Re: [Caml-list] C interface, threads and ocaml callbacks

["Daniel Bünzli" <daniel.buenzli@epfl.ch>]

Le 5 mars 04, à 03:13, annan@hushmail.com a écrit :

>   If I understand what you are trying to do, most likely you can't do
> it.  If you build ocaml code with -threads, you like with the thread-
> aware basic library.  Once you do this, you can't use threads you 
> created
> in the C execution context, as there is information required by Ocaml
> when executing library functions that gets created when *IT* uses 
> threads,
>  that isn't there when C creates threads.
>
>   So if you create all of your threads in Ocaml, then you can use 
> -threads,
>  but if C creates threads, those threads cannot execute OCaml code.

My understanding of the -thread flag --- drawn from message [1] --- is 
that when you compile with it, then you just get this master lock on 
the runtime that you can release/get by using 
enter/leave_blocking_section functions. Though the initial purpose of 
the latter functions was --- it seems --- to give the opportunity to 
ocaml signal handlers to execute during long running C code that 
doesn't interact with ocaml, it seems that you can also use this 
mechanism to serialize concurrent callbacks from C into ocaml (by 
wrapping the callbacks around leave/enter_blocking_section to 
get/release the master lock).

>   You can have C threads execute inside Ocaml, if you don't use the 
> OCaml
> threading library.
This seems match what is described in that message [2], though --- due 
to my lack of knowledge about the internals of ocaml's threading 
libraries --- I don't really understand the reason.

> So if you can get away with just one OCaml thread,
>  but need multiple C threads, then gatekeep the entry of C threads into
> the OCaml code (ie have the OCaml code go through a C function that 
> pthread_cond_waits
A lock/unlock on a mutex around the callbacks from C into caml should 
be sufficent, hence the use of -thread and 
enter/leave_blocking_section.

Unfortunately I hadn't any time these days to experiment the approach I 
described in my previous message.

Thanks for you answer.

Daniel

[1] <http://caml.inria.fr/archives/200106/msg00199.html>
[2] <http://caml.inria.fr/archives/200308/msg00258.html>

-------------------
To unsubscribe, mail caml-list-request@inria.fr Archives: http://caml.inria.fr
Bug reports: http://caml.inria.fr/bin/caml-bugs FAQ: http://caml.inria.fr/FAQ/
Beginner's list: http://groups.yahoo.com/group/ocaml_beginners