Re: [Caml-list] recursive mutexes in ocaml

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

Xavier,

Wow, that's excellent.  Made my morning!  I've implemented Java's
monitors, so yeah, they're as you describe, and I cannot imagine doing
it any other way.  What were they -thinkin'- in the POSIX committee?

Ah, well.

OK.  I'll just add the (few) lines of code needed to get basic
recursive mutexes for locking purposes, and leave it at that.  Given
Butenhof's analysis, I -fully- agree with leavin' 'em out of ocaml.
Yikes!

--chet--

On Wednesday, July 17, 2013 07:33:41 PM Xavier Leroy wrote:
> Hi Chet,
> 
> > I figured I'd ask -why- ocaml's mutexes aren't recursive.  Or at
> > least, why there isn't an option for recursive mutexes?
> 
> Dave Butenhof, one of the main authors of POSIX threads, makes rather
> strong points against recursive mutexes:
> http://tinyurl.com/butenhof-recursive-mutexes
> 
> The killer issue in my opinion is how to make recursive mutexes play
> well with pthread_cond_wait() / Condition.wait: POSIX's
> pthread_cond_wait() only releases the mutex once, which leads to all
> sorts of deadlock if it's been acquired N>1 times.  I believe Java
> has a more useful behavior, whereas the mutex is released N times,
> then reacquired N times.  But this cannot be implemented easily on top
> of POSIX threads.  And don't get me started on Win32 threads...
> 
> It might make sense to have recursive mutexes as a separate type,
> different from ordinary mutexes, so that it's not usable in
> conjunction with Condition.wait.  Apparently, Batteries has/had that.
> 
> - Xavier Leroy