Re: [Tcl9-cloverfield] [TCLCORE] ANNOUNCE: Colibri version 0.12

[Larry M. <lm...@bi...>]

On Wed, Nov 30, 2011 at 02:34:45PM +0100, Fr?d?ric Bonnet wrote:
> 1. Implemented page write protection-based parent tracking
> 
>      This is a major internal change in the generational GC. In previous
>      versions, parents (i.e. cells having children in younger generations) had
>      to be tracked explicitly by client code calls to Col_WordSetModified each
>      time a mutable cell was modified. This caused a significant performance
>      penalty and added complexity in client code. Now parents are automatically
>      tracked thanks to memory page write protection: pages of older generations
>      are write-protected, and write attempts are caught (using exceptions on
>      Windows and SIGSEGV signal handling on Unix). When such a page write is
>      detected, the write protection is removed and the page marked as modified,
>      and the client code resumes. All this happens transparently, so that client
>      code no longer have to track modifications. Upon GC, cells belonging to
>      modified pages are followed as potential parents so that younger reachable
>      cells are marked as well.

Strangely enough, we're playing similar games with write protection.  Can
you tell me

- which platforms have you tested this on?
  [ ] linux/x86
  [ ] linux/x86_64
  [ ] macos/x86
  [ ] windows/x86 2000? XP? 2003server? 2008server? Win7?
  [ ] HPUX/ia64
  [ ] AIX
  [ ] NET/Free/OpenBSD

- Is there a regression suite that tests the protection stuff?  If so, 
  a readme on how to run it and we'll see if we can test it on our
  cluster (all of the above and more)

- did you see any perf gain or loss?

Cool stuff!