Re: [Caml-list] Re: Haskell vs OCaml

[Chet Murthy <murthy.chet@gmail.com>]

On Saturday, March 23, 2013 01:02:51 AM Daniel Bünzli wrote:
> Le samedi, 23 mars 2013 à 00:47, Chet Murthy a écrit :
> > [short answer: because each time that a paying customer is affected by
> > a buggy driver, the vendor of the CLR or JVM spends -serious- money
> > finding and shooting that bug. They spend -serious- money building
> > internal tooling to help them do this. It's expensive and
> > time-consuming, and furthermore consumes -extremely-skilled- people.
> > No OSS language runtime can afford that.]
> 
> I can't speak for what happens with CLR or JVM, but if you bind to OpenGL
> from OCaml you are basically programming with OpenGL's C interface. There's
> no direct interaction between the OpenGL drivers and OCaml's language
> runtime. So I don't think your remark makes sense in that context.

For most C drivers I've seen, the same is true in the JVM.  The
"direct interaction" consists in manipulating/allocating some
heap-objects, and .... stray pointers.  Obv. it's the latter that
cause all the problem.

For example, long ago a (POSIX-threaded) JVM used signals to stop
threads in order to do mark-and-sweep GC.  Unfortunately, there were a
number of C drivers that were not "signal-safe" (didn't restart their
syscalls properly), and so you'd get "interesting" segfaults at GC
time.  The solution involved some hacking to avoid sending signals to
threads in C driver code, but still stopping them when they emerged
(or called code in the JVM, e.g. to allocate heap-memory).

--chet--