Compiled netx hdf5 models cannot be serialized.

[Gavinator98]

Describe the bug
netx hdf5 models create an error when attempting to save a compiled executable

To reproduce current behavior
Steps to reproduce the behavior:

1. Load and compile a netx hdf5 model

from lava.lib.dl import netx
from lava.magma.core.run_configs import Loihi2HwCfg
from lava.utils.serialization import save

net = netx.hdf5.Network(net_config='some_model.net')
executable = net.compile(run_cfg=Loihi2HwCfg())

save(processes=[net],
     executable=executable,
     filename="compiled_model.pkl")

2. Produces this error

h5py objects cannot be pickled

As a workaround, the h5py object can be removed before saving.

net.net_config = None 

Expected behavior
netx.hdf5.Network objects work with the save() function without modification.

Environment (please complete the following information):

• Device: Server
• OS: Linux
• Lava version 0.8.0
• Lava-Loihi version 0.5.0

[PhilippPlank]

Thank you for reporting this issue. We are aware that this is not possible and it is an h5py issue in general. We are testing some mitigation strategies, but h5py is explicitly preventing to use pickle and changing the h5py package is not a general solution. If you need that, you can do it though (remove the exception in the h5py code).