From: Matt Hurd (matt.hurd_at_[hidden])
Date: 2005-08-22 17:39:46
>On 23/08/05, Teemu Torma <teemu_at_[hidden]> wrote:
> > On Monday 22 August 2005 03:59, Matt Hurd wrote:
> > I'm using 1.33. Is this a known bug or am I using the read_write
> > stuff incorrectly?
>
> I noticed same thing last week--a multithreaded program deadlocks with
> multiple threads waiting for write lock but no one holding it. It was
> not immediately clear to me why.
>
> This was reproduced on win32, linux and solaris, so it appears to be
> read_write_lock specific problem. Underlying primitives (mutex and
> condition) work fine on all platforms.
>
> The same code worked fine with boost 1.32.
>
> Teemu
Thanks for the confirmation Teemu.
Perhaps boost would be better off with a simpler but similar approach
totally rewritten to use posix on all platforms including win32.
Posix is viable on Mac these days I assume too. Support would be
easier I'd think. I'd be prepared to put an initial cut together. At
least Alexander Terekhov would be happy ;-) There is a certain
attractiveness is making posix do the heavy lifting...
A 90% solution that covers basic atomic ops, simple fencing and mutex
/ lock implementations better suited to policy implementations
wouldn't be too hard.
I guess at a minimum tests should detect if multithreaded apps
correctly fail when not protected so that the efficacy of other
concurrency testing is known. I use this approach in my testing where
it warns if unprotected concurrency works where it should fail.
matt.