def test_validate_state_no_reward():
state = CompilerEnvState(
benchmark="cBench-v0/dijkstra",
walltime=1,
commandline="opt input.bc -o output.bc",
)
env = gym.make("llvm-v0")
try:
result = validate_state(env, state)
finally:
env.close()
assert result.success
assert not result.failed
assert not result.reward_validated
assert str(result) == "✅ cBench-v0/dijkstra"
def test_validate_state_with_reward():
state = CompilerEnvState(
benchmark="cBench-v0/crc32",
walltime=1,
reward=0,
commandline="opt input.bc -o output.bc",
)
env = gym.make("llvm-v0", reward_space="IrInstructionCount")
try:
env.require_dataset("cBench-v0")
result = validate_state(env, state)
finally:
env.close()
assert result.success
assert not result.failed
assert result.reward_validated
assert not result.reward_validation_failed
assert str(result) == "✅ cBench-v0/crc32 0.0000"
def test_validate_state_invalid_reward():
state = CompilerEnvState(
benchmark="cBench-v0/dijkstra",
walltime=1,
reward=1,
commandline="opt input.bc -o output.bc",
)
env = gym.make("llvm-v0", reward_space="IrInstructionCount")
try:
result = validate_state(env, state)
finally:
env.close()
assert not result.success
assert result.failed
assert result.reward_validated
assert result.reward_validation_failed
assert (
str(result) ==
"❌ cBench-v0/dijkstra Expected reward 1.0000 but received reward 0.0000"
)