[Python-Dev] pymalloc killer
Kevin Jacobs
jacobs@penguin.theopalgroup.com
Fri, 29 Mar 2002 06:57:54 -0500 (EST)
Some wild-ass thoughts:
On Fri, 29 Mar 2002, Tim Peters wrote:
> A nasty subtlety: if PyMem_NukeIt is called when the GIL isn't held, it's
> going to take a lot of thought to see whether this can be made safe when
> called simultaneously by multiple threads; e.g., the finger can change
> "mid-stream"
Then make multiple thread-specific fingers, which will presumably result in
higher hit rates due to better locality. To prevent fingers from being
invalidated, do not remove arenas that are deallocated from the tree -- just
mark them inactive.
> then, and, worse, some thread that *does* hold the GIL may
> legitimately cause a new arena to be allocated midstream, mutating the
> address vector during the search. This may be intractable enough to kill
> the idea of mapping PyMem_XXX to this.
Why not use a smaller (non-global) lock to protect arena modification and
finger flushing? Ideally, the lock should be cheap for readers (which I
presume will be the vast majority of accesses) and relatively expensive for
writers (which I assume will occur only when arenas are added or
deactivated).
-Kevin
--
Kevin Jacobs
The OPAL Group - Enterprise Systems Architect
Voice: (216) 986-0710 x 19 E-mail: jacobs@theopalgroup.com
Fax: (216) 986-0714 WWW: http://www.theopalgroup.com