Add tests/test_env.py
Browse files- tests/test_env.py +82 -0
tests/test_env.py
ADDED
|
@@ -0,0 +1,82 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import pytest
|
| 2 |
+
from codereview_env.env import CodeReviewEnv
|
| 3 |
+
from codereview_env.models import TaskId, Action, ActionType, Category, Severity, Verdict
|
| 4 |
+
|
| 5 |
+
def test_env_reset():
|
| 6 |
+
env = CodeReviewEnv()
|
| 7 |
+
res = env.reset(TaskId.BUG_DETECTION, seed=0)
|
| 8 |
+
assert res.task_id == TaskId.BUG_DETECTION
|
| 9 |
+
assert res.seed == 0
|
| 10 |
+
assert res.observation.step_count == 0
|
| 11 |
+
assert res.observation.noise_budget == 5
|
| 12 |
+
|
| 13 |
+
def test_env_step_bug_detection():
|
| 14 |
+
env = CodeReviewEnv()
|
| 15 |
+
res = env.reset(TaskId.BUG_DETECTION, seed=1)
|
| 16 |
+
# Seed 1 selects bug_003: None dereference in auth.py
|
| 17 |
+
|
| 18 |
+
# Flag the bug correctly
|
| 19 |
+
action = Action(
|
| 20 |
+
action_type=ActionType.FLAG_ISSUE,
|
| 21 |
+
body="None dereference null check guard clause",
|
| 22 |
+
filename="auth.py",
|
| 23 |
+
line_number=16,
|
| 24 |
+
category=Category.BUG,
|
| 25 |
+
severity=Severity.HIGH
|
| 26 |
+
)
|
| 27 |
+
step_res = env.step(action)
|
| 28 |
+
assert step_res.observation.step_count == 1
|
| 29 |
+
assert step_res.reward > 0
|
| 30 |
+
assert step_res.done == False
|
| 31 |
+
|
| 32 |
+
# Terminal action
|
| 33 |
+
action_term = Action(
|
| 34 |
+
action_type=ActionType.APPROVE,
|
| 35 |
+
body="LGTM",
|
| 36 |
+
verdict=Verdict.LGTM
|
| 37 |
+
)
|
| 38 |
+
step_term = env.step(action_term)
|
| 39 |
+
assert step_term.done == True
|
| 40 |
+
|
| 41 |
+
final = env.get_final_result()
|
| 42 |
+
assert final.final_score > 0
|
| 43 |
+
|
| 44 |
+
def test_env_noise_budget_exhaustion():
|
| 45 |
+
env = CodeReviewEnv()
|
| 46 |
+
env.reset(TaskId.BUG_DETECTION, seed=0)
|
| 47 |
+
|
| 48 |
+
# Flag 5 false positives
|
| 49 |
+
action_fp = Action(
|
| 50 |
+
action_type=ActionType.FLAG_ISSUE,
|
| 51 |
+
body="fp",
|
| 52 |
+
filename="nonexistent",
|
| 53 |
+
line_number=999,
|
| 54 |
+
category=Category.BUG,
|
| 55 |
+
severity=Severity.LOW
|
| 56 |
+
)
|
| 57 |
+
|
| 58 |
+
for i in range(4):
|
| 59 |
+
res = env.step(action_fp)
|
| 60 |
+
assert res.done == False
|
| 61 |
+
assert res.observation.noise_budget == 5 - (i + 1)
|
| 62 |
+
|
| 63 |
+
res_final = env.step(action_fp)
|
| 64 |
+
assert res_final.done == True
|
| 65 |
+
assert res_final.observation.noise_budget == 0
|
| 66 |
+
|
| 67 |
+
def test_env_max_steps():
|
| 68 |
+
env = CodeReviewEnv()
|
| 69 |
+
env.reset(TaskId.BUG_DETECTION, seed=0)
|
| 70 |
+
|
| 71 |
+
action_neutral = Action(
|
| 72 |
+
action_type=ActionType.ASK_QUESTION,
|
| 73 |
+
body="what's this?"
|
| 74 |
+
)
|
| 75 |
+
|
| 76 |
+
for i in range(9):
|
| 77 |
+
res = env.step(action_neutral)
|
| 78 |
+
assert res.done == False
|
| 79 |
+
|
| 80 |
+
res_final = env.step(action_neutral)
|
| 81 |
+
assert res_final.done == True
|
| 82 |
+
assert res_final.observation.step_count == 10
|