Re: Proposal: giop-thread-invocation-stack

[Michael Meeks <michael ximian com>]

Hi Frank,

On Sat, 2003-11-15 at 10:58, Frank Rehberger wrote:
> I would like to commit this stuff (please look at attached 
> /tmp/giop-invocation-stack.diff.gz)

	Ok; I have a few things:

	Can you split this to two lines:

	if (mutex) g_mutex_free (mutex);

	Cosmetic but how we should do things I think.

	// FIXME, Frank, what about free-ing giop_main_thread and pool
	// of giop_threads?.

	An interesting FIXME :-) but using '//' in C code is a gcc extension
and will break Forte / Ansi compilers.

	For the 'giop_thread_invocation_stack_first_equal' thing it's normal to
have a typedef for the iterator function.

	The other thing is - it's good to pass the *tdata to these functions to
save a slow g_thread_private lookup, but since the stuff is now
per-thread, no-one else can be accessing that data at a given time - so
we don't in fact need the locking :-) to signify that we should add a _T
suffix to the method names I think.

	Finally - I think we should use a static GIOPThread *main_thread; for
the case where we are not in a threaded-mode, simply because I think
this could simplify a chunk of logic in other places where we have
similar special cases that currently add code & complexity.

	This would then remove the need for:

	GSList *stack = tdata ? tdata->current_invocations : 
			giop_current_invocations;

	and this sort of special case comment:

+        /* multi threaded setup only */

	Anyhow - it's _really_ great to have you working on this, sorry if
that's a rather daunting set of changes, but I think we have to work
hard to try and keep this as simple as we possibly can, otherwise it's
just going to get unmanageable / out of control :-)

	Thanks muchly,

		Michael.

-- 
 michael@ximian.com  <><, Pseudo Engineer, itinerant idiot