## Summary
Reported as "deliver_files is broken". It is not: **the tool never ran.**
Every `deliver_files` call on one install was refused before dispatch -- five in a row in a single session -- and nothing was ever delivered:
```
Error: Invalid parameters for tool 'deliver_files': files should be array
```
What the model sent was the array, spelled as a string:
```json
{"files": "[{\"path\": \"...poster.png\", \"title\": \"...\"}]", "message": "..."}
```
`cast_params` already forgives that spelling for every scalar type -- `"3"` for an integer, `"true"` for a boolean -- and did not for arrays or objects, which are exactly the types a model is most likely to stringify.
## Type
- [x] Fix
- [ ] Feature
## The edge of the forgiving is the design
When the value is a string, try `json.loads`, and take the result **only if it is the type the schema asked for**. A string that is not JSON, or is JSON of the wrong shape, is returned unchanged so the validator still names the real problem -- the point is to widen what is accepted, not to swallow a type error.
**Nothing here can raise.** Turning "your argument has the wrong type" into a traceback would be worse than the refusal it replaces, so the parse is wrapped and failure falls through.
**A `type: "string"` field is never parsed.** JSON is ordinary text in a message body or a file's content, and replacing it would hand the tool something it never asked for. Two independent things prevent it today (the string branch returns before the parse is reached, and the parse only wins on a type match), which is why the case pinning it is a tripwire rather than a load-bearing assertion -- said so in its own docstring rather than left to look like coverage.
## Not just deliver_files
Six built-in tools declare an array or object parameter and were all exposed:
| tool | parameter |
|---|---|
| `deliver_files` | `files` |
| `ask_user` | `questions` |
| `image_generate` | `images` |
| `message` | `media` |
| `create_playbook` | `skills` |
| `video_generate` | `params` |
The same failure was already seen from the other side: `acp_client/acp_agent.py` records `update_task_state` and `operations should be array` from a real dispatch.
## Verification
```
make lint-python ruff check + ruff format --check, 1850 files
make lint-imports 9 contracts kept, 0 broken
pytest tests/ 19,699 passed, 7 failed (all pre-existing, see below)
```
- [x] Relevant tests pass locally
- [x] Relevant lint / type checks pass locally
- [ ] User-facing docs or screenshots -- no user-facing string
**End to end, through the real tool**, with the exact payload that failed five times in production:
```
files sent as: str -> [{"path": ".../poster.png", "title": "..."}]
returned: Delivered 1 file(s): poster.png (72 B).
still "should be array": False
registered in the store: True
```
Twelve cases, five mutations, each killing at least one:
| mutation | red |
|---|---|
| drop the whole branch (the old behaviour) | 3 |
| raise instead of falling through on a parse failure | 5 |
| accept the parse whatever type it yielded | 1 |
| do not re-cast the parsed value, so items are not cast | 1 |
| widen the tuple to include `string` | **survived -- unreachable** |
The last is worth stating rather than hiding: that mutation cannot change behaviour, because the `type == "string"` branch returns earlier and the type check would refuse the result anyway. I also tried moving the new block above that branch, alone and combined with widening the tuple, and neither reddens anything either. The invariant holds by construction; the case that documents it says so.
**Pre-existing failures, checked against the base.** 7 failed in the full run, all in `tests/test_security_network.py` (`test_a_redirect_to_a_dangerous_address_is_refused` and siblings). Running that file alone on this same tree gives 4 -- the count is order-dependent -- and the change touches only `raven/agent/tools/params.py` plus a new test file, neither of which `raven/security/network.py` imports. The failures are the box's resolver: it returns poisoned answers roughly half the time, which I measured separately (`en.wikipedia.org` came back as `2001::1` on 3 of 6 lookups). Not introduced here, and worth someone's attention on its own.
## Risk
- [x] Security impact considered -- widens what reaches a tool's own validation, not what bypasses it. A parsed value is still validated against the same schema, and a parse that yields the wrong type is refused exactly as before
- [x] Backward compatibility considered -- a call that worked before takes the identical path; the new branch is reached only by a `str` where an array or object is declared, which previously always failed
- [x] Rollback path is clear -- single-commit revert
## Related Issues
N/A
## Swept
`mr-review-patterns` pre-submit sweep, over `git diff origin/main...HEAD`.
- **1 (a write location moved)** -- nothing moved; a branch was added to an existing dispatch on `target_type`. Grepped every caller of `cast_params`: the registry is the only one, and it runs `validate_params` right after, which is the reader this affects.
- **2 (a value with several entry points)** -- the cast is the single funnel every tool argument passes through, which is why the fix belongs here and not in `deliver_files`.
- **3 (a guard)** -- checked both directions: what it wrongly lets through (JSON of the wrong type, non-JSON text -- both refused, with cases) and what it wrongly refuses (a real list, unaffected, with a case).
- **4 (tests)** -- five mutations above, one honestly reported as unreachable. Failing-ID sets diffed against the base rather than counts.
- **6 (stale text)** -- the module docstring enumerated what is coerced ("a string that should be an integer"); it now names arrays and objects too. Grepped the tracked tree for other statements of the rule: none.
- **9 (error handling)** -- the new path adds one way to end and it is caught: `json.loads` raises `ValueError` on bad text and `TypeError` on a non-string, and both fall through to the untouched value.
- **11** -- `git ls-files | grep -c node_modules` is 0; the commit-message gate was run, not eyeballed.
Not applicable, with reasons: **3b, 3c, 3d, 3e** (no tool schema produced, no state moved, one lane, no assertion lost its symbol); **4b, 4c, 4d** (no platform branch, no shell, no rendered number); **5, 5b** (no dependency; `json` is stdlib and already imported across this package); **6b, 6c** (no backup, no generated artifact); **7, 7b** (nothing on disk); **8** (no asset, no secret); **10** (no prompt text).
## Deliberately not fixed
The error text still ends with `[Analyze the error above and try a different approach.]`, and the model plainly did not -- it sent the same shape five times. A message naming the expected spelling (`files must be a JSON array, not a string containing one`) would give it something to act on. That is a change to the registry's error surface, affects every tool, and deserves its own argument.
## Moved from !594
Opened on a compliant source branch. AGENTS.md 2.1 asks for a 3-5 word snake_case description and the previous branch had seven; GitLab cannot change an existing merge request's source branch, so the change moves rather than the name.
Same commits, rebased onto `c2a9768f`. Both code findings from the review on !594 are already in them -- `RecursionError` caught with its regression case, and the implementation comment reduced to the two non-obvious reasons. Nothing else changed, so that review carries.
532 lines
21 KiB
Python
532 lines
21 KiB
Python
"""Unit tests for the unified evolution launcher (evolver.launch)."""
|
|
|
|
from __future__ import annotations
|
|
|
|
import json
|
|
import subprocess
|
|
import sys
|
|
from pathlib import Path
|
|
|
|
import pytest
|
|
import yaml
|
|
|
|
from evolver.launch.config import (
|
|
SMOKE_BUILTIN,
|
|
RunSpecError,
|
|
deep_merge,
|
|
load_run_spec,
|
|
)
|
|
from evolver.launch.contract import validate_whitelist
|
|
from evolver.launch.registry import load_bench
|
|
from evolver.launch.state import RunMeta, atomic_write_json, config_fingerprint
|
|
|
|
# The appworld bench plugin lives at the repo root (benchmarks/), outside the
|
|
# installed raven package; load_bench imports it via the subject repo root,
|
|
# and the direct `import benchmarks...` statements below need it on sys.path.
|
|
REPO_ROOT = Path(__file__).resolve().parents[1]
|
|
if str(REPO_ROOT) not in sys.path:
|
|
sys.path.insert(0, str(REPO_ROOT))
|
|
|
|
|
|
@pytest.fixture()
|
|
def subject_repo(tmp_path: Path) -> tuple[Path, str]:
|
|
"""A tiny git repo standing in for the evolved subject."""
|
|
repo = tmp_path / "subject"
|
|
(repo / "raven/agent").mkdir(parents=True)
|
|
(repo / "raven/agent/loop.py").write_text("x = 1\n")
|
|
env = {"GIT_AUTHOR_NAME": "t", "GIT_AUTHOR_EMAIL": "t@t", "GIT_COMMITTER_NAME": "t", "GIT_COMMITTER_EMAIL": "t@t"}
|
|
for cmd in (["git", "init", "-q"], ["git", "add", "-A"], ["git", "commit", "-qm", "init"]):
|
|
subprocess.run(cmd, cwd=repo, check=True, env={**env, "PATH": "/usr/bin:/bin"}, capture_output=True)
|
|
sha = subprocess.run(
|
|
["git", "rev-parse", "HEAD"], cwd=repo, check=True, capture_output=True, text=True
|
|
).stdout.strip()
|
|
return repo, sha
|
|
|
|
|
|
def _write_spec(tmp_path: Path, repo: Path, sha: str, **extra) -> Path:
|
|
data = {
|
|
"bench": "appworld",
|
|
"repo_root": str(repo),
|
|
"base_sha": sha,
|
|
"work_dir": str(tmp_path / "work"),
|
|
**extra,
|
|
}
|
|
path = tmp_path / "spec.yaml"
|
|
path.write_text(yaml.safe_dump(data))
|
|
return path
|
|
|
|
|
|
class TestRunSpec:
|
|
def test_minimal_spec_loads_with_defaults(self, tmp_path, subject_repo):
|
|
repo, sha = subject_repo
|
|
spec = load_run_spec(_write_spec(tmp_path, repo, sha))
|
|
assert spec.bench == "appworld"
|
|
assert spec.funnel.k_confirm == 3
|
|
assert spec.funnel.termination.patience == 10
|
|
assert spec.funnel.sealed_output_dir == spec.work_dir / "sealed"
|
|
|
|
def test_omitted_base_sha_resolves_to_head(self, tmp_path, subject_repo):
|
|
repo, sha = subject_repo
|
|
path = tmp_path / "spec.yaml"
|
|
path.write_text(
|
|
yaml.safe_dump(
|
|
{
|
|
"bench": "appworld",
|
|
"repo_root": str(repo),
|
|
"work_dir": str(tmp_path / "work"),
|
|
}
|
|
)
|
|
)
|
|
spec = load_run_spec(path)
|
|
assert spec.base_sha == sha
|
|
assert spec.base_sha_defaulted is True
|
|
assert spec.snapshot()["base_sha"] == sha
|
|
|
|
def test_explicit_base_sha_kept_verbatim(self, tmp_path, subject_repo):
|
|
repo, sha = subject_repo
|
|
spec = load_run_spec(_write_spec(tmp_path, repo, sha[:7]))
|
|
assert spec.base_sha == sha[:7]
|
|
assert spec.base_sha_defaulted is False
|
|
|
|
def test_missing_required_keys(self, tmp_path):
|
|
path = tmp_path / "bad.yaml"
|
|
path.write_text(yaml.safe_dump({"bench": "appworld"}))
|
|
with pytest.raises(RunSpecError, match="missing required"):
|
|
load_run_spec(path)
|
|
|
|
def test_funnel_scalar_rejected(self, tmp_path, subject_repo):
|
|
repo, sha = subject_repo
|
|
path = _write_spec(tmp_path, repo, sha, funnel=5)
|
|
with pytest.raises(RunSpecError, match="must be a mapping"):
|
|
load_run_spec(path)
|
|
|
|
def test_funnel_bad_value_type_readable(self, tmp_path, subject_repo):
|
|
repo, sha = subject_repo
|
|
path = _write_spec(tmp_path, repo, sha, funnel={"k_confirm": "three"})
|
|
with pytest.raises(RunSpecError, match="funnel"):
|
|
load_run_spec(path)
|
|
|
|
def test_funnel_nonpositive_rejected(self, tmp_path, subject_repo):
|
|
repo, sha = subject_repo
|
|
path = _write_spec(tmp_path, repo, sha, funnel={"termination": {"patience": 0}})
|
|
with pytest.raises(RunSpecError, match=">= 1"):
|
|
load_run_spec(path)
|
|
|
|
def test_model_role_scalar_rejected(self, tmp_path, subject_repo):
|
|
repo, sha = subject_repo
|
|
path = _write_spec(tmp_path, repo, sha, models={"driver": "claude"})
|
|
with pytest.raises(RunSpecError, match="models.driver"):
|
|
load_run_spec(path)
|
|
|
|
def test_relative_paths_resolve_against_config_dir(self, tmp_path, subject_repo, monkeypatch):
|
|
repo, sha = subject_repo
|
|
path = tmp_path / "spec.yaml"
|
|
path.write_text(
|
|
yaml.safe_dump(
|
|
{
|
|
"bench": "appworld",
|
|
"repo_root": str(repo),
|
|
"base_sha": sha,
|
|
"work_dir": "evo_work",
|
|
}
|
|
)
|
|
)
|
|
elsewhere = tmp_path / "elsewhere"
|
|
elsewhere.mkdir()
|
|
monkeypatch.chdir(elsewhere)
|
|
spec = load_run_spec(path)
|
|
assert spec.work_dir == (tmp_path / "evo_work").resolve()
|
|
|
|
def test_unknown_funnel_key_rejected(self, tmp_path, subject_repo):
|
|
repo, sha = subject_repo
|
|
path = _write_spec(tmp_path, repo, sha, funnel={"k_confrim": 3})
|
|
with pytest.raises(RunSpecError, match="unknown keys"):
|
|
load_run_spec(path)
|
|
|
|
def test_unknown_model_role_rejected(self, tmp_path, subject_repo):
|
|
repo, sha = subject_repo
|
|
path = _write_spec(tmp_path, repo, sha, models={"designer": {}})
|
|
with pytest.raises(RunSpecError, match="unknown roles"):
|
|
load_run_spec(path)
|
|
|
|
def test_smoke_applies_builtin_shrink_then_user_overlay(self, tmp_path, subject_repo):
|
|
repo, sha = subject_repo
|
|
path = _write_spec(
|
|
tmp_path,
|
|
repo,
|
|
sha,
|
|
funnel={"k_confirm": 3},
|
|
smoke={"funnel": {"termination": {"max_rounds": 2}}},
|
|
)
|
|
spec = load_run_spec(path, smoke=True)
|
|
assert spec.funnel.k_confirm == SMOKE_BUILTIN["funnel"]["k_confirm"]
|
|
assert spec.funnel.budget.max_why_per_round == 1
|
|
assert spec.funnel.termination.max_rounds == 2 # user overlay wins
|
|
assert spec.work_dir.name.endswith("_smoke")
|
|
|
|
def test_non_smoke_ignores_smoke_section(self, tmp_path, subject_repo):
|
|
repo, sha = subject_repo
|
|
path = _write_spec(tmp_path, repo, sha, smoke={"funnel": {"k_confirm": 1}})
|
|
spec = load_run_spec(path)
|
|
assert spec.funnel.k_confirm == 3
|
|
assert "smoke" not in spec.raw
|
|
|
|
def test_deep_merge_nested(self):
|
|
merged = deep_merge({"a": {"b": 1, "c": 2}}, {"a": {"b": 9}, "d": 3})
|
|
assert merged == {"a": {"b": 9, "c": 2}, "d": 3}
|
|
|
|
|
|
class TestRunMeta:
|
|
def test_create_load_roundtrip(self, tmp_path):
|
|
meta = RunMeta.create(tmp_path, {"bench": "x"})
|
|
again = RunMeta.load(tmp_path)
|
|
assert again is not None
|
|
assert again.config_hash == meta.config_hash
|
|
assert again.unsealed_at is None
|
|
|
|
def test_config_drift_detected(self, tmp_path):
|
|
meta = RunMeta.create(tmp_path, {"bench": "x", "funnel": {"k": 3}})
|
|
assert meta.check_config({"bench": "x", "funnel": {"k": 3}})
|
|
assert not meta.check_config({"bench": "x", "funnel": {"k": 1}})
|
|
|
|
def test_unseal_stamp_is_persisted(self, tmp_path):
|
|
meta = RunMeta.create(tmp_path, {})
|
|
meta.stamp_unsealed(reason="user_finalized")
|
|
again = RunMeta.load(tmp_path)
|
|
assert again.unsealed_at
|
|
assert again.finalize_reason == "user_finalized"
|
|
|
|
def test_fingerprint_is_order_insensitive(self):
|
|
assert config_fingerprint({"a": 1, "b": 2}) == config_fingerprint({"b": 2, "a": 1})
|
|
|
|
def test_atomic_write_leaves_no_tmp_on_success(self, tmp_path):
|
|
target = tmp_path / "x.json"
|
|
atomic_write_json(target, {"ok": True})
|
|
assert json.loads(target.read_text()) == {"ok": True}
|
|
assert list(tmp_path.glob("*.tmp")) == []
|
|
|
|
|
|
class TestWhitelistValidation:
|
|
def test_live_prefix_passes(self, subject_repo):
|
|
repo, sha = subject_repo
|
|
validate_whitelist(repo, sha, ("raven/agent/",))
|
|
|
|
def test_dead_prefix_refuses(self, subject_repo):
|
|
repo, sha = subject_repo
|
|
with pytest.raises(ValueError, match="match no files"):
|
|
validate_whitelist(repo, sha, ("nonexistent/agent/",))
|
|
|
|
def test_empty_whitelist_refuses(self, subject_repo):
|
|
repo, sha = subject_repo
|
|
with pytest.raises(ValueError, match="empty"):
|
|
validate_whitelist(repo, sha, ())
|
|
|
|
|
|
class TestAppWorldEntry:
|
|
def _ctx(self, tmp_path, subject_repo, bench_config, smoke=False):
|
|
from evolver.launch.contract import LaunchContext
|
|
|
|
repo, sha = subject_repo
|
|
path = _write_spec(tmp_path, repo, sha, bench_config=bench_config)
|
|
spec = load_run_spec(path, smoke=smoke)
|
|
return LaunchContext(
|
|
spec=spec,
|
|
models={"driver": None, "design": None, "verdict": None},
|
|
)
|
|
|
|
def _bench_config(self, tmp_path):
|
|
cfg = tmp_path / "subject_cfg.json"
|
|
cfg.write_text("{}")
|
|
(tmp_path / "appworld" / "data" / "tasks").mkdir(parents=True, exist_ok=True)
|
|
bin_dir = tmp_path / "appworld" / "appworld-venv" / "bin"
|
|
bin_dir.mkdir(parents=True, exist_ok=True)
|
|
(bin_dir / "appworld").write_text("#!/bin/sh\n")
|
|
return {
|
|
"config_path": str(cfg),
|
|
"appworld_data_root": str(tmp_path / "appworld"),
|
|
"train_task_ids": ["t1", "t2"],
|
|
"whitelist": ["raven/agent/"],
|
|
}
|
|
|
|
def test_build_bundle_shape(self, tmp_path, subject_repo):
|
|
build = load_bench("appworld", repo_root=REPO_ROOT)
|
|
bundle = build(self._ctx(tmp_path, subject_repo, self._bench_config(tmp_path)))
|
|
assert bundle.root_node_id == "C0"
|
|
assert bundle.cold_start_total == 2 * 3 # 2 tasks x k_confirm=3
|
|
assert bundle.cold_start_done() == 0
|
|
assert bundle.unseal is None # no test ids -> no sealed test
|
|
|
|
def test_cold_start_counts_existing_trials(self, tmp_path, subject_repo):
|
|
build = load_bench("appworld", repo_root=REPO_ROOT)
|
|
ctx = self._ctx(tmp_path, subject_repo, self._bench_config(tmp_path))
|
|
bundle = build(ctx)
|
|
van = ctx.spec.work_dir / "runs" / "vanilla"
|
|
van.mkdir(parents=True)
|
|
(van / "t1_k0.json").write_text("{}")
|
|
(van / "t1_k1.json").write_text("{}")
|
|
assert bundle.cold_start_done() == 2
|
|
|
|
def test_unknown_bench_config_key_rejected(self, tmp_path, subject_repo):
|
|
build = load_bench("appworld", repo_root=REPO_ROOT)
|
|
bc = self._bench_config(tmp_path)
|
|
bc["train_tasks"] = ["x"]
|
|
with pytest.raises(ValueError, match="unknown keys"):
|
|
build(self._ctx(tmp_path, subject_repo, bc))
|
|
|
|
def test_train_test_overlap_rejected(self, tmp_path, subject_repo):
|
|
build = load_bench("appworld", repo_root=REPO_ROOT)
|
|
bc = self._bench_config(tmp_path)
|
|
bc["test_task_ids"] = ["t2", "t3"]
|
|
with pytest.raises(ValueError, match="overlap"):
|
|
build(self._ctx(tmp_path, subject_repo, bc))
|
|
|
|
def test_placeholder_task_ids_refused(self, tmp_path, subject_repo):
|
|
build = load_bench("appworld", repo_root=REPO_ROOT)
|
|
bc = self._bench_config(tmp_path)
|
|
bc["train_task_ids"] = ["<failing-task-1>", "t2"]
|
|
with pytest.raises(ValueError, match="placeholder"):
|
|
build(self._ctx(tmp_path, subject_repo, bc))
|
|
|
|
def test_missing_appworld_install_refuses_at_build(self, tmp_path, subject_repo, monkeypatch):
|
|
build = load_bench("appworld", repo_root=REPO_ROOT)
|
|
bc = self._bench_config(tmp_path)
|
|
bc["appworld_data_root"] = str(tmp_path / "nowhere")
|
|
monkeypatch.delenv("APPWORLD_ROOT", raising=False)
|
|
with pytest.raises(ValueError, match="no AppWorld install"):
|
|
build(self._ctx(tmp_path, subject_repo, bc))
|
|
|
|
def test_dead_whitelist_refuses_at_build(self, tmp_path, subject_repo):
|
|
build = load_bench("appworld", repo_root=REPO_ROOT)
|
|
bc = self._bench_config(tmp_path)
|
|
bc["whitelist"] = ["nonexistent/dir/"]
|
|
with pytest.raises(ValueError, match="match no files"):
|
|
build(self._ctx(tmp_path, subject_repo, bc))
|
|
|
|
|
|
class TestScorerImmutability:
|
|
def test_scorer_surface_is_immutable(self):
|
|
from evolver.applier.path_guard import check_patch_paths
|
|
|
|
offenders = check_patch_paths(
|
|
[
|
|
"benchmarks/appworld/evolve/grade.py",
|
|
"benchmarks/appworld/evolve/adapter.py",
|
|
"benchmarks/appworld/batch.py",
|
|
"evolver/orchestrator/gates/pipeline.py",
|
|
]
|
|
)
|
|
assert len(offenders) == 4
|
|
|
|
def test_agent_surface_is_editable(self):
|
|
from evolver.applier.path_guard import check_patch_paths
|
|
|
|
assert (
|
|
check_patch_paths(
|
|
[
|
|
"benchmarks/appworld/agent_cli.py",
|
|
"benchmarks/appworld/tool.py",
|
|
]
|
|
)
|
|
== []
|
|
)
|
|
|
|
|
|
class TestSecretRedaction:
|
|
def _spec_with(self, tmp_path, repo, sha, **driver):
|
|
path = _write_spec(
|
|
tmp_path,
|
|
repo,
|
|
sha,
|
|
models={"driver": {"provider": "openai_compat", "base_url": "http://h/v1", "model": "m", **driver}},
|
|
)
|
|
return load_run_spec(path)
|
|
|
|
def test_api_key_never_reaches_snapshot(self, tmp_path, subject_repo):
|
|
repo, sha = subject_repo
|
|
snap = self._spec_with(tmp_path, repo, sha, api_key="sk-SECRET").snapshot()
|
|
assert "sk-SECRET" not in json.dumps(snap)
|
|
assert snap["models"]["driver"]["api_key"] == "<redacted>"
|
|
|
|
def test_fingerprint_stable_across_key_rotation(self, tmp_path, subject_repo):
|
|
repo, sha = subject_repo
|
|
f1 = config_fingerprint(self._spec_with(tmp_path, repo, sha, api_key="sk-AAA").snapshot())
|
|
f2 = config_fingerprint(self._spec_with(tmp_path, repo, sha, api_key="sk-BBB").snapshot())
|
|
f3 = config_fingerprint(self._spec_with(tmp_path, repo, sha, api_key="sk-AAA", model="other").snapshot())
|
|
assert f1 == f2
|
|
assert f1 != f3
|
|
|
|
|
|
class TestColdStartPrecheck:
|
|
def _bundle(self, tmp_path, subject_repo, monkeypatch, calls):
|
|
import benchmarks.appworld.evolve.entry as entry_mod
|
|
import benchmarks.appworld.evolve.precheck as precheck_mod
|
|
|
|
monkeypatch.setattr(
|
|
precheck_mod,
|
|
"make_appworld_precheck",
|
|
lambda cfg, **kw: lambda: calls.__setitem__("precheck", calls["precheck"] + 1),
|
|
)
|
|
monkeypatch.setattr(
|
|
entry_mod,
|
|
"eval_with_infra_rerun",
|
|
lambda *a, **k: calls.__setitem__("eval", calls["eval"] + 1),
|
|
)
|
|
helper = TestAppWorldEntry()
|
|
ctx = helper._ctx(tmp_path, subject_repo, helper._bench_config(tmp_path))
|
|
return load_bench("appworld", repo_root=REPO_ROOT)(ctx), ctx
|
|
|
|
def test_precheck_fires_before_fill_and_skips_when_clean(self, tmp_path, subject_repo, monkeypatch):
|
|
calls = {"precheck": 0, "eval": 0}
|
|
bundle, ctx = self._bundle(tmp_path, subject_repo, monkeypatch, calls)
|
|
|
|
bundle.run_cold_start()
|
|
assert calls == {"precheck": 1, "eval": 1}
|
|
|
|
van = ctx.spec.work_dir / "runs" / "vanilla"
|
|
van.mkdir(parents=True, exist_ok=True)
|
|
for tid in ("t1", "t2"):
|
|
for k in range(3):
|
|
(van / f"{tid}_k{k}.json").write_text(json.dumps({"task_id": tid, "success": True}))
|
|
bundle.run_cold_start()
|
|
assert calls["precheck"] == 1 # complete + infra-clean: no probe
|
|
assert calls["eval"] == 2
|
|
|
|
def test_precheck_fires_when_ladder_has_salvage_work(self, tmp_path, subject_repo, monkeypatch):
|
|
calls = {"precheck": 0, "eval": 0}
|
|
bundle, ctx = self._bundle(tmp_path, subject_repo, monkeypatch, calls)
|
|
van = ctx.spec.work_dir / "runs" / "vanilla"
|
|
van.mkdir(parents=True, exist_ok=True)
|
|
for tid in ("t1", "t2"):
|
|
for k in range(3):
|
|
rec = {"task_id": tid, "success": False}
|
|
if tid == "t2":
|
|
rec["infra_error"] = "runner: TimeoutExpired"
|
|
(van / f"{tid}_k{k}.json").write_text(json.dumps(rec))
|
|
bundle.run_cold_start()
|
|
assert calls["precheck"] == 1 # infra residue -> probe before reruns
|
|
|
|
|
|
class TestBatchTrialResume:
|
|
def test_existing_result_is_returned_without_rerun(self, tmp_path, monkeypatch):
|
|
from benchmarks.appworld import batch
|
|
|
|
out_dir = tmp_path
|
|
rec = {"task_id": "t1", "success": True}
|
|
(out_dir / "t1_k0.json").write_text(json.dumps(rec))
|
|
|
|
def boom(*a, **k): # noqa: ANN002, ANN003
|
|
raise AssertionError("subprocess must not run for a completed trial")
|
|
|
|
monkeypatch.setattr(batch.subprocess, "run", boom)
|
|
|
|
class Args:
|
|
config = "unused"
|
|
workspace = str(tmp_path / "chanwork")
|
|
experiment = "e"
|
|
model = None
|
|
env = ""
|
|
task_timeout = 1
|
|
|
|
got = batch._run_one("t1", 0, 9999, Args(), str(out_dir))
|
|
assert got == rec
|
|
|
|
def test_corrupt_result_is_rerun(self, tmp_path, monkeypatch):
|
|
from benchmarks.appworld import batch
|
|
|
|
(tmp_path / "t1_k0.json").write_text("{half written")
|
|
calls = []
|
|
|
|
def fake_run(*a, **k): # noqa: ANN002, ANN003
|
|
calls.append(1)
|
|
|
|
monkeypatch.setattr(batch.subprocess, "run", fake_run)
|
|
|
|
class Args:
|
|
config = "unused"
|
|
workspace = str(tmp_path / "chanwork")
|
|
experiment = "e"
|
|
model = None
|
|
env = ""
|
|
task_timeout = 1
|
|
|
|
got = batch._run_one("t1", 0, 9999, Args(), str(tmp_path))
|
|
assert calls, "corrupt file must trigger a re-run"
|
|
assert got.get("infra_error") # agent_cli never wrote a fresh result
|
|
|
|
|
|
class TestCli:
|
|
def test_parser_subcommands(self):
|
|
from evolver.cli import build_parser
|
|
|
|
p = build_parser()
|
|
args = p.parse_args(["run", "--config", "x.yaml", "--smoke", "--force"])
|
|
assert args.command == "run" and args.smoke and args.force
|
|
args = p.parse_args(["finalize", "--config", "x.yaml", "--yes"])
|
|
assert args.command == "finalize" and args.yes
|
|
|
|
def test_status_on_fresh_dir_reports_not_started(self, tmp_path, subject_repo, capsys):
|
|
from evolver.launch.runner import cmd_status
|
|
|
|
repo, sha = subject_repo
|
|
path = _write_spec(tmp_path, repo, sha)
|
|
assert cmd_status(str(path)) == 0
|
|
assert "not started" in capsys.readouterr().out
|
|
|
|
def _buildable_spec(self, tmp_path, repo, sha):
|
|
"""A spec whose bundle+models build cleanly, so guard tests reach
|
|
the meta guard (which now runs after bundle validation)."""
|
|
cfg = tmp_path / "subject_cfg.json"
|
|
cfg.write_text("{}")
|
|
(tmp_path / "appworld" / "data" / "tasks").mkdir(parents=True, exist_ok=True)
|
|
bin_dir = tmp_path / "appworld" / "appworld-venv" / "bin"
|
|
bin_dir.mkdir(parents=True, exist_ok=True)
|
|
(bin_dir / "appworld").write_text("#!/bin/sh\n")
|
|
return _write_spec(
|
|
tmp_path,
|
|
repo,
|
|
sha,
|
|
models={"driver": {"provider": "openai_compat", "base_url": "http://localhost:1/v1", "model": "m"}},
|
|
bench_config={
|
|
"config_path": str(cfg),
|
|
"appworld_data_root": str(tmp_path / "appworld"),
|
|
"train_task_ids": ["t1"],
|
|
"whitelist": ["raven/agent/"],
|
|
},
|
|
)
|
|
|
|
def test_run_refuses_after_unseal(self, tmp_path, subject_repo, capsys):
|
|
from evolver.launch.runner import cmd_run
|
|
|
|
repo, sha = subject_repo
|
|
path = self._buildable_spec(tmp_path, repo, sha)
|
|
work = tmp_path / "work"
|
|
work.mkdir()
|
|
meta = RunMeta.create(work, {})
|
|
meta.stamp_unsealed(reason="user_finalized")
|
|
with pytest.raises(SystemExit) as exc:
|
|
cmd_run(str(path))
|
|
assert exc.value.code == 2
|
|
assert "unsealed" in capsys.readouterr().err
|
|
|
|
def test_run_refuses_on_config_drift(self, tmp_path, subject_repo, capsys):
|
|
from evolver.launch.runner import cmd_run
|
|
|
|
repo, sha = subject_repo
|
|
path = self._buildable_spec(tmp_path, repo, sha)
|
|
work = tmp_path / "work"
|
|
work.mkdir()
|
|
RunMeta.create(work, {"bench": "appworld", "different": True})
|
|
with pytest.raises(SystemExit) as exc:
|
|
cmd_run(str(path))
|
|
assert exc.value.code == 2
|
|
assert "config drift" in capsys.readouterr().err
|
|
|
|
def test_first_launch_config_mistake_leaves_no_meta(self, tmp_path, subject_repo):
|
|
from evolver.launch.runner import cmd_run
|
|
|
|
repo, sha = subject_repo
|
|
path = _write_spec(tmp_path, repo, sha) # no bench_config -> build fails
|
|
with pytest.raises(SystemExit):
|
|
cmd_run(str(path))
|
|
assert RunMeta.load(tmp_path / "work") is None
|