Discussed two ways to solve gc of weakly ref'ed things.
Did away with preemptive thread suspend and resume.
Sorted out the issue with concurrent gc access to pointers.
Various small changes.
Some clarifications for shared mapping/multiset data blocks.
Googled a bit on tls support. OS X is a problem. :P
Talk about the string append case for single-refcount optimizations.
More on weak refs.
State compatibility as a goal.
More on single-refcount optimizations.
Detailing asynchronous lock space pointer access.
More details on which pike stack frames the micro-gc has to scan.
TLS on OS X is a no-go in the foreseeable future. :(
Various changes based on feedback and new ideas.
Described new gc approach.