Remove code no control flow can reach, and the check that proves it

CHANGELOG.md

@@ -8,6 +8,29 @@ as a CLI toolkit can. A **patch** release means fixes — it does not mean every
 flag and every default is frozen. Where a patch changes behaviour an existing
 user would notice, the release notes say so first.
 
+## [0.1.1] — 2026-09-16
+
+### Fixed
+
+- **Fifteen lines of unreachable code removed from `playwright_scraper.py`.**
+  A function's `def` line had been lost at some point before this repo's
+  first commit, leaving its docstring and its `try: return page.content()`
+  body indented into the end of `_mask_credentials`, where the control flow
+  can never arrive. Nothing called it — `_content_when_settled` below it
+  does the job — so no behaviour changes. The same fifteen lines, byte for
+  byte, were in six repos of this family.
+
+### Added
+
+- **A check for a statement the control flow can never reach.** The
+  undefined-name walk beside it cannot see this class by design: it pools
+  every binding in a file rather than tracking scopes, so a name used inside
+  dead code passes as long as anything else in the module binds it. The new
+  one is narrow — a statement after a `return`/`raise`/`break`/`continue` in
+  the SAME block — and measured across the eighteen repos of this family it
+  found six real problems and zero false positives. Verified by control:
+  appending `return 1` followed by a statement turns the suite red.
+
 ## [0.1.0] — 2026-09-14
 
 The first release of this repository as a member of the

playwright_scraper.py

@@ -603,21 +603,6 @@ def _mask_credentials(text: str) -> str:
     return _CREDENTIALS_IN_URL_RE.sub(r"\1***:***@", text or "")
 
 
-    """The page's HTML, or None when it cannot be read right now.
-
-    Playwright RAISES rather than returning empty while a navigation is in
-    flight ("Unable to retrieve content because the page is navigating"), and
-    a DataDome interstitial resolves by navigating — so the one moment this
-    is called is the one moment it can fail. Returning None keeps
-    `page_flow.settle_datadome` waiting instead of crashing the run, which is
-    what the first live run of this engine did.
-    """
-    try:
-        return page.content()
-    except (PWError, PWTimeout):
-        return None
-
-
 def _content_when_settled(page, attempts: int = 4, pause_ms: int = 700):
     """page.content() that tolerates a page mid-navigation.
 

pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "dubizzle-scraper"
-version = "0.1.0"
+version = "0.1.1"
 description = "dubizzle UAE classifieds scraper: cars, property, electronics, jobs and community listings to JSON or CSV."
 readme = "README.md"
 license = "MIT"

smoke_test.py

@@ -1801,6 +1801,14 @@ def test_no_undefined_names():
                            for k, v in sorted(missing.items()))
         ok &= check("%s references no undefined name%s"
                     % (name, ": " + detail if missing else ""), not missing)
+
+    # And a statement that can never RUN — see `_unreachable_statements`.
+    for name in sorted(f for f in os.listdir(REPO_ROOT) if f.endswith(".py")):
+        dead = _unreachable_statements(os.path.join(REPO_ROOT, name))
+        ok &= check("%s has no statement the control flow can never reach%s"
+                    % (name, "" if not dead else ": line %d" % dead[0]),
+                    not dead)
+
     return ok
 
 
@@ -2731,6 +2739,39 @@ def _undefined_names(path):
     return missing
 
 
+
+def _unreachable_statements(path):
+    """Line numbers of statements that can never run.
+
+    A statement sitting after a `return`/`raise`/`break`/`continue` in the
+    SAME block. Deliberately narrow: it makes no claim about conditions or
+    reachability in general, only about a block whose control flow has
+    already left. Measured across the eighteen repos in this family on
+    2026-09-16 it reported six problems and zero false positives.
+
+    `_undefined_names` above cannot see this class at all, by design — it
+    pools every binding in the file rather than tracking scopes, so a name
+    used inside dead code passes as long as anything else in the module
+    binds it. What was hiding there: a function whose `def` line had been
+    lost, leaving its docstring and body absorbed into the end of the
+    function above it. Identical in six repos, present since each one's
+    first commit, invisible to import, `--help`, `compileall` and every
+    green run of this suite.
+    """
+    tree = ast.parse(open(path, encoding="utf-8").read())
+    dead = []
+    for node in ast.walk(tree):
+        for field in ("body", "orelse", "finalbody"):
+            block = getattr(node, field, None)
+            if not isinstance(block, list):
+                continue
+            for i, stmt in enumerate(block[:-1]):
+                if isinstance(stmt, (ast.Return, ast.Raise,
+                                     ast.Continue, ast.Break)):
+                    dead.append(block[i + 1].lineno)
+                    break
+    return sorted(dead)
+
 def test_public_names_have_consumers():
     group("no public name is dead code or unenforced policy")
     ok = True