-
I am running more tests to try to identify any exception that may be raised, although, IMHO it's dangerous that FastMM retains a lock if an exception is raised within a thread, it should release the lock somehow to avoid to block all other threads - maybe it should spawn its onw "guardian" thread?
It could be an issue of mine, but I wouldn't posted here if I didn't find all my thread...
2009-08-19 19:56:07 UTC in FastMM
-
Once started, threads are never suspended calling Suspend(), they may be put on wait using WaitForSingleObject() on a TEvent but only when they've finished their task and a new one is not available, but I should see it in the stack trace, as far as I can see in those traces all threads looks to be "alive and running" and blocked in the middle of their processing.
Any unhandled...
2009-08-19 14:05:42 UTC in FastMM
-
At last I've been able to reproduce the error and collect call stacks (using map2dbg and SysInternals ProcessExplorer). I found nine threads all blocked inside FastMM, ProcessExplorer just shows a lot of context switches but each thread don't get past the sleep call. This data where collected with NeverSleepOnThreadContention *not* set.
=====================================
thread 3452...
2009-08-19 07:55:49 UTC in FastMM
-
> What I would recommond you should do as a first step is to enable the
> NeverSleepOnThreadContention option and then to run the application
> until it deadlocks
I run a test for four days on four machines with NeverSleepOnThreadContention enabled and the application never deadlocked. Looking at the code I see that if it is set FastMM never calls Sleep().
My application may...
2009-07-28 09:52:25 UTC in FastMM
-
Forgot to add I am using 4.92.
A suggestion: Windows allows for critical sections with a "spin count" (see InitializeCriticalSectionAndSpinCount) but in Windows 95 AFAIK. Have you ever tried it to "lock" memory blocks instead on relying on Sleep()? It could be give the same benefits of NeverSleepOnThreadContention but still going to a wait if the lock cannot be acquired...
2009-07-22 08:48:17 UTC in FastMM
-
I am encountering something that looks like a deadlock when a service of mine is run with 8 concurrent threads on a dual Xeon dual core server. Sometimes the application stops responding, and inspecting it with SysInternals Process Explorer everytime I found each thread blocked in a Sleep() call.
I can trace that call in a DLL of mine which is used to store data in TMemoryStreams. This is the...
2009-07-22 07:55:44 UTC in FastMM
