FirstThread

[Tom Tromey]

>>>>> "Jeff" == Jeff Sturm <jsturm@sigma6.com> writes:

Jeff> I don't think the thread layer would change much.  Some
Jeff> functions may no longer be needed, like _Jv_ThreadWait.

We'll still need something.  Our (still internal) cooperative thread
system doesn't use the "main" thread.  Maybe I'm misunderstanding; I
think I'd need to see a patch, or more details.

Jeff> I've been thinking some about the interaction of threads with
Jeff> boehm-gc though... every new thread system has to be ported
Jeff> twice.  There is a linux-threads.c port in boehm-gc and a
Jeff> corresponding posix-threads.c port in libjava.  Similar for
Jeff> quick-threads.  Since everything boehm-gc needs from the thread
Jeff> system can be done easily in libgcj (at least once we have
Jeff> suspend/resume), why not just create one gcj-threads port in
Jeff> boehm-gc and have it call back into libgcj?

That would be nice, because it would let us reuse the thread code
if/when we write a new GC.

We also need a way to get the stack limits, though, which might be
hard.

And there might be nasty interactions with shared libraries (a part of
the GC which I haven't explored yet).

GC/thread/libjava interactions have never been specified.  We took a
simplistic approach at first since we knew we only had to deal with
one GC.  The current scheme was never intended to be the final one.

Tom