Menhir:How to use inline to add locations?

Learning
1

I saw this trick from Your favorite Menhir tricks and fanciness , which is very impressive (cc @EmileTrotignon ):

%public %inline located(X):
x=X { Position.with_poss $startpos $endpos x }

I tried to make a demo to test how to use it, and there’re not so many examples outside the world, except this new syntax one cc @fpottier :

let located(x) ==
  ~ = x; { { loc = $loc; value = x } }

let atomic_expr :=
  | LPAREN; ~ = expr; RPAREN; <>
  | located(
    | ~ = INT; <ELiteral>
    | ~ = unary_op; ~ = atomic_expr; <EUnOp>
    )

I was wondering is there any way to make the located thing work in the old syntax?

Here’re my code:

syntax.ml

type prim2 = Plus | Minus | Times
type 'a located = { loc : Lexing.position * Lexing.position; value : 'a }

type lexpr = raw_expr located
and raw_expr = Num of int64 | EPrim2 of prim2 * lexpr * lexpr

lexer.mll

{
open Parser
}
let whitespace = [' ' '\t' '\n' '\r' ]
let digit = ['0'-'9']
let number = digit+

rule read =
  parse
    | whitespace { read lexbuf }
    | number as lxm { NUMBER (Int64.of_string lxm) }
    | '(' { LPAREN }
    | ')' { RPAREN }
    | '+' { PLUS }
    | eof { EOF }

parser.mly


%{
open Syntax
%}
%token <int64> NUMBER
%token LPAREN
%token RPAREN
%token PLUS
%token EOF

%start <lexpr> start

%%

start:
| e = expr EOF { e}

expr:
| n = NUMBER { Num n }
| LPAREN e=expr RPAREN {e}
| e1=expr PLUS e2=expr { EPrim2 (Plus, e1, e2)}


%inline located(X):
| x=X {{ loc = $loc; value = x }}

I mean, what’s the appropriate way to add located in the expr rule?

I did tried a lot of ways, like:

expr:
| n = NUMBER { Num n }
| LPAREN e=expr RPAREN {e}
| e1=located(expr) PLUS e2=located(expr) { EPrim2 (Plus, e1, e2)}

But got error

Error: This expression should not be a constructor, the expected type is Dune__exe.Syntax.lexpr
2

It seems to me that the error you have here is related to your ocaml code, and that the menhir part is correct.
Reading more carefully, start is supposed to return a lexpr but actually return a raw_expr, you should also add located to it. If this is not correct, having the location of the error would be quite helpful.

1 Like
3

There is a big example using this trick in the menhir sources : benchmarks/parsers/houblix/parser.mly · master · POTTIER Francois / menhir · GitLab
This is a parser for an ML language using a syntax more similar to reason than to OCaml.

1 Like