specfuse-loop 0.23.0__tar.gz → 0.23.2__tar.gz

{specfuse_loop-0.23.0/specfuse_loop.egg-info → specfuse_loop-0.23.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: specfuse-loop
-Version: 0.23.0
+Version: 0.23.2
 Summary: Local-first executor for the Specfuse Plan + Work Unit gate-cycle methodology.
 Author: Specfuse contributors
 License: Apache-2.0

{specfuse_loop-0.23.0 → specfuse_loop-0.23.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "specfuse-loop"
-version = "0.23.0"
+version = "0.23.2"
 description = "Local-first executor for the Specfuse Plan + Work Unit gate-cycle methodology."
 readme = "README.md"
 requires-python = ">=3.10"

specfuse_loop-0.23.2/specfuse/loop/data/VERSION

@@ -0,0 +1 @@
+0.23.2

{specfuse_loop-0.23.0 → specfuse_loop-0.23.2}/specfuse/loop/labels.py

@@ -194,8 +194,8 @@ LABEL_REGISTRY: tuple[LabelSpec, ...] = (
         name=bug_lane.DECLINE_LABELS[bug_lane.REASON_CI_PENDING],
         colour="fbca04",
         description=(
-            "Bug-lane auto-merge declined: CI was still pending at the "
-            "deadline; re-run the lane once it concludes"
+            "Bug-lane auto-merge declined: CI still pending at the deadline; "
+            "re-run once it concludes"
         ),
         consumer="loop/bug_lane_run.py",
     ),
@@ -203,8 +203,8 @@ LABEL_REGISTRY: tuple[LabelSpec, ...] = (
         name=bug_lane.DECLINE_LABELS[bug_lane.REASON_TEST_NOT_RED_ON_BASE],
         colour="fbca04",
         description=(
-            "Bug-lane auto-merge declined: the PR's new test passes at the "
-            "merge base, so it proves nothing about the fix"
+            "Bug-lane auto-merge declined: the new test passes at the merge "
+            "base, so it proves nothing"
         ),
         consumer="loop/bug_lane_run.py",
     ),
@@ -213,7 +213,7 @@ LABEL_REGISTRY: tuple[LabelSpec, ...] = (
         colour="fbca04",
         description=(
             "Bug-lane auto-merge declined: the red-on-base check was required "
-            "and could not be run; declare rules.bugs.red_on_base_command"
+            "and could not be run"
         ),
         consumer="loop/bug_lane_run.py",
     ),

{specfuse_loop-0.23.0 → specfuse_loop-0.23.2}/specfuse/loop/loop.py

@@ -125,7 +125,7 @@ SPECFUSE_DIR = Path(".specfuse")
 REPO_ROOT = SPECFUSE_DIR.parent
 FEATURES_DIR = SPECFUSE_DIR / "features"
 VERIFICATION_PATH = SPECFUSE_DIR / "verification.yml"
-DRIVER_VERSION = "0.23.0"
+DRIVER_VERSION = "0.23.2"
 # Oldest scaffold layout this driver can drive. init.sh stamps the scaffold's own
 # version into `.specfuse/VERSION`; check_scaffold_version() fails loud at startup if
 # the consumer's scaffold is older than this, pointing at `specfuse upgrade`. Bump

{specfuse_loop-0.23.0 → specfuse_loop-0.23.2}/specfuse/loop/red_on_base.py

@@ -27,6 +27,13 @@ Three deliberate constraints:
 * **Never guess the command.** The operator declares it. A guessed command that
   silently matched no tests would exit non-zero, read as "red on base", and
   wave every PR through -- strictly worse than the gap it was meant to close.
+* **Run it through a shell.** Operator-authored commands use pipes, `||`, `$(...)`
+  and inline environment overrides, exactly as the gate commands in
+  `verification.yml` do -- `_run_gate_set` runs those with `shell=True` for the
+  same stated reason. A first cut of this module used `shlex.split` and no
+  shell, which could not express the command of the very repository it was
+  built for: every merge would have declined `red_on_base_unverified` while
+  looking configured.
 * **Fail closed.** Every unknown returns `None`, which the guardrail reads as
   unverified and declines. An unverifiable claim is not a satisfied one.
 
@@ -77,9 +84,16 @@ def test_was_red_on_base(
     """`True` if *test_files* fail at *base_sha*, `False` if they pass, else `None`.
 
     `command_template` is the operator's own command with a `{tests}`
-    placeholder -- `pytest {tests}`, `mvn -q test -Dtest={tests}`, whatever the
-    repository actually uses. Absent, this returns `None` and the merge
-    declines: see the module docstring on why guessing is worse than the gap.
+    placeholder, run through `sh -c` so pipes, `||` and `$(...)` all work --
+    a Maven repository needs `$(...)` just to turn the changed file paths into
+    the class names `-Dtest=` accepts. Absent, this returns `None` and the
+    merge declines: see the module docstring on why guessing is worse than the
+    gap.
+
+    `sh -c` rather than `subprocess(shell=True)` so the command still arrives
+    through the injected runner as an argv, which is what makes this testable
+    without executing anything. It also means the template is POSIX shell; a
+    Windows runner would need its own spelling, and none is invented here.
 
     `None` means "could not establish", never "fine". A worktree that will not
     create, a runner that raises, no declared command and no test file in the
@@ -103,9 +117,10 @@ def test_was_red_on_base(
         return None
 
     try:
-        rendered = command_template.replace("{tests}", " ".join(files))
+        rendered = command_template.replace(
+            "{tests}", " ".join(shlex.quote(p) for p in files))
         try:
-            result = runner(shlex.split(rendered), cwd=str(root), check=False)
+            result = runner(["sh", "-c", rendered], cwd=str(root), check=False)
         except Exception:  # noqa: BLE001 - same reason as above
             return None
         code = getattr(result, "returncode", None)

{specfuse_loop-0.23.0 → specfuse_loop-0.23.2/specfuse_loop.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: specfuse-loop
-Version: 0.23.0
+Version: 0.23.2
 Summary: Local-first executor for the Specfuse Plan + Work Unit gate-cycle methodology.
 Author: Specfuse contributors
 License: Apache-2.0

{specfuse_loop-0.23.0 → specfuse_loop-0.23.2}/specfuse_loop.egg-info/SOURCES.txt

@@ -382,6 +382,7 @@ tests/test_judge_cost_fold_committed.py
 tests/test_judge_module.py
 tests/test_judge_path_registry.py
 tests/test_judge_terminal_flip_state.py
+tests/test_label_description_length.py
 tests/test_label_provisioning_runner_contract.py
 tests/test_label_registry.py
 tests/test_label_registry_covers_consumers.py

specfuse_loop-0.23.2/tests/test_label_description_length.py

@@ -0,0 +1,64 @@
+#
+# Copyright 2026 Specfuse contributors
+# Licensed under the Apache License, Version 2.0. See LICENSE.
+"""Every registered label description fits what GitHub will accept.
+
+`gh label create` rejects a description over **100 characters** with
+`HTTP 422: Validation Failed — description is too long (maximum is 100
+characters)`. `provision_labels` reports the failure and carries on, which is
+the right behaviour for one repository being briefly unreachable and the wrong
+outcome entirely for a spec that can never be created anywhere.
+
+Measured on a consumer repository at upgrade time:
+
+    label provisioning — created [], failed ['bug-lane:ci-pending',
+    'bug-lane:test-not-red-on-base', 'bug-lane:red-on-base-unverified']
+
+`bug-lane:ci-pending` is 101 characters and had been failing in every
+repository since it shipped. Nothing noticed, because the consequence only
+appears later and elsewhere: a declining bug-lane path calls `gh pr edit
+--add-label` against a label that was never created, which is #1420 exactly —
+the failure `tests/test_bug_lane_labels_registered.py` was written for, arriving
+through the one door that test does not watch.
+
+That test asserts every declining reason HAS a spec. This one asserts the spec
+is one GitHub will actually accept. A registry entry that cannot be created is
+indistinguishable, at the moment it matters, from no entry at all.
+"""
+
+from __future__ import annotations
+
+import unittest
+
+from specfuse.loop.labels import LABEL_REGISTRY
+
+#: GitHub's documented maximum for a label description.
+GITHUB_LABEL_DESCRIPTION_MAX = 100
+
+
+class LabelDescriptionsFitGitHubsLimit(unittest.TestCase):
+
+    def test_no_description_exceeds_the_limit(self):
+        too_long = [
+            f"{spec.name}: {len(spec.description)} chars"
+            for spec in LABEL_REGISTRY
+            if len(spec.description) > GITHUB_LABEL_DESCRIPTION_MAX
+        ]
+
+        self.assertEqual(
+            too_long, [],
+            f"`gh label create` rejects a description over "
+            f"{GITHUB_LABEL_DESCRIPTION_MAX} characters with HTTP 422, so "
+            f"these labels cannot be created in ANY repository — and a "
+            f"declining path that tries to apply one fails the way #1420 did. "
+            f"Over the limit:\n  " + "\n  ".join(too_long),
+        )
+
+    def test_the_registry_is_not_empty(self):
+        # Guards the assertion above against passing vacuously if the registry
+        # were ever emptied or failed to import.
+        self.assertGreater(len(LABEL_REGISTRY), 0)
+
+
+if __name__ == "__main__":  # pragma: no cover
+    unittest.main()

{specfuse_loop-0.23.0 → specfuse_loop-0.23.2}/tests/test_red_on_base_probe.py

@@ -178,5 +178,75 @@ class TheProbe(unittest.TestCase):
             self.assertNotIn("reset", argv)
 
 
+class TheCommandGetsAShell(unittest.TestCase):
+    """Operator commands use shell features; the probe must provide them.
+
+    The first cut of this module ran `shlex.split(rendered)` with no shell. It
+    passed every test above, because none of them used a shell feature — and it
+    could not express the command of the repository it was built for. A Maven
+    project needs `$(...)` just to turn changed file paths into the class names
+    `-Dtest=` accepts, so turning the guardrail on there would have declined
+    every merge as `red_on_base_unverified` while looking configured.
+    """
+
+    def _tmp(self) -> Path:
+        tmp = TemporaryDirectory()
+        self.addCleanup(tmp.cleanup)
+        return Path(tmp.name)
+
+    def _capture(self, template):
+        seen = {}
+
+        def runner(argv, **kw):
+            if argv[:2] == ["git", "worktree"]:
+                return SimpleNamespace(returncode=0, stdout="", stderr="")
+            seen["argv"] = list(argv)
+            return SimpleNamespace(returncode=1, stdout="", stderr="")
+
+        test_was_red_on_base(
+            runner, base_sha="abc123", test_files=["src/test/java/FooTest.java"],
+            command_template=template, worktree_root=self._tmp(),
+        )
+        return seen["argv"]
+
+    def test_the_command_is_handed_to_a_shell(self):
+        argv = self._capture("pytest {tests}")
+
+        self.assertEqual(["sh", "-c"], argv[:2])
+        self.assertEqual(3, len(argv), "the whole command is one shell string")
+
+    def test_shell_syntax_survives_intact(self):
+        template = (
+            './mvnw test -Dtest=$(echo "{tests}" | sed -E "s@.*/([^/]+)\\.java@\\1@")'
+            ' || { echo failed; exit 1; }'
+        )
+
+        argv = self._capture(template)
+
+        self.assertIn("$(", argv[2])
+        self.assertIn("||", argv[2])
+        self.assertIn("|", argv[2])
+
+    def test_a_path_needing_quoting_is_quoted(self):
+        seen = {}
+
+        def runner(argv, **kw):
+            if argv[:2] == ["git", "worktree"]:
+                return SimpleNamespace(returncode=0, stdout="", stderr="")
+            seen["argv"] = list(argv)
+            return SimpleNamespace(returncode=1, stdout="", stderr="")
+
+        test_was_red_on_base(
+            runner, base_sha="abc", test_files=["tests/a b.py", "tests/c;d.py"],
+            command_template="pytest {tests}", worktree_root=self._tmp(),
+        )
+
+        # Unquoted, `a b.py` would become two arguments and `c;d.py` would end
+        # the command — a path the repository controls must not be able to do
+        # either.
+        self.assertIn("'tests/a b.py'", seen["argv"][2])
+        self.assertIn("'tests/c;d.py'", seen["argv"][2])
+
+
 if __name__ == "__main__":  # pragma: no cover
     unittest.main()

specfuse_loop-0.23.0/specfuse/loop/data/VERSION

@@ -1 +0,0 @@
-0.23.0