[dmccoystephenson]

Summary

• A Stage B (unit-test expansion) cycle: a characterization test is added for the per-request bound in _send, which no existing test asserted.
• test_a_server_that_never_answers_costs_one_report_not_the_sender shortens TraceClient.TIMEOUT_SECONDS to 0.5 s, holds the loopback server's first request unanswered, and checks three things: the hung request is logged as could not deliver, the next report is still delivered, and every log record is at DEBUG.
• Before this, a regression that dropped or raised the urlopen timeout would have passed the suite. The existing hang test only bounds close(), not the sender.
• Only tests change. No production code or docs are touched.

Test plan

• python3 -m unittest -v locally (Python 3.8.10): Ran 68 tests, OK
• Mutation check: with urlopen(..., timeout=30) in place of self.TIMEOUT_SECONDS, the new test failed (FAILED (failures=1)); with the source restored, it passed
• CI test (3.8), test (3.10), test (3.12) green

Deferred

• The version tag is added even when it pushes an event past MAX_TAGS #16 is still deferred. It needs an owner decision on whether the README promise that tags "always holds at least version" holds at the MAX_TAGS cap, and the change has to land together with trace-client-java.

Java parity

No Java-visible behavior changed. This adds a test only.

No tracking issue: the gap was found during triage.

This PR description was drafted during a Gardener session (https://github.com/Stephenson-Software/gardener).

🤖 Generated with Claude Code

---

drafted by Claude on behalf of Daniel Stephenson

[dmccoystephenson]

Self-review rubric (one pass, anchored on CI run 41180910676):

• Scope: PASS. gh pr diff 29 --name-only lists only tests/test_trace_client.py (+20 lines, one new test).
• Tests-new: PASS (not applicable). No new public names were added.
• Tests-fix: PASS (not applicable as a bug fix). This is a characterization test. A mutation check was still run: with the urlopen timeout replaced by 30, the new test failed (FAILED (failures=1)), and with the source restored it passed.
• Sibling structure: PASS. The test sits in TraceClientTest, uses _Capture.release to simulate a hanging server, uses mock.patch.object and self.log, closes its client, and has explanatory assertion messages.
• Sibling renames: PASS (not applicable). Nothing was renamed.
• Docs: PASS. No behavior changed. The README "Bounded" and "never raises" rows already describe what the test asserts.
• Issue resolution: PASS (not applicable). No Closes. The body records the triage-found gap and the deferral of The version tag is added even when it pushes an event past MAX_TAGS #16.
• CI: PASS. test (3.8), test (3.10) and test (3.12) are all green on the PR head.
• Stdlib-only: PASS. No added import lines.
• Py3.8 floor: PASS. test (3.8) is green, and the local run was also on 3.8.10.
• Never-raises / never-blocks: PASS (not applicable). No production code changed.
• Log level: PASS. No _LOG calls were added. The test itself asserts that every record is DEBUG.
• Version agreement / Exports: PASS (not applicable). Neither was touched.
• Java parity noted: PASS. The body states that no Java-visible behavior changed.

Judgment call for a reader: the 10 s polling deadline is twice the default timeout and twenty times the patched one, so it should be generous enough for a slow runner. Patching the class attribute does not change close()'s default timeout, because that default was bound when the function was defined. close() therefore still allows the full 5 s for "second" to drain.

This comment was drafted during a Gardener session (https://github.com/Stephenson-Software/gardener).

---

drafted by Claude on behalf of Daniel Stephenson