Re: Recursive mutexes?

[Tom Lord]

   Tom Lord <address@hidden> writes:

   > In my "dream scheme" system, I'm thinking they aren't worth the
   > effort.  They slow down access to the store and complicate
   > programming.  I don't even want to think about how to reconcile them
   > with continuations, dynamic-wind, or fluids.

   It seems to me that we should not decide that a problem is too hard to
   solve well before trying.

Trying or thinking through, sure.   I just wanted to put the
alternative on the table -- not discourage careful consideration of
all alternatives.



   > I do want multiple interpreters (without a shared store) in separate
   > threads.  I do want low-level routines running in separate threads
   > (e.g., give a CPU to I/O or to reving cellular automata generations).   
   > But I'm having trouble seeing Scheme semantics as other than "optimal
   > for SISD".

   How will these different interpreters share data?


Dunno.  Lots of options exist.  Here's a few:

1) through optimizedj intra-process I/O

2) through segregated data structures that can be passed back and
   forth between threads (a high-level "segmented" store).

3) through a subset of values that _are_ shared.

3a) through java objects

3b) through objects of types defined by extensions to guile

-t