Hi, After lots of debates off-list (specially with BrowserUk), I got to the following conclusions:
- Each OS thread runs a cooperative-event-scheduler where coroutines
might be enqueued.
- map returns the buffer immediatly and enqueues a coroutine to process
the data.
- Explicit parallel markings such as the feed operator, means a new OS
thread with its own cooperative scheduler is spawned.
- Data that is shared between thread is selectively marked that way to
make sure they are thread safe.
- The actor model can be implemented in a different module.
So, a quick summary of how the following code would run is attached:
my @a := map { cheap_op($_) } <== map { expensive_op($_) }, $*IN
The "value sharing framework" is still to be designed, but it's clear
that forcing actor model for everything was too intrusive, so we allow a
balance of "shared state" and "massage-passing" concurrency. This
balance will be implemented by modules.
comments are appreciated.
daniel
<<attachment: Perl6ThreadingModel.svg>>
