yade-dev team mailing list archive

[William Chèvremont <william.chevremont@xxxxxxxxxxxxxxxxxxxxxx>]

Ok got it.

The problem come from the Global Interpreter Lock from Python. run() 
seems to run the code in an other thread as step() run the code in the 
same thread as python. So, when calling the function via step(), the 
lock is already acquired by the thread. On the other hand, run() is into 
a different thread, and the lock has to be acquired manually.

Using boost, I was expecting an exception instead of crash for this kind 
of error.

Functions should look like:

virtual Real contactForce(Real const& u) const {
        TRACE;
        gilLock lock; // Acquire GIL. (from pyutils/gil.hpp)
        LOG_TRACE("GIL State: " << PyGILState_Check());
        return get_override("contactForce")(u);
    }

I'll update the doc about subclassing c++ in python, as this is 
mandatory for all python code called from C++ while running O.run().
William.



On 08/10/2019 13:22, William Chèvremont wrote:
> Hi,
>
> Yes, there are the same behaviour with and without wait.
>
> William
>
>
> On 08/10/2019 12:57, Bruno Chareyre wrote:
> > Hi William,
> > I don't know precisely but in case it can help I would raise that the 
> > main difference between step() and run() is Py_BEGIN_ALLOW_THREADS, 
> > at least if O.run(...,wait=True).
> > Do you have the same problem with and without "wait"?
> >
> > Simply passing to c++ a python expression (just a string with python 
> > commands) or a python function returning a value - instead of a 
> > derived class object - could be a simple workaround maybe.
> >
> > Bruno
> >
> >
> >
> >
> >
> >
> >
> >
> > On Tue, 8 Oct 2019 at 10:56, William Chèvremont 
> > <william.chevremont@xxxxxxxxxxxxxxxxxxxxxx 
> > <mailto:william.chevremont@xxxxxxxxxxxxxxxxxxxxxx>> wrote:
> >
> >     Hi,
> >
> >     @Janek
> >
> >     Yes, I'm aware of the doc about subclassing types in python.
> >     There are
> >     no problem about that, since it works when calling O.step()
> >     instead of
> >     O.run().
> >
> >     @Anton
> >
> >     The complete backtrace is attached to this mail. Functions of
> >     interests
> >     are around line 400.
> >
> >     Best Regards,
> >
> >     William
> >
> >
> >     On 07/10/2019 18:19, Janek Kozicki (yade) wrote:
> >     > Only a quick question to make sure - you have read
> >     https://yade-dem.org/doc/prog.html#subclassing-c-types-in-python
> >     in documentation?
> >     >
> >     >
> >     > best regards
> >     > Janek
> >     >
> >     > _______________________________________________
> >     > Mailing list: https://launchpad.net/~yade-dev
> >     > Post to     : yade-dev@xxxxxxxxxxxxxxxxxxx
> >     <mailto:yade-dev@xxxxxxxxxxxxxxxxxxx>
> >     > Unsubscribe : https://launchpad.net/~yade-dev
> >     > More help   : https://help.launchpad.net/ListHelp
> >     _______________________________________________
> >     Mailing list: https://launchpad.net/~yade-dev
> >     Post to     : yade-dev@xxxxxxxxxxxxxxxxxxx
> >     <mailto:yade-dev@xxxxxxxxxxxxxxxxxxx>
> >     Unsubscribe : https://launchpad.net/~yade-dev
> >     More help   : https://help.launchpad.net/ListHelp
> >
> >
> >
> > --
> > --
> > _______________
> > Bruno Chareyre
> > Associate Professor
> > ENSE³ - Grenoble INP
> > Lab. 3SR
> > BP 53
> > 38041 Grenoble cedex 9
> > Tél : +33 4 56 52 86 21
> > ________________
> >
> > Email too brief?
> > Here's why: email charter 
> > <https://marcuselliott.co.uk/wp-content/uploads/2017/04/emailCharter.jpg> 
>
> _______________________________________________
> Mailing list: https://launchpad.net/~yade-dev
> Post to     : yade-dev@xxxxxxxxxxxxxxxxxxx
> Unsubscribe : https://launchpad.net/~yade-dev
> More help   : https://help.launchpad.net/ListHelp