Re: glib win32 condvar implementation

[Owen Taylor <otaylor redhat com>]

On Thu, 2006-08-24 at 22:10 +0300, Tor Lillqvist wrote:
> Alexander Larsson writes:
>  > Am I on crack, or is this totally broken?
> 
> I can't say, I haven't looked closely yet. But revelations/doubts like
> this sure makes you wonder whether the current gthread-win32.c should
> be scrapped, and we instead should just use gthread-posix.c on Win32,
> too, on top of pthreads-win32.
> 
> Reading the well-documented pthreads-win32 code one gets the
> impression of it really having been thoroughly designed (or evolved)
> to cater for all corner cases, being more reliable, and more
> thoroughly tested than gthread-win32.

On the other hand, Win32 threads are not pthreads, and cannot possibly
be 100% compatible with pthreads. After all, substantial kernel changes
were need on *Linux* to get threads fully compliant with POSIX for the
NPTL project.

There was at least some (small) attempt to make the GLib thread API 
more easily cross-platform than pthreads. So, pthreads-win32 might
go through a lot of trouble of emulation that we don't need in some
cases.

How big is pthreads-win32? Are you considering depending on it as
an external DLL/dependency, or incorporating it inside gthread.dll?

					- Owen