threads & gc

[Michael Livshin]

hello all.

the more I think about the recent "gc issues" thread here, the more
confused I become.  the following might be not very coherent, sorry.

it was mentioned earlier (by Mikael, I think) that the plan is to
make sure, before doing GC, that all mutator threads are stopped.

now, I can think of two ways to stop all threads:

* assuming we know about all running threads, we just suspend them
  "from the outside".

  this would require:

  * either a way to enumerate all threads in the process or a
    requirement that all Guile-using threads are started with a
    Guile-specific "pthread_create" wrapper.

  * a way to suspend a thread from another thread.  I can't find such
    a thing in pthreads, maybe I'm looking in a wrong place?

* the "cooperative" way: there are "checkpoints" before each cons and
  periodically in the evaluator.

  this means that when GC occurs, no thread can be in the "just got a
  new cell and it's half-initialized and now I'm suspended" state, so
  life is good.

please set me straight,
--mike

-- 
Who's afraid of the garbage collector?