Re: [Caml-list] extending records with Obj.magic

[Goswin von Brederlow <goswin-v-b@web.de>]

Martin Jambon <martin.jambon@ens-lyon.org> writes:

> Nicolas Ojeda Bar wrote:
>> Hello,
>> 
>> I need extensible records, and the code below seems to
>> work. Are there any pitfall that I should be aware of?
>> Could this mess up the GC?
>> 
>> # type t0 = { a : int };
>> # type t1 = { a2 : int; b : int };
>> # value x1 = { a2 = 3; b = 5 };
>> # value x0 : t0 = Obj.magic x1;
>> value x0 = { a = 3 }
>> # value x0' : t1 = Obj.magic x0;
>> value x0' = { a2 = 3; b = 5 }
>> 
>> (supposedly t1 is an extension of t0). The types are
>> being generated by a program, so I am not worried about
>> actually having to _write_ this myself.

You need to make sure t1 is structurally an extension of t0 and then it
will be safe to use in the current implementation of the GC (see counter
example below). As others have said beware of functions that do
structural compares on the magiced values. But GC compaction will keep
the value intact as it only looks at the in-memory structure and
Obj.magic doesn't change that.

> # type t0 = { a : float };;
> type t0 = { a : float; }
> # type t1 = { a2 : float; b : int };;
> type t1 = { a2 : float; b : int; }
> # let x1 = { a2 = 3.; b = 5 };;
> val x1 : t1 = {a2 = 3.; b = 5}
> # let x0 : t0 = Obj.magic x1;;
> val x0 : t0 = {a = 3.}
>
> And now the magic:
>
> # x0.a;;
> - : float = 6.94364726476075e-310

This is a special case because records with ONLY floats are stored as
unboxed float array. Mixed records on the other hand use boxed
floats. So in you example t0 is structurally not a prefix of t1.

MfG
        Goswin