Re: [Caml-list] Threads and "transaction isolation" in OCaml

[Markus Mottl <markus.mottl@gmail.com>]

On Mon, Aug 19, 2013 at 2:10 AM, Mark Shinwell <mshinwell@janestreet.com> wrote:
> I have wondered in the past if we should introduce a syntactic
> construct to delimit areas of code that rely on the no-thread-switch
> property.  I think it would probably be reasonable to actually
> check that the expression inside the construct was guaranteed under
> the current runtime not to cause a context switch (such expressions are
> likely to be, and should be, straightforward).  At runtime, the
> construct would have no semantic effect.  For the scenario described
> with bytecode, the construct could just cause a compiler error, at
> least at present.

This might just require adding two new primitives
(enter/leave_transaction) to the standard library.  If e.g. the native
code compiler sees that there is no allocation in-between, these could
even become no-ops.  Otherwise the runtime system might just need to
check one extra flag when deciding whether to do a context switch,
which would have negligible cost and apparently low complexity.  Of
course, using these primitives requires great care so maybe they
should be prefixed with "unsafe_".

Regards,
Markus

-- 
Markus Mottl        http://www.ocaml.info        markus.mottl@gmail.com