code-coordinator 0.5.223__tar.gz → 0.5.224__tar.gz

{code_coordinator-0.5.223 → code_coordinator-0.5.224}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-coordinator
-Version: 0.5.223
+Version: 0.5.224
 Summary: Coordinate Claude Code workers across multiple machines and repos.
 Author-email: John Donaghy <johnfdonaghy@gmail.com>
 License-Expression: LicenseRef-FSL-1.1-MIT

{code_coordinator-0.5.223 → code_coordinator-0.5.224}/code_coordinator.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-coordinator
-Version: 0.5.223
+Version: 0.5.224
 Summary: Coordinate Claude Code workers across multiple machines and repos.
 Author-email: John Donaghy <johnfdonaghy@gmail.com>
 License-Expression: LicenseRef-FSL-1.1-MIT

{code_coordinator-0.5.223 → code_coordinator-0.5.224}/code_coordinator.egg-info/scm_version.json

@@ -1,7 +1,7 @@
 {
-  "tag": "0.5.223",
+  "tag": "0.5.224",
   "distance": 0,
-  "node": "g744223e42f4d266aef3954a9aaed09de74827289",
+  "node": "gcb4f4b8b669bb9319880ac1b1b10496b1131b557",
   "dirty": false,
   "branch": "HEAD",
   "node_date": "2026-08-22"

{code_coordinator-0.5.223 → code_coordinator-0.5.224}/scripts/coord-test-runner.sh

@@ -829,6 +829,32 @@ run_python() {
         return $?
     fi
 
+    # #2562: the isolation re-run above is not all-or-nothing evidence — its
+    # own `FAILED ` lines say EXACTLY which node ids survived. Rebuild
+    # $failed from those before doing anything else: a test that dropped out
+    # PASSED in isolation and is by definition the flake case this whole
+    # block exists to catch. Handing the ORIGINAL (unpruned) set to the
+    # baseline comparison below is the bug this closes — one genuine failure
+    # was keeping a flake in the set, and the baseline comparison is itself
+    # all-or-nothing (#2170's unanimity), so the flake alone could suppress a
+    # correct BASELINE-RED downgrade.
+    local still_failed
+    still_failed="$(grep '^FAILED ' "$rerun" | awk '{print $2}' | sort -u || true)"
+    if [[ -z "$still_failed" ]]; then
+        # Non-zero exit but no parseable FAILED lines — e.g. a collection/
+        # import error on the isolated re-run. Nothing to prune against;
+        # fall back to the original set rather than claim a pruning that
+        # did not happen.
+        still_failed="$failed"
+    fi
+    local pruned_count; pruned_count="$(printf '%s\n' "$still_failed" | grep -c . || true)"
+    if [[ "$pruned_count" -lt "$count" ]]; then
+        log "$((count - pruned_count)) of $count failing test(s) passed in isolation — treating as flakes, comparing the remaining $pruned_count against the baseline"
+        FLAKES+=("python:$((count - pruned_count))")
+    fi
+    failed="$still_failed"
+    count="$pruned_count"
+
     # Not a flake. Before calling it the branch's fault, ask the question the
     # Test gate is actually asked: is this worse than $BASE_REF on THIS machine?
     # (#2170 — see the baseline section above for why this only ever downgrades

{code_coordinator-0.5.223 → code_coordinator-0.5.224}/tests/test_coord_test_runner_baseline.py

@@ -97,6 +97,15 @@ if [[ "$1" == "-m" && "$2" == "pytest" ]]; then
             printf '%s passed\n' "2"
             exit 0
         fi
+        # #2562: a PARTIAL isolation re-run — one node id genuinely still
+        # fails, the rest passed. Distinct from FAKE_RERUN_PASSES (all pass,
+        # exit 0) and from the "no override" default (everything in
+        # FAKE_BRANCH_FAILED fails again, as if isolation changed nothing).
+        if [[ "$_targeted" -eq 1 && -n "${FAKE_RERUN_STILL_FAILING:-}" ]]; then
+            _emit_failed "$FAKE_RERUN_STILL_FAILING"
+            printf '1 failed, 1 passed\n'
+            exit 1
+        fi
         _emit_failed "${FAKE_BRANCH_FAILED:-}"
         printf '2 failed\n'
         exit 1
@@ -393,6 +402,74 @@ def test_a_flake_is_still_a_flake_and_never_consults_the_baseline(repo: Path) ->
     assert "baseline:" not in result.stdout + result.stderr
 
 
+# ── #2562: partial-flake pruning feeds baseline comparison ──────────────────
+#
+# The all-or-nothing bug: `pytest $failed` re-ran the whole failing set
+# together, so ONE genuine failure alongside a flake kept the flake in
+# `$failed` — and that poisoned set then went to `python_baseline_is_red`,
+# itself unanimity-gated, where the flake (which passes on the baseline same
+# as everywhere else) short-circuited a correct BASELINE-RED downgrade. The
+# fix rebuilds `$failed` from the isolation re-run's own `FAILED ` lines
+# before doing anything else with it.
+
+
+def test_mixed_flake_and_genuine_prunes_to_baseline_red(repo: Path) -> None:
+    """{genuine, flake} prunes to {genuine}, and BASELINE-RED fires on just
+    that remainder — the exact CC#2556 shape from the issue: a real flake
+    sitting next to a test that fails identically on the merge-base must not
+    suppress the downgrade.
+    """
+    result = _run(
+        repo,
+        # Isolation re-run: test_one still fails, test_two (the flake) passed
+        # and dropped out.
+        FAKE_RERUN_STILL_FAILING="tests/test_ambient.py::test_one",
+        # Baseline reproduces the one that's left.
+        FAKE_BASELINE_FAILED="tests/test_ambient.py::test_one",
+    )
+
+    assert result.returncode == 4, (result.returncode, result.stdout, result.stderr)
+    assert "RESULT: BASELINE-RED (python)" in result.stdout
+    assert "1 of 2 failing test(s) passed in isolation" in result.stdout
+    assert "comparing the remaining 1 against the baseline" in result.stdout
+    assert "BASELINE-RED(python): all 1 failing test(s) fail identically" in result.stdout
+    # The pruned flake must not appear as part of the (now genuine-only) set.
+    assert "tests/test_ambient.py::test_one" in result.stdout
+    assert "RESULT: FAIL" not in result.stdout
+
+
+def test_mixed_flake_and_genuine_prunes_to_a_smaller_genuine_fail(repo: Path) -> None:
+    """Same pruning, but the remainder is NOT baseline-red: still a FAIL, but
+    for the pruned count and node id only — the reported list must match
+    what actually failed on the isolation re-run, not the original set.
+    """
+    result = _run(
+        repo,
+        FAKE_RERUN_STILL_FAILING="tests/test_ambient.py::test_one",
+        # FAKE_BASELINE_FAILED unset ⇒ baseline is green for test_one too.
+    )
+
+    assert result.returncode == 1
+    assert "RESULT: FAIL (python)" in result.stdout
+    assert "BASELINE-RED" not in result.stdout
+    assert "1 of 2 failing test(s) passed in isolation" in result.stdout
+    assert "FAIL(python): 1 test(s) fail on re-run — genuine" in result.stdout
+    assert "tests/test_ambient.py::test_one" in result.stdout
+    assert "tests/test_ambient.py::test_two" not in (
+        result.stdout.split("FAIL(python): 1 test(s) fail on re-run — genuine")[1]
+    )
+
+
+def test_all_genuine_failing_set_is_unaffected_by_pruning(repo: Path) -> None:
+    """Both failures still fail in isolation ⇒ nothing pruned, and behaviour
+    matches the pre-#2562 baseline-red path exactly (no pruning log line)."""
+    result = _run(repo, FAKE_BASELINE_FAILED=BRANCH_FAILED)
+
+    assert result.returncode == 4
+    assert "RESULT: BASELINE-RED (python)" in result.stdout
+    assert "passed in isolation" not in result.stdout
+
+
 # ── #2180 review fix: the Test stage must agree with CI on tests/acceptance ─
 #
 # Before this fix, `--ignore=tests/acceptance` dropped the sealed cli-pytest