fix: keep macos release gate awake

Author: ebursztein

CHANGELOG.md

@@ -7,6 +7,11 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Fixed
+- Kept unattended macOS candidate gates awake, ran the release benchmark once,
+  and preserved service/VM logs when exception teardown precedes pytest's
+  failure report.
+
 ## [1.5.1784663414] - 2026-07-21
 
 ### Fixed

justfile

@@ -656,6 +656,14 @@ _bootstrap:
 test:
     #!/bin/bash
     set -euo pipefail
+    if [ "$(uname -s)" = "Darwin" ] && [ -z "${CAPSEM_TEST_CAFFEINATED:-}" ]; then
+        command -v caffeinate >/dev/null || {
+            echo "macOS just test requires caffeinate to prevent an unattended release gate from sleeping" >&2
+            exit 1
+        }
+        echo "=== Holding macOS awake for the complete candidate gate ==="
+        exec caffeinate -dimsu env CAPSEM_TEST_CAFFEINATED=1 just test
+    fi
     if [ -n "$(git status --porcelain --untracked-files=all)" ]; then
         echo "just test requires a clean working tree; commit the complete candidate first." >&2
         git status --short >&2
@@ -874,7 +882,7 @@ _test-candidate: _bound-docker-test-storage _bootstrap _install-tools _clean-sta
     CAPSEM_REQUIRE_ARTIFACTS=1 uv run python -m pytest \
         tests/capsem-serial/ \
         tests/ironbank/test_route_health.py \
-        -v --tb=short -m serial
+        -v --tb=short -m serial -k 'not test_capsem_bench_baseline'
 
     echo "=== Python: Build chain and release tests (serial) ==="
     CAPSEM_REQUIRE_ARTIFACTS=1 uv run python -m pytest tests/capsem-build-chain/ tests/capsem-release/ -v --tb=short

tests/helpers/service.py

@@ -97,7 +97,7 @@ def materialize_test_profiles(tmp_dir: Path) -> Path:
     return profiles_dir
 
 
-def preserve_tmp_dir_on_failure(tmp_dir):
+def preserve_tmp_dir_on_failure(tmp_dir, *, force: bool = False):
     """Copy tmp_dir to test-artifacts/ when this worker saw any failure.
 
     Called by integration-test fixture teardowns BEFORE they rmtree the
@@ -132,7 +132,7 @@ def preserve_tmp_dir_on_failure(tmp_dir):
     # tmp_dir regardless of that worker's own failure state. Used during
     # concurrency investigations where a failure on worker B needs to be
     # correlated against what worker A was doing at the same time.
-    force = os.environ.get("CAPSEM_TEST_PRESERVE_ALWAYS")
+    force = force or bool(os.environ.get("CAPSEM_TEST_PRESERVE_ALWAYS"))
     if not force and not FAILED_NODEIDS:
         return
     import stat as statmod
@@ -346,7 +346,14 @@ def stop(self, *, cleanup: bool = True):
         if not cleanup:
             return
 
-        preserve_tmp_dir_on_failure(self.home_dir)
+        # Tests commonly stop the service from a ``finally`` block.  That
+        # happens before pytest's makereport hook records FAILED_NODEIDS, so
+        # use the actively-propagating exception as authoritative failure
+        # evidence instead of deleting the only service/process logs.
+        if sys.exc_info()[0] is not None:
+            preserve_tmp_dir_on_failure(self.home_dir, force=True)
+        else:
+            preserve_tmp_dir_on_failure(self.home_dir)
 
         if self.home_dir.exists():
             shutil.rmtree(self.home_dir, ignore_errors=True)

tests/test_release_gate_integrity.py

@@ -38,6 +38,25 @@ def test_just_test_binds_clean_tree_to_one_commit_without_archiving_benchmarks()
     assert "benchmarks/**/data_*.json" in _read(".gitignore")
 
 
+def test_full_gate_runs_capsem_bench_baseline_exactly_once() -> None:
+    justfile = _read("justfile")
+    candidate = justfile.split("\n_test-candidate:", maxsplit=1)[1].split(
+        "\nbuild-host-image:", maxsplit=1
+    )[0]
+
+    assert candidate.count("tests/capsem-serial/test_capsem_bench_baseline.py") == 1
+
+
+def test_macos_full_gate_holds_a_system_sleep_assertion() -> None:
+    justfile = _read("justfile")
+    wrapper = justfile.split("\ntest:", maxsplit=1)[1].split(
+        "\n_test-candidate:", maxsplit=1
+    )[0]
+
+    assert "caffeinate" in wrapper
+    assert "CAPSEM_TEST_CAFFEINATED" in wrapper
+
+
 def test_toolchain_and_workflow_inputs_are_immutable_and_consistent() -> None:
     toolchain = tomllib.loads(_read("rust-toolchain.toml"))
     assert toolchain["toolchain"]["channel"] == PINNED_RUST

tests/test_service_profile_helper.py

@@ -83,3 +83,28 @@ def test_service_instance_can_keep_shutdown_flushed_state_for_assertions(
 
     service.stop()
     assert not home.exists()
+
+
+def test_service_instance_preserves_artifacts_during_exception_teardown(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """A failure active inside ``finally`` must survive pre-report teardown."""
+    home = tmp_path / "capsem-home"
+    home.mkdir()
+    monkeypatch.setattr(service_helper, "make_capsem_tmp_dir", lambda _prefix: home)
+    preserved: list[tuple[Path, bool]] = []
+
+    def record_preserve(path: Path, *, force: bool = False) -> None:
+        preserved.append((Path(path), force))
+
+    monkeypatch.setattr(service_helper, "preserve_tmp_dir_on_failure", record_preserve)
+    service = service_helper.ServiceInstance()
+
+    with pytest.raises(RuntimeError, match="benchmark failed"):
+        try:
+            raise RuntimeError("benchmark failed")
+        finally:
+            service.stop()
+
+    assert preserved == [(home, True)]
+    assert not home.exists()