[ocaml-ctypes] ocaml as callback to C code with event loop

[Jeremy Yallop]

On 11 May 2017 at 10:20, Serge Sivkov <ssp.mryau at gmail.com> wrote:
> thank for your reply, it allow me partially resolve problem: code works on
> x86 without any problems (yet with Gc.compact() calls insert after each line
> of code). But ARM version allways raises CallToExpiredClosure on the first
> call to callback.
> I've tried next variants to store callback if struct field:
> - existing toplevel function by name
> - store function name to toplevel ref and save it to struct field by
> !ref_name
> - by register closure to function (fun a b c d -> top_level_function a b c
> d) with Ctypes.Root.create
> - by store same closure to Hashtbl.
> - by register finaliser with Gc.finalise for closure (and it has not been
> called)
> All this variants work on x86 and do not work on ARM, i.e. first call to
> OCaml callback from C raises CallToExpiredClosure.
> It seems I don't understand problem at all.
>
> So here is base questions:
> - in case I store in C struct field existing toplevel function by name, may
> it be GC'ed?

Yes.  The only way to prevent a value from being collected is to
ensure that it's visible to the GC, which requires keeping a reference
to it in an OCaml object.  The GC doesn't scan C memory, such as
struct fields, so storing a function in a struct field won't stop it
being collected.

> -what may cause CallToExpiredClosure raise too if I has only one callback to
> call and at C side this struct in GDB seems to be correct?

If there's an OCaml reference to the function passed to C then
CallToExpiredClosure should never occur.  But it has to be exactly the
same function -- for example, storing a function in a struct field,
then reading from the struct field will create a new copy of the
function, which won't prevent the original from being collected.

If you have a (smallish) running example somewhere of the code that's
causing problems I'd be happy to take a look.

Kind regards,

Jeremy.