parse for both cases. The first parse can be problematic when using PPX, because the variable and expression may need to have different types. In particular, in

  `let%lwt foo : int = bar in ...`

foo should have type int, and bar should have type int Lwt.t. See ocsigen/lwt#564.

Steps to reproduce

1. source.ml:

let () =
  let foo : int = 42 in
  let (foo : int) = 42 in
  ignore foo

2. ocamlopt -dparsetree -c source.ml

3. Result is:

[
  structure_item (source.ml[1,0+0]..[4,59+12])
    Pstr_value Nonrec
    [
      <def>
        pattern (source.ml[1,0+4]..[1,0+6])
          Ppat_construct "()" (source.ml[1,0+4]..[1,0+6])
          None
        expression (source.ml[2,9+2]..[4,59+12])
          Pexp_let Nonrec
          [
            <def>
              pattern (source.ml[2,9+6]..[2,9+20]) ghost
                Ppat_constraint
                pattern (source.ml[2,9+6]..[2,9+9])
                  Ppat_var "foo" (source.ml[2,9+6]..[2,9+9])
                core_type (source.ml[2,9+6]..[2,9+20]) ghost
                  Ptyp_poly
                  core_type (source.ml[2,9+12]..[2,9+15])
                    Ptyp_constr "int" (source.ml[2,9+12]..[2,9+15])
                    []
              expression (source.ml[2,9+6]..[2,9+20]) ghost
                Pexp_constraint   <----------------------------- !!!!
                expression (source.ml[2,9+18]..[2,9+20])
                  Pexp_constant PConst_int (42,None)
                core_type (source.ml[2,9+12]..[2,9+15])
                  Ptyp_constr "int" (source.ml[2,9+12]..[2,9+15])
                  []
          ]
          expression (source.ml[3,33+2]..[4,59+12])
            Pexp_let Nonrec
            [
              <def>
                pattern (source.ml[3,33+6]..[3,33+17])
                  Ppat_constraint
                  pattern (source.ml[3,33+7]..[3,33+10])
                    Ppat_var "foo" (source.ml[3,33+7]..[3,33+10])
                  core_type (source.ml[3,33+13]..[3,33+16])
                    Ptyp_constr "int" (source.ml[3,33+13]..[3,33+16])
                    []
                expression (source.ml[3,33+20]..[3,33+22])
                  Pexp_constant PConst_int (42,None)
            ]
            expression (source.ml[4,59+2]..[4,59+12])
              Pexp_apply
              expression (source.ml[4,59+2]..[4,59+8])
                Pexp_ident "ignore" (source.ml[4,59+2]..[4,59+8])
              [
                <arg>
                Nolabel
                  expression (source.ml[4,59+9]..[4,59+12])
                    Pexp_ident "foo" (source.ml[4,59+9]..[4,59+12])
              ]
    ]
]

Additional information

Further complicating this in practice is that in 4.05.0, (1) is parsed as

  `let foo = 42 : int in`

i.e. the constraint is removed from the pattern and appears on the expression only.

[vicuna]

Comment author: @Drup

It would be nice to sort this out before the next release, as it messes up with ppx in a way that is difficult to take into account, even with ocaml-migrate-parsetree.

If someone has a concrete proposition on how to make this cleaner, I'm ready to implement it.

[vicuna]

Comment author: @alainfrisch

See https://blog.janestreet.com/plans-for-ocaml-408/, "https://blog.janestreet.com/plans-for-ocaml-408/":

We plan to change the AST to have an explicit representation for:...

FWIW, I'm not 100% convinced by the proposal, since it would mean treating very differently

let (x1 : t), x2 = e1, e2 in ...

from

let (x1 : t) = x1 in ...

which might be surprising for ppx authors and for users (in the first case, the annotation would not be propagated to the expression e1).

An alternative would be to keep the annotations only in the pattern at the AST level, but somehow extract them (with a strategy to define) and apply them to the bound expression.

[github-actions]

This issue has been open one year with no activity. Consequently, it is being marked with the "stale" label. What this means is that the issue will be automatically closed in 30 days unless more comments are added or the "stale" label is removed. Comments that provide new information on the issue are especially welcome: is it still reproducible? did it appear in other contexts? how critical is it? etc.

[garrigue]

I vote to close this. Here parentheses are relevant.

[gasche]

@garrigue there are different things here. One is a special status of let p : t = .., but the other is whether the propagation of type information happens through a syntactic desugaring in the parsetree, or by some later means.

We can decide that we like or not the special typing discipline in the let p : t case (vs. let (p : t)), but I think it is rather consensual than having this as a parse-time desugaring is not a good approach, especially if it makes the life of ppx authors harder.

[github-actions]

This issue has been open one year with no activity. Consequently, it is being marked with the "stale" label. What this means is that the issue will be automatically closed in 30 days unless more comments are added or the "stale" label is removed. Comments that provide new information on the issue are especially welcome: is it still reproducible? did it appear in other contexts? how critical is it? etc.

[garrigue]

First, to clarify, originally the two syntaxes were already treated differently.
I.e., before the introduction of polymorphic recursion, (1) was indeed parsed as

let foo = (42 : int) in

which is coherent with the handling in functions:

let foo x : int = x+1

Now, for ppx's, wouldn't it be sufficient to disable the duplication of the type into the pattern when there is an extension?
This would make sense, since the semantics of let is changed, so that x is no longer guaranteed to be of type int (while 42 ought to be).

[github-actions]

This issue has been open one year with no activity. Consequently, it is being marked with the "stale" label. What this means is that the issue will be automatically closed in 30 days unless more comments are added or the "stale" label is removed. Comments that provide new information on the issue are especially welcome: is it still reproducible? did it appear in other contexts? how critical is it? etc.