Thank you for reply. I was bit worried you didn't replied to message about dictionaries, where i described what causing problems with my dicts. A reason is quite surprising: because they too fast :)
On 24/08/07, Chris Muller ma.chris.m@gmail.com wrote:
Hi Sig, I guess I did not see your email about the Dictionary's, but here is my reply to Keith after he asked.
I won't have time to get too much into benchmarking in the near future (plate is overflowing) but am grateful for your work (as I'm sure all Magma users are).
I want to establish an appropriate benchmark (maybe add an additional to MagmaBenchmarker) and try all of the below-mentioned solutions; Martin Loewies', yours, HashTable and any other possibilities.
Best obviously is if the MagmaBenchmarker can identify which ones are most suitable but I think the existing tests in there are a bit too short (not enough entires in the Dictionary's, they don't test that) to really reveal what would be the best dictionary, so we need to add an additional test to the output.
Thanks. Chris
---------- Forwarded message ---------- From: Chris Muller ma.chris.m@gmail.com Date: Aug 2, 2007 10:42 PM Subject: Re: coordinating Magma development - configurations To: Keith Hodges keith_hodges@yahoo.co.uk
Yes. I am keenly interested in the work Sig has done on the WeakKeyIdentityDictionary's. Magma spends a considerable portion of its time spinning through these Dictionary's, so I believe speeding them up will have a direct, impactful benefit to Magma.
Release 40 is about to go out the door so it is way too late to try to evaluate his new "MaDictionary" upgrade against the other Dictionary options like the one from Martin Loewis at
http://bugs.squeak.org/view.php?id=6348
Also, there was mention of a HashTable (?) and others that need investigating.
I have, however, included his enhancement that lets one plug whatever kind of Dictionary one wants to use, which is a perfect segway to "Release 41", so we'll have time to plug in the various options and see how they all perform compared to each other.
Oh, this enhancement is not complete , i added a choice to select WeakKeyIdentityDictionary class, but not WeakValueDictionary. While testing Db, which Florian gave me i noticed, that most speed impact lies on replacing WeakValueDictionary.
Thanks, Chris
PS - Sig is welcome to contact me anytime, better yet, this discussion would be great for the mailing list.
On 8/2/07, Keith Hodges keith_hodges@yahoo.co.uk wrote:
Sig asked me whether you were going to look at his replacement dictionaries.
I told him that you were probably being methodical and would get around to it.
best regards
Keith
magma@lists.squeakfoundation.org