Original bug ID: 6874
Reporter: @mmottl
Assigned to: @garrigue
Status: closed (set by @xavierleroy on 2016-12-07T10:49:25Z)
Resolution: fixed
Priority: normal
Severity: minor
Platform: Mac
OS: Mac OS X
OS Version: Yosemite
Version: 4.02.1
Target version: 4.03.0+dev / +beta1
Fixed in version: 4.03.0+dev / +beta1
Category: middle end (typedtree to clambda)
Monitored by: @yakobowski @mmottl

Bug description

Consider the following three implementations of essentially the same function:

module type ARG = sig val f : unit -> unit end

let foo_bad (module Arg : ARG) b =
  if b then Arg.f ()

let foo_good1 (arg : (module ARG)) b =
  let module Arg = (val arg) in
  if b then Arg.f ()

let foo_good2 b (module Arg : ARG) =
  if b then Arg.f ()

"foo_bad" generates a lot of unnecessary and expensive machine code whereas "foo_good1" doesn't. "foo_good2" just changes the argument order compared to "foo_bad" and generates equivalently efficient code as "foo_good1".

This seems to indicate that the compiler doesn't check whether subsequent function arguments (initializers for default arguments) actually refer to the unpacked module. The compiler then unnecessarily allocates closures to deal with that situation.

One could improve the compiler by introducing a check whether the module argument is being referred to by others. Maybe one can even solve this issue without any closure allocations in general.