Some tests fail when tensorflow is imported before catalyst

[rmoyard]

The imports

import tensorflow
import catalyst

do not always work, and some tests are failing if tensorflow is imported first. We can see that in when testing the wheels.

FAILED frontend/test/pytest/test_autograph.py::TestConditionals::test_qubit_manipulation_cond[lightning.qubit] - RuntimeError: random_device could not be read
FAILED frontend/test/pytest/test_conditionals.py::TestCond::test_qubit_manipulation_cond[lightning.qubit]
FAILED frontend/test/pytest/test_jax_integration.py::TestJAXJIT::test_multiple_calls[lightning.qubit]
FAILED frontend/test/pytest/test_mid_circuit_measurement.py::TestMidCircuitMeasurement::test_basic[lightning.qubit] - RuntimeError: random_device could not be read
FAILED frontend/test/pytest/test_mid_circuit_measurement.py::TestMidCircuitMeasurement::test_more_complex[lightning.qubit] - RuntimeError: random_device could not be read
FAILED frontend/test/pytest/test_pytree_args.py::TestPyTreesReturnValues::test_return_value_float[lightning.qubit] - RuntimeError: random_device could not be read
FAILED frontend/test/pytest/test_pytree_args.py::TestPyTreesReturnValues::test_return_value_tuples[lightning.qubit]
FAILED frontend/test/pytest/test_measurements_results.py::TestOtherMeasurements::test_multiple_return_values[lightning.qubit]
FAILED frontend/test/pytest/test_jit_behaviour.py::TestCallsiteCompileVsFunctionDefinitionCompile::test_equivalence[lightning.qubit] - RuntimeError: random_device could not be read
FAILED frontend/test/pytest/test_jit_behaviour.py::TestDecorator::test_function_is_cached[lightning.qubit] - RuntimeError: random_device could not be read
FAILED frontend/test/pytest/test_pytree_args.py::TestPyTreesFuncArgs::test_args_used_in_measure[lightning.qubit] - RuntimeError: random_device could not be read
FAILED frontend/test/pytest/test_jit_behaviour.py::TestCaching::test_function_is_cached[lightning.qubit] - RuntimeError: random_device could not be read
FAILED frontend/test/pytest/test_pytree_args.py::TestPyTreesFuncArgs::test_args_used_indirectly_in_measure[lightning.qubit] - RuntimeError: random_device could not be read
FAILED frontend/test/pytest/test_jit_behaviour.py::TestShots::test_shots_in_decorator_in_sample[lightning.qubit] - RuntimeError: random_device could not be read
FAILED frontend/test/pytest/test_jit_behaviour.py::TestShots::test_shots_in_callsite_in_sample[lightning.qubit]
FAILED frontend/test/pytest/test_qnode.py::test_variable_capture[0-False]
FAILED frontend/test/pytest/test_pytree_args.py::TestPyTreesReturnValues::test_return_value_dict[lightning.qubit]
FAILED frontend/test/pytest/test_loops.py::TestWhileLoops::test_alternating_loop[lightning.qubit]
FAILED frontend/test/pytest/test_conditionals.py::TestCond::test_identical_branch_names[lightning.qubit] - RuntimeError: random_device could not be read
FAILED frontend/test/pytest/test_contexts.py::TestTracing::test_fixed_tracing[lightning.qubit] - RuntimeError: random_device could not be read
FAILED frontend/test/pytest/test_contexts.py::TestTracing::test_cond_inside_while_loop[lightning.qubit]
FAILED **frontend/test/pytest/test_variable_wires.py::TestBasicCircuits::test_measurements[lightning.qubit-args0]**

[erick-xanadu]

I have contributed to the discussion on this ticket after our investigation https://github.com/tensorflow/tensorflow/issues/61741#issuecomment-1749307431

[dime10]

Looks like this issue has been fixed upstream.