On Mon, May 03, 2004 at 10:46:28AM -0400, Dan Sugalski wrote:
> 3) The embedding wrapper is responsible for setting and resetting the
> top of stack.
I don't think that this is quite right. The embedding wrapper needs to
set (and reset) the top of stack only if it's not set.
Otherwise this scenario could happen:
Eternal Program
|
+-> Embedding wrapper (sets stack top)
|
+-> More parrot code with something unanchored on stack *
|
+-> Call back out to a subroutine in the external program
|
+-> Embedding wrapper (sets stack top)
|
+-> More parrot code that triggers GC
|
+-> GC
And the GC routine won't spot that pointer on the stack at *
If the embedding wrapper (well wrappers onto the data access functions and
onto the runloop) both set a stack top if it's not yet set, and clear it on
exit if they set it, then I think that parrot won't fall foul of the above.
Effectively the embedder is playing by the same rules as Leo said that we
have to have for a nested call to the runloop - anything calling the runloop
must ensure that all PMCs are anchored.
Nicholas Clark
