[Caml-list] empty mli for executable

[Caml-list] empty mli for executable

[Milan Stanojević]

ocaml compiler executables have empty mlis with the following comment
(*
  this "empty" file is here to speed up garbage collection in ocamlopt.opt
*)

Can someone explain this in more detail?
How does empty interface lead to faster gc-ing?

Thanks,
     Milan

Re: [Caml-list] empty mli for executable

[Alain Frisch]

On 10/03/2013 06:50 PM, Milan Stanojević wrote:
> ocaml compiler executables have empty mlis with the following comment
> (*
>    this "empty" file is here to speed up garbage collection in ocamlopt.opt
> *)
>
> Can someone explain this in more detail?
> How does empty interface lead to faster gc-ing?

The empty interface implies that toplevel values computed by those 
modules don't need to be stored in the global symbol slots.  Those slots 
are roots for the GC and are thus scanned every time it runs.  The fewer 
global symbols, the quicker the GC will be.  Well, this is the theory, 
I'd be really surprised if this materialized in a speed up  which can be 
observed.

Having an empty interface has a much more practical interest, in 
general: it allows the compiler to detect unused declarations in the 
implementation (if the corresponding warnings are enabled), and thus to 
reduce code rot.


-- Alain

Re: [Caml-list] empty mli for executable

[Milan Stanojević]

> The empty interface implies that toplevel values computed by those modules
> don't need to be stored in the global symbol slots.  Those slots are roots
> for the GC and are thus scanned every time it runs.  The fewer global
> symbols, the quicker the GC will be.  Well, this is the theory, I'd be
> really surprised if this materialized in a speed up  which can be observed.

Thanks, Alain.
Just one question about this. Let's say I have a toplevel value in my
executable and an empty mli. How is this toplevel value then retained?
Is it part of some other root set ?


> Having an empty interface has a much more practical interest, in general: it
> allows the compiler to detect unused declarations in the implementation (if
> the corresponding warnings are enabled), and thus to reduce code rot.

Yes, I've been doing this, it is a very useful feature of the compiler.

Re: [Caml-list] empty mli for executable

[Alain Frisch]

On 10/3/2013 8:42 PM, Milan Stanojević wrote:
> Thanks, Alain.
> Just one question about this. Let's say I have a toplevel value in my
> executable and an empty mli. How is this toplevel value then retained?
> Is it part of some other root set ?

Well, actually, after double-checking, it appears that in native code, 
the non-exported values are still stored in the global record 
corresponding to the unit.  See this comment in translmod.ml:

    Identifiers that are not exported are assigned positions at the
    end of the block (beyond the positions of all exported idents).


This is just one possible compilation strategy, designed to reduce 
pressure on registers (and it invalidates my explanation).  Another one 
would be to simply consider those value as local variables (in 
registers, maybe spilled to the stack).

The current strategy has the effect that a non-exported toplevel value 
is never reclaimed by the GC, even if it is no longer accessible (this 
is not the case in bytecode).


Alain