Re: [Caml-list] GC assumptions ?

[Pierre Chambart <pierre.chambart@ocamlpro.com>]

On 17/01/2016 12:01, Gerd Stolpmann wrote:
> Am Samstag, den 16.01.2016, 22:42 +0100 schrieb Nicolas Ojeda Bar:
>> Hi list,
>>
>>
>> I am toying with the idea of using the OCaml runtime for a dynamically
>> typed language targeting Lambda.  I was wondering what the OCaml
>> runtime (ie the GC) assumes about the language being compiled.
>>
>>
>> In particular I was wondering if it assumes that a value cannot go
>> from pointing to a block to being an int or vice-versa.
> This possibility exists in OCaml, e.g.
>
> type t = Block of bool | Int0
>
> Now, an assignment x := Block false; x := Int0 would do the change you
> are thinking about (argument-less variants are represented as ints).
>
> Gerd
>
Note that in that case, the value is built as an Const_pointer instead of
an integer. It announce to the compiler that this value could contain a
block later.

But the result of integer returning primitive is assumed to always be an
integer, in particular, it is not safe to do something like: (imagine the
corresponding lambda code).

let f x =
  let r = ref (x + 1) in (* r is local mutable variable after simplify *)
  r := Obj.magic (x,x);
  call_a_function ();
  !r

Since r is considered as containing an integer, it is not registered in
the frame descriptors as a potential GC root.

This is not the only place I think that assumes that, so you should
be quite carefull.
-- 
Pierre