Original bug ID: 6782
Reporter: @alainfrisch
Status: acknowledged (set by @damiendoligez on 2015-02-18T22:57:54Z)
Resolution: open
Priority: normal
Severity: feature
Category: lexing and parsing
Monitored by: @yallop @hcarty

Bug description

Since method call binds more tightly than function application, it is inconvenient to chain method calls with arguments:

(((f x0) # m1 x1) # m2 x2) # m3 x3

which would be written "f(x0).m1(x1).m2(x2).m3(x3)" in many other languages.

We have been using the following syntactic variant at LexiFi for some time:

(f x0) # (m1 x1) # (m2 x2) # (m3 x3)

and I'd like to propose it for upstream inclusion.

This can be implemented by one extra rule in parser.mly:

| simple_expr SHARP LPAREN method_label simple_labeled_expr_list RPAREN
{ mkexp (Pexp_apply(mkexp(Pexp_send($1, $4)), List.rev $5)) }

(or perhaps some extra work so that the location of the Pexp_send node covers only "obj # (meth").

One counterargument is that it's a new use of parentheses, and they don't even correspond to an actual subtree of the AST. But I don't see a better syntax.

Note: #6583 is a proposal to let js_of_ocaml's new ppx support use the syntax "obj ## meth arg" for Javascript method invocation, and in Javascript, chaining method calls is quite common. The proposal is to consider ## as a custom operator, which means that "obj ## (meth arg)" could already be supported by the ppx, and Drup told me that he indeed plans to do so. Keeping the syntactic similarity between calls of Javascript methods and normal methods calls in OCaml is an extra argument for the present proposal about "obj # (meth arg)".