.gitea/workflows/build-images.yml

@@ -5,7 +5,9 @@ name: Build Images
 # devx and all dependencies into the image.
 #
 # Triggers:
-# - On push to master (after post-merge release completes)
+# - After post-merge workflow completes successfully (workflow_run)
+#   This ensures images are only rebuilt AFTER the release is published
+#   to PyPI, so the image always has the latest released version.
 # - Manually via workflow_dispatch
 #
 # The workflow builds 3 tier images in sequence:
@@ -15,12 +17,10 @@ name: Build Images
 # After pushing, a cleanup job removes old versions (keeps last 2 + latest).
 
 on:
-  push:
+  workflow_run:
+    workflows: ["Post-merge"]
+    types: [completed]
     branches: [master]
-    paths:
-      - docker/**
-      - pyproject.toml
-      - src/devx/**
   workflow_dispatch:
 
 concurrency:
@@ -49,7 +49,11 @@ jobs:
 
   build-and-push:
     needs: [detect-type]
-    if: needs.detect-type.outputs.is-release == 'false'
+    if: >-
+      needs.detect-type.outputs.is-release == 'false' && (
+        github.event_name == 'workflow_dispatch' ||
+        (github.event_name == 'workflow_run' && github.event.workflow_run.conclusion == 'success')
+      )
     runs-on: docker
     timeout-minutes: 30
     steps:

src/devx/tools/check_test_speed.py

@@ -32,7 +32,10 @@ _TIMING_RE = re.compile(r"(\d+) passed.* in ([0-9.]+)s")
 
 # Matches per-test duration lines from --durations=0:
 #   0.51s call     tests/test_foo.py::test_bar
-_DURATION_LINE_RE = re.compile(r"^(\d+\.?\d*)s\s+(?:setup|call|teardown)\s+(.+)$")
+# Only "call" duration is counted — "setup" includes import/collection
+# overhead (coverage init, module imports) which is environment-dependent
+# and not a test quality signal.
+_DURATION_LINE_RE = re.compile(r"^(\d+\.?\d*)s\s+call\s+(.+)$")
 
 
 def run_tests() -> tuple[str, str]:

tests/unit/test_check_test_speed.py

@@ -69,16 +69,16 @@ class TestParsePerTestDurations:
         assert len(durations) == 1
         assert durations[0] == ("tests/test_foo.py::test_bar", 0.01)
 
-    def test_parses_setup_and_teardown(self) -> None:
+    def test_ignores_setup_and_teardown(self) -> None:
+        """Only 'call' durations are counted — setup includes import overhead."""
         output = (
-            "0.02s setup    tests/test_foo.py::test_bar\n"
+            "0.68s setup    tests/test_foo.py::test_bar\n"
             "0.01s call     tests/test_foo.py::test_bar\n"
             "0.00s teardown tests/test_foo.py::test_bar\n"
         )
         durations = parse_per_test_durations(output)
-        assert len(durations) == 3
-        names = [d[0] for d in durations]
-        assert "tests/test_foo.py::test_bar" in names
+        assert len(durations) == 1
+        assert durations[0] == ("tests/test_foo.py::test_bar", 0.01)
 
     def test_sorted_slowest_first(self) -> None:
         output = "0.01s call     tests/test_a.py::test_slow\n0.50s call     tests/test_b.py::test_fast\n"