0007907: Specialization vs Indirections

View Issue Details [ Jump to Notes ]

[ Issue History ] [ Print ]

Project

Category

View Status

Date Submitted

Last Update

0007907

OCaml

middle end (typedtree to clambda)

public

2019-02-02 18:44

2019-02-05 11:04

Reporter

drup

Assigned To

Priority

normal

Severity

minor

Reproducibility

always

Status

new

Resolution

open

Platform

OS Version

Product Version

Target Version

Fixed in Version

Summary

0007907: Specialization vs Indirections

Description

Several people proposed to have functors that would easily introduce infix operators based on given primitives. That scheme is also used in the monomorphic library (https://github.com/kit-ty-kate/ocaml-monomorphic [^]) and a few others.

We already know that specialization of primitives and abstraction don't play particularly well together (https://caml.inria.fr/mantis/view.php?id=7440 [^]), but apparently, basic indirections are really bad too:

module Make (A : sig type t end) : sig
val (=) : A.t -> A.t -> bool
(* external ( = ) : A.t -> A.t -> bool = "%equal" (* SPECIALIZE IF UNCOMENTED *) *)
end = struct
external ( = ) : A.t -> A.t -> bool = "%equal"
end

module F = Make(Float)
let equal1 (a : float) b = F.(=) a b (* NOT SPECIALIZED *)
let equal2 (a : float) b = (=) a b (* SPECIALIZED *)

inlining often doesn't help at all with functors, but sometime does in less complex examples. The result is the same with clambda and flambda.

Tags

No tags attached.

Attached Files

Relationships

Notes
(0019581) stedolan (developer) 2019-02-05 11:04	Internally, this issue is the same as 0007440. When an external is exposed by a signature using "val" rather than "external", it is eta-expanded. In other words, the module: module Make (A : sig type t end) : sig val (=) : A.t -> A.t -> bool end = struct external ( = ) : A.t -> A.t -> bool = "%equal" end is internally converted to: module Make (A : sig type t end) : sig val (=) : A.t -> A.t -> bool end = struct external ( eq ) : A.t -> A.t -> bool = "%equal" let (=) a b = eq a b end This turns it into the same pattern as 0007440.

Issue History
Date Modified	Username	Field	Change
2019-02-02 18:44	drup	New Issue
2019-02-05 11:04	stedolan	Note Added: 0019581