Reuse of Thread IDs

[Ross Johnson <rpj@callisto.canberra.edu.au>]

Hi all,

Nicolas Barry has discovered that a couple of routines 
(pthread_getschedparam etc) had rather iffy thread validation checks 
that could ultimately segfault due to pthread_t being a simple pointer 
to alloced memory that was being freed on thread exit. This weakness in 
the library hasn't been entirely unanticipated though, but it's time has 
arrived.

These routines also returned an incorrect error value (EINVAL instead or 
ESRCH).

As a result, I've changed the way that threads are managed slightly.

The validation check is now stronger and goes as far as checking the 
validity of the underlying Win32 thread HANDLE stored in the POSIX 
thread handle. To make work, the library (as of the current CVS version) 
now no longer frees pthread structs when a thread exits. Instead it 
pushes disused structs onto a LIFO stack (implemented as a simple 
guarded linked list). Thread handles are still calloced if no disused 
handles are available.

There is now also an exported pthread_kill() routine, but it only 
accepts a zero value for the signal arg. Any other value will return 
EINVAL. It's intended to be used by applications that want to test the 
validity of a POSIX thread handle (as intended by the standard).

I've added a few more tests to the test suite, and the library passes 
all of the tests so hopefully these changes will be pretty transparent 
to your applications. I hope to get a new snapshot out soon, with 
pre-built DLLs and libs as usual.

I've also made changes that fix a number of problems that came up when I 
changed to using the current MinGW + MsysDTK environment.

Regards.
Ross