Overflow errors with random seeds about openmm HOT 5 OPEN

Can you show your code? This is probably coming from SWIG. The C++ method declares its argument to be an int. If you call the Python wrapper for it and pass a value that can't be converted, SWIG will reasonably throw an exception.

from openmm.

Can you show your code? This is probably coming from SWIG. The C++ method declares its argument to be an int. If you call the Python wrapper for it and pass a value that can't be converted, SWIG will reasonably throw an exception.

I just updated the description with a minimal example of how to produce the error for LangevinMiddleIntegrator.

from openmm.

@peastman Wouldn't it make sense to make the C++ functions take a uint32_t instead of an int?

I know this is not critical bug, but it is at least surprising for developers to get overflow errors when using uint32 seeds. And the error messages could be more helpful.

from openmm.

As far as the C++ is concerned, it makes no difference. The seed is just a sequence of bits. It doesn't really interpret it as a number, so there's no difference between signed and unsigned.

If we changed the type to uint32_t, then SWIG would throw an exception if you passed in an int that happened to be negative.

from openmm.

I agree that it shouldn't make a difference. In any case, it seems like you can avoid the SWIG overflow error by converting the seed to a signed integer:

seed = int(seed.view(np.int32))

from openmm.