Re: More problems related to MT

[Vladimir T. <vtz...@gm...>]

On Thu, 26 Feb 2009 10:28:00 +0200, Don Cohen  
<don...@is...> wrote:
> Sam Steingold writes:
>  > interrupt can occur at any gc-safe spot.
> Then which spots are gc safe?

Currently GC safe spots are (new may be added if needed):
1. Any heap allocation
2. Any JMP bytecode instruction (this includes all variants of JMP  
instruction - JMPIF, JMPIF1, etc). This is too strong - but seems it does  
not cause noticeable performance decrease.
3. GO spec form in interpreted code
4. While the execution is in marked as "possibly blocking" foreign  
non-lisp code (read from socket for example).

> More to the point, how does one look at a piece of lisp code and
> determine in which states it could interrupt?

you cannot (and should not rely on this - since in future safe spots may  
change).