Dan says: >'Kay, here's the thought. Exactly one filter may be (optionally) marked as >an interpreter terminal filter. It runs in the context of the interpreter >that made the I/O request, and gets (or returns) a real PMC. All the other >filters run in the context of a separate interpreter, and while they can >also pass PMCs between each other, communication between the terminal >filter and the source filter interpreter is via a non-reallocable string. >(Yes, the string buffer is static. We can cope... :)
-snip- >I think that makes sense. I might be wrong, but I think so. Good enough? What differentiates a "trusted" filter from a "non-trusted". Who is allowed to mark a filter as terminal, user code or language implementors? Lastly, how much overhead in interpreter creation/teardown. Probably enough to keep around a stashed "IO" interpreter right? -Melvin Smith IBM :: Atlanta Innovation Center [EMAIL PROTECTED] :: 770-835-6984
