Excerpts from Gabriel Kerneis's message of Tue Jul 31 18:55:05 +0200 2007:
> Hello,
> 
> I've got a little problem with camlp4. The following works perfectly
> well :
> #####################################################################
> ~/ocaml/camlp4% cat pa_bind.ml
> module Id = struct
>   let name = "pa_bind"
>   let version = "$Id:$"
> end
> 
> module Example (Syntax : Camlp4.Sig.Camlp4Syntax) = struct
>  include Syntax
> 
>  EXTEND Gram
>  expr: LEVEL "top"
>    [[ t=expr; "foo"; f=expr ->
>         <:expr<
>             Lwt.bind $t$ (fun () -> $f$)   
>         >>]];  
>  END
> end
> 
> module M = Camlp4.Register.OCamlSyntaxExtension(Id)(Example)
> 
> ~/ocaml/camlp4% ocamlc -c -pp camlp4of -I +camlp4 pa_bind.ml 
> 
> ~/ocaml/camlp4% cat test_bind.ml 
> let r = (Lwt_unix.sleep 5) foo (print_endline "42")
> 
> ~/ocaml/camlp4% camlp4of -parser pa_bind.cmo test_bind.ml 
> let r = Lwt.bind (Lwt_unix.sleep 5) (fun () -> print_endline "42")
> 
> #####################################################################
> 
> But if I change "foo" to ">>" (which is the usual way to write an
> anonymous bind for monads), it doesn't work anymore. I first thought the
> reason was that ">>" is also the symbol for closing a quotation, but
> trying "++" instead proved useless as well.
> 
> I'm certainly not a camlp4 guru so... what am I missing?

When  you  use  "foo",  you  create a new keyword. You can therefor choose its
priority.  However  the  parsing  of  ">>"  is predefined since OCaml supports
user-defined  operators.  In  your example using t=expr is too greedy since it
contains  ">>"  in  the infixop0 rule.

However  if  you  use  SELF instead of expr it works fine. Indeed SELF handles
the  left  recursion and don't blindly parse an expr but remember that ">>" is
waited.

...
EXTEND Gram
expr: LEVEL "top"
  [[ t = SELF; ">>"; f = SELF ->
       <:expr<
           Lwt.bind $t$ (fun () -> $f$)   
       >>]];  
END
...

Best regards,

-- 
Nicolas Pouillard aka Ertai