[Python-Dev] New dictionaries patch on SF

[Fred L. Drake, Jr.]

  I've been playing with dictionaries lately trying to stamp out a
bug:

http://sourceforge.net/bugs/?func=detailbug&bug_id=112558&group_id=5470

  It looks like any fix that really works risks a fair bit of
performance, and that's not good.  My best-effort fix so far is on
SourceForge:

http://sourceforge.net/patch/?func=detailpatch&patch_id=101277&group_id=5470

but doesn't quite work, according to Guido (I've not yet received
instructions from him about how to reproduce the observed failure).
  None the less, performance is an issue for dictionaries, so I came
up with the idea to use a specialized version for string keys.  When I
saw how few of the dictionaries created by the regression test ever
had anything else, I tried to simply make all dictionaries the
specialized variety (they can degrade themselves as needed).  What I
found was that just over 2% of the dictionaries created by running the
regression test ever held any non-string keys; this may be very
different for "real" programs, but I'm curious about how different.
  I've also done *no* performance testing on my patch for this yet,
and don't expect it to be a big boost without something like the bug
fix I mentioned above, but I could be wrong.  If anyone would like to
play with the idea, I've posted my current patch at:

http://sourceforge.net/patch/?func=detailpatch&patch_id=101309&group_id=5470

  Enjoy!  ;)


  -Fred

-- 
Fred L. Drake, Jr.  <fdrake at beopen.com>
BeOpen PythonLabs Team Member