http://issues.apache.org/SpamAssassin/show_bug.cgi?id=4696
[EMAIL PROTECTED] changed:
What |Removed |Added
----------------------------------------------------------------------------
Attachment #3343 is|0 |1
obsolete| |
AssignedTo|[email protected] |[EMAIL PROTECTED]
Status|NEW |ASSIGNED
------- Additional Comments From [EMAIL PROTECTED] 2006-01-28 02:30 -------
Created an attachment (id=3348)
--> (http://issues.apache.org/SpamAssassin/attachment.cgi?id=3348&action=view)
better fix
ok, this is a better fix. That code:
(a) could leak alarm die() noise (a la bug 4631);
(b) the extra few seconds were still not enough;
(c) it wasn't protecting itself against any other active timeouts, so a
timeout-within-a-timeout wasn't going to work;
(d) it should have been stopping the timeout timer before the eval { } scope
was exited, instead of after;
(e) it should have been treating an alarm timeout event as equivalent to the
select() timeout, instead of issuing warnings, since under heavy load, the
alarm() timeout may be noticed first.
So, all in all, (b) and (e) were the big issues that causes this bug. All the
others, (a), (c), and (d), are ones that we have fixed in other places (e.g.
cf. bug 4631), but not yet here.
BTW obviously that's not good. I plan to abstract all these timeout timers
into a new class in svn trunk, I'm sick of fixing the same bugs again and again
due to duplication :(
------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.
