http://issues.apache.org/SpamAssassin/show_bug.cgi?id=4488
[EMAIL PROTECTED] changed:
What |Removed |Added
----------------------------------------------------------------------------
Summary|Process size growing with |[review] Process size
|each processed message |growing with each processed
| |message
Status Whiteboard| |needs 2 votes
------- Additional Comments From [EMAIL PROTECTED] 2006-06-27 04:33 -------
Hrm. After spending a good bit thinking about this and going through the code,
I've decided that this patch does not fully addresses the issue, and ends up
causing some other issues. Specifically, the problem is that
M::SA::finish_learner() (which calls M::SA::Bayes::finish()) is called
specifically to only untie the DB, but the rest of the object should be left
intact. Only when M::SA::finish() is called should the Bayes and BayesStore
objects be wripped down.
So my thought (patch attached shortly) is for finish_learner() to call
Bayes::sanity_check_is_untied() (with a slight API addition) which will just
untie if we're currently tied. finish() will call Bayes::finish() which will go
ahead and untie as well as drop the reference to BayesStore. That will then let
BayesStore get cleaned up, and finish() then drops the Bayes reference and Bayes
gets cleaned up, and things are fine.
Please give this patch a try. :)
------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.
