Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.77%. Comparing base (868f28f) to head (530b7ff).
⚠️ Report is 1 commits behind head on master.

@@           Coverage Diff           @@
##           master    #1796   +/-   ##
=======================================
  Coverage   99.77%   99.77%           
=======================================
  Files          33       33           
  Lines        3540     3549    +9     
  Branches      498      500    +2     
=======================================
+ Hits         3532     3541    +9     
  Misses          5        5           
  Partials        3        3           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

Merging this PR will not alter performance

✅ 21 untouched benchmarks

_{Comparing fix/self-heal-dead-transport (530b7ff) with master (868f28f)¹}

PR Review — fix: prune dead transports from the engine on connection loss

Solid, well-scoped pruning mechanism; main open question is whether its trigger fires in the case it targets.

Strengths:

• Matching readers/senders by shared underlying transport identity (w.transport is not transport) is exactly right given each socket's transport is wrapped three times (reader, sender, and listener.transport).
• Rebuilding self.readers/self.senders via list comprehension (new list object) rather than in-place .remove() means any in-flight iterator over the old list — e.g. async_send's for send_transport in self.engine.senders — is unaffected.
• Loop-thread-only execution keeps it free-threading safe; no-op guards handle the never-registered / already-removed cases, and the noop test pins that.
• Coverage is complete and CodSpeed shows no regression.

Needs attention:

• warning Fix ServiceInfo __repr__ #1: the motivating scenario (interface down → EHOSTUNREACH on send) routes through asyncio's error_received, not connection_lost, so the dead transport may never reach this hook — verify the real trigger fires or prune from error_received.
• suggestion: trim the _async_remove_listener docstring to a single line per CLAUDE.md style.
• The added test exercises connection_lost directly, which validates the pruning logic but not that connection_lost is actually invoked on interface-down.

🟡 Important

except OSError as exc:
    self._protocol.error_received(exc)
    return
except Exception as exc:
    self._fatal_error(exc, 'Fatal write error on datagram transport')

def connection_lost(self, exc: Exception | None) -> None:
    """Prune this transport from the engine so a dead socket is not reused."""
    self.zc.engine._async_remove_listener(self)

🟢 Suggestions

Checklist

• Logic correct for connection-loss pruning
• Trigger fires in the targeted failure mode — warning ＃1
• Thread/free-threading safety
• No mutation-during-iteration hazard
• Cython .pxd updated where needed (_engine not cythonized; new listener methods not in .pxd)
• Tests cover new branches
• Docstring/comment style per CLAUDE.md — suggestion

To rebase specific severity levels, mention me: @bluetoothbot rebase critical (fixes 🔴 only), @bluetoothbot rebase important (fixes 🔴 + 🟡), or just @bluetoothbot rebase for all.

Automated review by Kōan (Claude) HEAD=37e3ceb 3 min 3s

Closing per maintainer review. As Kōan correctly flagged, EHOSTUNREACH/ENETUNREACH on send route to asyncio's error_received(), not connection_lost() (verified against CPython _SelectorDatagramTransport), so this hook never fires for the targeted scenario. These errnos are transient route conditions and should not trigger pruning of an otherwise-good interface. The runtime rescan API (#1797) and the optional monitor (#1798) cover genuine interface add/remove/IP-change, and are being reworked to stand alone on master without this change.