Re: the state of the concurrency branch

[Paul Eggert <eggert@cs.ucla.edu>]

On 08/27/13 11:23, Tom Tromey wrote:
> Could you say what is wrong with it as it stands?

It's that there are two sets of functions
where there needs to be just one.  Where there's a need for a
separate level for the Lisp layer (e.g., the stuff in sys_thread_create)
that should be kept.  But there's no need to distinguish (say)
sys_thread_equal from pthread_equal.

If there's a need for Emacs to port to OS threads that are not
compatible with pthreads, we should use GNU Pth, rather than
reinventing the wheel.

> If you're concerned about the layer of indirection... well, first, don't
> be, it hardly matters; but otherwise, you can just inline most of the
> stuff in systhread.h if you really care.  That would be ok with me.

It's not the inefficiency I'm worried about, it's just that the extra
layer makes the code harder to read and understand, and doesn't
buy us anything.  But I hear what you're saying about the system-dependent
bits and will try to preserve that in the proposed patch.

> I presume the modified code will still leave a place for a
> non-pthreads implementation, right?

Sure, platforms without pthreads will still work.