feat: bound autonomous release command runtime (#329)

tests/test_release_engine.py

@@ -1,6 +1,8 @@
 import json
 import sys
 import threading
+import time
+from dataclasses import replace
 from http.server import BaseHTTPRequestHandler, HTTPServer
 from pathlib import Path
 
@@ -15,6 +17,7 @@ from src.release_engine import (
     RunState,
     ShellRunner,
     StateStore,
+    _parse_args,
     branch_name,
     select_issue,
 )
@@ -41,6 +44,33 @@ def test_shell_runner_treats_metacharacters_as_literal_arguments(tmp_path):
     assert not injected.exists()
 
 
+def test_shell_runner_times_out_and_terminates_spawned_children(tmp_path):
+    marker = tmp_path / "child-survived.txt"
+    child = (
+        "import time; from pathlib import Path; "
+        f"time.sleep(0.3); Path({str(marker)!r}).write_text('survived')"
+    )
+    parent = (
+        "import subprocess, sys, time; "
+        f"subprocess.Popen([sys.executable, '-c', {child!r}]); "
+        "time.sleep(10)"
+    )
+
+    started = time.monotonic()
+    result = ShellRunner().run(
+        [sys.executable, "-c", parent],
+        tmp_path,
+        timeout_seconds=0.05,
+        termination_grace_seconds=0.05,
+    )
+    elapsed = time.monotonic() - started
+    time.sleep(0.35)
+
+    assert result.timed_out is True
+    assert elapsed < 0.5
+    assert not marker.exists()
+
+
 def test_stale_token_error_explains_how_to_restore_cron_authentication():
     class UnauthorizedHandler(BaseHTTPRequestHandler):
         def do_GET(self):
@@ -94,10 +124,12 @@ class FakeRunner:
         self.results = list(results)
         self.commands = []
         self.environments = []
+        self.options = []
 
-    def run(self, command, cwd, env=None):
+    def run(self, command, cwd, env=None, **options):
         self.commands.append(command)
         self.environments.append(env)
+        self.options.append(options)
         return self.results.pop(0)
 
 
@@ -164,6 +196,41 @@ def test_empty_agent_command_fails_before_claim(tmp_path):
     assert client.claimed == []
 
 
+@pytest.mark.parametrize(
+    ("field", "value"),
+    [
+        ("agent_timeout_seconds", 0),
+        ("test_timeout_seconds", -1),
+        ("termination_grace_seconds", 0),
+    ],
+)
+def test_non_positive_command_deadlines_fail_before_claim(tmp_path, field, value):
+    issue = Issue(14, "Engine", "body", "open", [], None)
+    client = FakeClient([issue])
+    cfg = replace(config(tmp_path), **{field: value})
+
+    with pytest.raises(ValueError, match=field.replace("_", " ")):
+        ReleaseEngine(cfg, client, FakeRunner([])).execute(issue_number=14)
+
+    assert client.claimed == []
+
+
+def test_command_deadlines_are_configurable_from_cli(monkeypatch):
+    monkeypatch.setattr(sys, "argv", [
+        "release-engine",
+        "--repo", "stackchain/stackchain-dashboard",
+        "--agent-timeout", "120",
+        "--test-timeout", "45.5",
+        "--termination-grace", "2",
+    ])
+
+    args = _parse_args()
+
+    assert args.agent_timeout == 120
+    assert args.test_timeout == 45.5
+    assert args.termination_grace == 2
+
+
 def test_agent_failure_blocks_tests_and_pr(tmp_path):
     issue = Issue(14, "Engine", "body", "open", [], None)
     client = FakeClient([issue])
@@ -183,6 +250,33 @@ def test_agent_failure_blocks_tests_and_pr(tmp_path):
     assert client.created_prs == []
 
 
+def test_agent_timeout_records_terminal_state_and_blocks_tests(tmp_path):
+    issue = Issue(14, "Engine", "body", "open", [], None)
+    client = FakeClient([issue])
+    runner = FakeRunner([
+        CommandResult(0, "", ""),
+        CommandResult(-9, "partial output", "stalled", timed_out=True),
+    ])
+    cfg = config(tmp_path)
+
+    with pytest.raises(RuntimeError, match="coding agent timed out"):
+        ReleaseEngine(cfg, client, runner).execute(issue_number=14)
+
+    state = StateStore(cfg.state_path).load()
+    assert state is not None
+    assert state.status == "agent_timed_out"
+    assert "partial output" in state.evidence
+    assert runner.commands == [
+        ["git", "checkout", "-B", "timmy/14-engine"],
+        ["agent", "--issue-from-env"],
+    ]
+    assert runner.options[1] == {
+        "timeout_seconds": cfg.agent_timeout_seconds,
+        "termination_grace_seconds": cfg.termination_grace_seconds,
+    }
+    assert client.created_prs == []
+
+
 def test_issue_content_reaches_agent_only_through_environment(tmp_path):
     title = "$(touch /tmp/title-injected); `id`"
     body = "line one\n&& touch /tmp/body-injected > stolen"
@@ -229,6 +323,31 @@ def test_failed_tests_block_push_and_pr(tmp_path):
     assert client.created_prs == []
 
 
+def test_test_timeout_records_terminal_state_and_blocks_push(tmp_path):
+    issue = Issue(14, "Engine", "body", "open", [], None)
+    client = FakeClient([issue])
+    runner = FakeRunner([
+        CommandResult(0, "", ""),
+        CommandResult(0, "implemented", ""),
+        CommandResult(-15, "3 passed before stall", "", timed_out=True),
+    ])
+    cfg = config(tmp_path)
+
+    with pytest.raises(RuntimeError, match="tests timed out"):
+        ReleaseEngine(cfg, client, runner).execute(issue_number=14)
+
+    state = StateStore(cfg.state_path).load()
+    assert state is not None
+    assert state.status == "tests_timed_out"
+    assert "3 passed before stall" in state.evidence
+    assert runner.options[2] == {
+        "timeout_seconds": cfg.test_timeout_seconds,
+        "termination_grace_seconds": cfg.termination_grace_seconds,
+    }
+    assert not any(command[:2] == ["git", "push"] for command in runner.commands)
+    assert client.created_prs == []
+
+
 def test_passing_run_pushes_and_opens_linked_pr(tmp_path):
     issue = Issue(14, "Engine MVP", "body", "open", ["P1"], None)
     client = FakeClient([issue])