I have a GC tuning branch which improves performance by about 10%, and we have a new GC in progress as well. Our target date for the latter is for the 3.0 release in January.
2. How do these work with the new concurrency (threading) system that as I remember Team Parrot was going to write?
Ideally, for the multicore future that is now today, I'd like to see multiple simultaneous mutators and scavengers without a stop the world requirement. That said, stock HotSpot doesn't supply the latter, although Azul has done some neat stuff to extend it and apparently Do A Right Thing (their Pauseless mark and compact GC and their open source release of that in their Managed Runtime Initiative).
It would be nice if Parrot can someday do this without extreme pain.
The GC is mostly precise. We haven't yet eliminated our use of the C stack (so full GC has to examine the appropriate registers and memory locations to find pointers not otherwise anchored), but when we do, we'll have a fully precise GC.
Our current GC in trunk right now still stops the world, even when concurrency is active, but that's an incidental detail of implementation and not a requirement for the algorithm. We could make it incremental with a small investment of time, but our priority now is to migrate to the improved generational GC on the branch first.
Oh yeah, as I recall Hotspot doesn't have production quality generational GC yet. That tends to pay off big in functional languages and if I'd had to guess before reading about your priorities I would have bet that generational would be fairly advantageous for a lot of Perl code, certainly the majority I've written (e.g. ETL programs).
Good luck! Nail down this and your concurrency by the time Clojure-in-Clojure is written (there's no guess when that'll happen) and I wouldn't be surprised if someone chooses Parrot as a porting target.
Generational GC will be very useful for Perl 6 and other languages running on Parrot, especially if we can add reliable escape analysis. The ability to perform the equivalent of stack allocation in C offers many performance benefits.
