purecall in dgThread

[JoeWright2]

Hi, got a strange intermittent bug in dgThread::dgThreadSystemCallback at:

Code: Select all

me->Execute(me->m_id);

This can cause a purecall in debug mode (and I presume the cause of a crash I'm having in release mode).

I am recreating the world many times over in short succession (for an optimisation process). The current run I'm doing can have very short simulation times. I'm wondering if there's some overlap problems with the threads, but I don't have expertise in the area.

Is this fundamentally a bug, or should I be configuring things differently? Can I turn off threading?

Thanks, Joe

[Julio Jerez]

you can try uncomment
//#define DG_USE_THREAD_EMULATION

in dgThread.h

[JoeWright2]

That worked, thanks.

[JoeJ]

There's this function: NewtonWaitForUpdateToFinish (world),
I call this before changing stuff that may be still in use.

[JoeWright2]

Hi thanks I tried that, and it certainly did the trick when simulation times were shorter, but still needed Julio's suggestion for short simulations.

[Julio Jerez]

what do you mean by short simulation?>

you I would do is to create a Pool of world, the you can flush a world by calling delete all body, and repopulate again. this will same lot of overhead o create and destroy works whish is not cheap.

[JoeWright2]

I wrote that last one in a rush. What I meant to say was that NewtonWaitForUpdateToFinish worked when I was simulating, say, more than a couple of seconds of physical time, but when it was shorter, I ran into the problem of the purecall when simulating a shorter time and recreating the world.

So you would suggest not destroying the world? Just empty it all and recreate bodies?

Joe

[Julio Jerez]

So you would suggest not destroying the world? Just empty it all and recreate bodies?Joe

yes.