File size: 15,120 Bytes
af83196 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 | """Tests for HarborEvaluator — solution path extraction, task.toml parsing, reward reading, and detection."""
import json
import textwrap
from unittest.mock import patch, MagicMock
import pytest
from skydiscover.config import EvaluatorConfig
from skydiscover.evaluation import _is_harbor_task, _is_containerized
from skydiscover.evaluation.harbor_evaluator import HarborEvaluator, _DEFAULT_SOLUTION_PATH
def _make_evaluator(task_dir: str) -> HarborEvaluator:
"""Create a HarborEvaluator without starting Docker."""
inst = object.__new__(HarborEvaluator)
inst.task_dir = task_dir
return inst
# ------------------------------------------------------------------
# task.toml timeout parsing
# ------------------------------------------------------------------
class TestTaskTomlTimeout:
def test_reads_verifier_timeout(self, tmp_path):
(tmp_path / "task.toml").write_text("[verifier]\ntimeout_sec = 3600\n")
inst = _make_evaluator(str(tmp_path))
config = EvaluatorConfig()
inst._apply_task_toml_timeout(config)
assert config.timeout == 3600
def test_no_task_toml_keeps_default(self, tmp_path):
inst = _make_evaluator(str(tmp_path))
config = EvaluatorConfig()
inst._apply_task_toml_timeout(config)
assert config.timeout == 360
def test_missing_key_keeps_default(self, tmp_path):
(tmp_path / "task.toml").write_text("[metadata]\nname = 'test'\n")
inst = _make_evaluator(str(tmp_path))
config = EvaluatorConfig()
inst._apply_task_toml_timeout(config)
assert config.timeout == 360
def test_inline_timeout(self, tmp_path):
(tmp_path / "task.toml").write_text("timeout_sec = 1200\n")
inst = _make_evaluator(str(tmp_path))
config = EvaluatorConfig()
inst._apply_task_toml_timeout(config)
assert config.timeout == 1200
def test_malformed_toml_keeps_default(self, tmp_path):
(tmp_path / "task.toml").write_bytes(b"\x80\x81\x82")
inst = _make_evaluator(str(tmp_path))
config = EvaluatorConfig()
inst._apply_task_toml_timeout(config)
assert config.timeout == 360
# ------------------------------------------------------------------
# Solution path extraction: solve.sh (tier 1)
# ------------------------------------------------------------------
class TestExtractPathFromSolveSh:
def _write_solve_sh(self, tmp_path, content: str):
solution_dir = tmp_path / "solution"
solution_dir.mkdir(exist_ok=True)
(solution_dir / "solve.sh").write_text(content)
return _make_evaluator(str(tmp_path))
def test_absolute_cat_redirect(self, tmp_path):
inst = self._write_solve_sh(tmp_path, "cat > /app/solver.py << 'EOF'\nprint('hi')\nEOF\n")
assert inst._extract_path_from_solve_sh() == "/app/solver.py"
def test_bare_redirect(self, tmp_path):
inst = self._write_solve_sh(tmp_path, "> /workspace/solution.py << 'EOF'\ncode\nEOF\n")
assert inst._extract_path_from_solve_sh() == "/workspace/solution.py"
def test_rust_extension(self, tmp_path):
inst = self._write_solve_sh(tmp_path, "cat > /app/src/main.rs << 'EOF'\nfn main(){}\nEOF\n")
assert inst._extract_path_from_solve_sh() == "/app/src/main.rs"
def test_cpp_extension(self, tmp_path):
inst = self._write_solve_sh(tmp_path, "cat > /solution/solve.cpp << 'EOF'\nint main(){}\nEOF\n")
assert inst._extract_path_from_solve_sh() == "/solution/solve.cpp"
def test_relative_path_with_cd(self, tmp_path):
content = textwrap.dedent("""\
#!/bin/bash
cd "/workspace/project"
cat > src/interfaces/base.rs << 'EOF'
code
EOF
""")
inst = self._write_solve_sh(tmp_path, content)
assert inst._extract_path_from_solve_sh() == "/workspace/project/src/interfaces/base.rs"
def test_relative_path_with_variable_assignment(self, tmp_path):
content = textwrap.dedent("""\
#!/bin/bash
RBENCH_DIR="/workspace/rbench_reference"
cat > src/main.py << 'EOF'
code
EOF
""")
inst = self._write_solve_sh(tmp_path, content)
assert inst._extract_path_from_solve_sh() == "/workspace/rbench_reference/src/main.py"
def test_relative_path_with_dockerfile_workdir(self, tmp_path):
(tmp_path / "solution").mkdir()
(tmp_path / "solution" / "solve.sh").write_text("cat > solver.py << 'EOF'\ncode\nEOF\n")
(tmp_path / "environment").mkdir()
(tmp_path / "environment" / "Dockerfile").write_text("FROM python:3.11\nWORKDIR /opt/app\n")
inst = _make_evaluator(str(tmp_path))
assert inst._extract_path_from_solve_sh() == "/opt/app/solver.py"
def test_no_solve_sh_returns_empty(self, tmp_path):
inst = _make_evaluator(str(tmp_path))
assert inst._extract_path_from_solve_sh() == ""
def test_no_redirect_returns_empty(self, tmp_path):
inst = self._write_solve_sh(tmp_path, "#!/bin/bash\necho hello\n")
assert inst._extract_path_from_solve_sh() == ""
# ------------------------------------------------------------------
# Solution path extraction: instruction.md (tier 2)
# ------------------------------------------------------------------
class TestExtractPathFromInstruction:
def test_backtick_path(self, tmp_path):
(tmp_path / "instruction.md").write_text("Write your solution in `/app/solver.py`.\n")
inst = _make_evaluator(str(tmp_path))
assert inst._extract_path_from_instruction() == "/app/solver.py"
def test_quoted_path(self, tmp_path):
(tmp_path / "instruction.md").write_text('Save your code to "/workspace/solve.py".\n')
inst = _make_evaluator(str(tmp_path))
assert inst._extract_path_from_instruction() == "/workspace/solve.py"
def test_preposition_path(self, tmp_path):
(tmp_path / "instruction.md").write_text("Place your solution at /opt/solution.py and run it.\n")
inst = _make_evaluator(str(tmp_path))
assert inst._extract_path_from_instruction() == "/opt/solution.py"
def test_no_path_returns_empty(self, tmp_path):
(tmp_path / "instruction.md").write_text("Solve this problem efficiently.\n")
inst = _make_evaluator(str(tmp_path))
assert inst._extract_path_from_instruction() == ""
def test_no_file_returns_empty(self, tmp_path):
inst = _make_evaluator(str(tmp_path))
assert inst._extract_path_from_instruction() == ""
# ------------------------------------------------------------------
# Full solution path extraction (tier priority)
# ------------------------------------------------------------------
class TestExtractSolutionPath:
def test_prefers_solve_sh_over_instruction(self, tmp_path):
(tmp_path / "solution").mkdir()
(tmp_path / "solution" / "solve.sh").write_text("cat > /from/solve.py << 'EOF'\nEOF\n")
(tmp_path / "instruction.md").write_text("Write to `/from/instruction.py`.\n")
inst = _make_evaluator(str(tmp_path))
assert inst._extract_solution_path() == "/from/solve.py"
def test_falls_back_to_instruction(self, tmp_path):
(tmp_path / "instruction.md").write_text("Write to `/from/instruction.py`.\n")
inst = _make_evaluator(str(tmp_path))
assert inst._extract_solution_path() == "/from/instruction.py"
def test_falls_back_to_default(self, tmp_path):
inst = _make_evaluator(str(tmp_path))
assert inst._extract_solution_path() == _DEFAULT_SOLUTION_PATH
# ------------------------------------------------------------------
# _read_reward
# ------------------------------------------------------------------
def _mock_docker_exec(outputs: dict):
"""Return a side_effect for subprocess.run that fakes `docker exec ... cat <path>`.
Args:
outputs: mapping from container path to (returncode, stdout) tuples.
"""
def side_effect(cmd, **kwargs):
# Detect "docker exec <cid> cat <path>" calls.
if cmd[:2] == ["docker", "exec"] and "cat" in cmd:
path = cmd[-1]
if path in outputs:
rc, stdout = outputs[path]
return MagicMock(returncode=rc, stdout=stdout)
return MagicMock(returncode=1, stdout="")
return side_effect
class TestReadReward:
def _make_inst(self):
inst = object.__new__(HarborEvaluator)
inst.container_id = "fake_container"
return inst
def test_reads_reward_txt(self):
inst = self._make_inst()
with patch("subprocess.run", side_effect=_mock_docker_exec({
"/logs/verifier/reward.json": (1, ""),
"/logs/verifier/reward.txt": (0, "0.75\n"),
})):
result = inst._read_reward()
assert result.metrics["combined_score"] == 0.75
def test_reads_reward_json_with_reward_key(self):
inst = self._make_inst()
payload = json.dumps({"reward": 0.9, "time_ms": 123})
with patch("subprocess.run", side_effect=_mock_docker_exec({
"/logs/verifier/reward.json": (0, payload),
})):
result = inst._read_reward()
assert result.metrics["combined_score"] == 0.9
assert result.metrics["time_ms"] == 123.0
def test_reads_reward_json_with_score_key(self):
inst = self._make_inst()
payload = json.dumps({"score": 0.5})
with patch("subprocess.run", side_effect=_mock_docker_exec({
"/logs/verifier/reward.json": (0, payload),
})):
result = inst._read_reward()
assert result.metrics["combined_score"] == 0.5
def test_json_preferred_over_txt(self):
inst = self._make_inst()
payload = json.dumps({"reward": 0.9})
with patch("subprocess.run", side_effect=_mock_docker_exec({
"/logs/verifier/reward.json": (0, payload),
"/logs/verifier/reward.txt": (0, "0.1\n"),
})):
result = inst._read_reward()
assert result.metrics["combined_score"] == 0.9
def test_missing_reward_key_defaults_to_zero(self):
inst = self._make_inst()
payload = json.dumps({"time_ms": 500})
with patch("subprocess.run", side_effect=_mock_docker_exec({
"/logs/verifier/reward.json": (0, payload),
})):
result = inst._read_reward()
assert result.metrics["combined_score"] == 0.0
def test_no_reward_files_returns_zero(self):
inst = self._make_inst()
with patch("subprocess.run", side_effect=_mock_docker_exec({
"/logs/verifier/reward.json": (1, ""),
"/logs/verifier/reward.txt": (1, ""),
})):
result = inst._read_reward()
assert result.metrics["combined_score"] == 0.0
assert "error" in result.artifacts
def test_malformed_json_falls_back_to_txt(self):
inst = self._make_inst()
with patch("subprocess.run", side_effect=_mock_docker_exec({
"/logs/verifier/reward.json": (0, "{bad json"),
"/logs/verifier/reward.txt": (0, "0.42\n"),
})):
result = inst._read_reward()
assert result.metrics["combined_score"] == 0.42
def test_non_numeric_txt_falls_through(self):
inst = self._make_inst()
with patch("subprocess.run", side_effect=_mock_docker_exec({
"/logs/verifier/reward.json": (1, ""),
"/logs/verifier/reward.txt": (0, "not a number"),
})):
result = inst._read_reward()
assert result.metrics["combined_score"] == 0.0
assert "error" in result.artifacts
# ------------------------------------------------------------------
# Harbor task detection
# ------------------------------------------------------------------
def _make_harbor_dir(tmp_path):
"""Create a minimal valid Harbor task directory."""
(tmp_path / "instruction.md").write_text("problem")
(tmp_path / "tests").mkdir()
(tmp_path / "tests" / "test.sh").write_text("#!/bin/bash\n")
(tmp_path / "environment").mkdir()
(tmp_path / "environment" / "Dockerfile").write_text("FROM python:3.11\n")
return str(tmp_path)
class TestHarborTaskDetection:
def test_valid_harbor_task(self, tmp_path):
assert _is_harbor_task(_make_harbor_dir(tmp_path)) is True
def test_missing_instruction_md(self, tmp_path):
(tmp_path / "tests").mkdir()
(tmp_path / "environment").mkdir()
(tmp_path / "environment" / "Dockerfile").write_text("FROM python:3.11\n")
assert _is_harbor_task(str(tmp_path)) is False
def test_missing_tests_dir(self, tmp_path):
(tmp_path / "instruction.md").write_text("problem")
(tmp_path / "environment").mkdir()
(tmp_path / "environment" / "Dockerfile").write_text("FROM python:3.11\n")
assert _is_harbor_task(str(tmp_path)) is False
def test_missing_test_sh(self, tmp_path):
(tmp_path / "instruction.md").write_text("problem")
(tmp_path / "tests").mkdir()
(tmp_path / "environment").mkdir()
(tmp_path / "environment" / "Dockerfile").write_text("FROM python:3.11\n")
assert _is_harbor_task(str(tmp_path)) is False
def test_missing_dockerfile(self, tmp_path):
(tmp_path / "instruction.md").write_text("problem")
(tmp_path / "tests").mkdir()
(tmp_path / "tests" / "test.sh").write_text("#!/bin/bash\n")
(tmp_path / "environment").mkdir()
assert _is_harbor_task(str(tmp_path)) is False
def test_not_a_directory(self, tmp_path):
f = tmp_path / "not_a_dir"
f.write_text("hi")
assert _is_harbor_task(str(f)) is False
class TestContainerizedDetection:
def test_valid_containerized(self, tmp_path):
(tmp_path / "Dockerfile").write_text("FROM python:3.11\n")
(tmp_path / "evaluate.sh").write_text("#!/bin/bash\n")
assert _is_containerized(str(tmp_path)) is True
def test_missing_evaluate_sh(self, tmp_path):
(tmp_path / "Dockerfile").write_text("FROM python:3.11\n")
assert _is_containerized(str(tmp_path)) is False
def test_missing_dockerfile(self, tmp_path):
(tmp_path / "evaluate.sh").write_text("#!/bin/bash\n")
assert _is_containerized(str(tmp_path)) is False
class TestDetectionPriority:
"""A dir that matches both Harbor and containerized should be detected as Harbor."""
def test_harbor_wins_over_containerized(self, tmp_path):
# Set up Harbor structure.
_make_harbor_dir(tmp_path)
# Also add containerized markers at root.
(tmp_path / "Dockerfile").write_text("FROM python:3.11\n")
(tmp_path / "evaluate.sh").write_text("#!/bin/bash\n")
assert _is_harbor_task(str(tmp_path)) is True
assert _is_containerized(str(tmp_path)) is True
# create_evaluator checks harbor first — verify the detection functions
# agree that both match, confirming the ordering in create_evaluator matters.
|