`AttributeError: module 'warp' has no attribute 'is_tensor'`

[Robin CREMESE]

Hello,
I'm new with wrap and already amazed by what it can achieve.
While I was trying the simulation module with warp V0.10.1 I ran into the following issue

Using the code snippet provided in the documentation I tried to print the output of the state.flatten() method. The error I get is :

File "/home/rcremese/mambaforge/envs/warp-env/lib/python3.10/site-packages/warp/sim/model.py", line 216, in flatten
    if wp.is_tensor(value):
AttributeError: module 'warp' has no attribute 'is_tensor

To reproduce here is the code I used :

# import warp
import warp as wp
import warp.sim

wp.init()
builder = warp.sim.ModelBuilder()

# anchor point (zero mass)
builder.add_particle((0, 1.0, 0.0), (0.0, 0.0, 0.0), 0.0)

# build chain
for i in range(1,10):
    builder.add_particle((i, 1.0, 0.0), (0.0, 0.0, 0.0), 1.0)
    builder.add_spring(i-1, i, 1.e+3, 0.0, 0)

# create model
model = builder.finalize("cuda")

state = model.state()
integrator = warp.sim.SemiImplicitIntegrator()

for i in range(100):
   print(state.flatten())
   state.clear_forces()
   integrator.simulate(model, state, state, dt=1.0/60.0)

I know the module is in active development but I wanted to point out this issue

[Miles Macklin]

Huh, this looks like a hangover from some long gone PyTorch code. For now I would avoid the method, although you could likely replace this particular call to isinstance(value, wp.array) to gather a list of all arrays on the State object. This may not be super useful though, since there are generally not too many attributes so you can generally enumerate them manually.

We used to use this flatten() method to gather a list of all PyTorch tensors for gradient tracking. We will likely remove this method in the next version, thanks for bringing this to our attention!

[Robin CREMESE]

I won't use this method as access to the state attributes is straightforward but I thought it would be usefull to know the bug.

[Eric Shi]

Closing because this has been fixed as of 1.0.0b2, which is available on PyPI. Using Model.flatten() also currently prints the following deprecation notice:

DeprecationWarning: Model.flatten() will be removed in a future Warp version.