Keep report() from raising on a non-string name and the sender alive on a scheme-less URL

[dmccoystephenson]

Summary

• report(): the blank-name check was moved inside the existing try. A non-string name such as report(123) used to raise AttributeError into the host program. Now it is dropped and logged at DEBUG (could not queue 123: ...), which is what the README's "Never raises" promise says happens.
• _send(): the urllib.request.Request is now built inside the existing try. A base_url with no scheme ("trace.example.org") used to make Request.__init__ raise ValueError on the sender thread, which killed the thread and printed a traceback to the host's stderr. Now each report is logged at DEBUG as could not deliver and the thread keeps running.
• The constructor is unchanged. It does not start rejecting URLs it used to accept, so no host program that constructs a client today can start failing at construction.

Closes #2
Closes #3

Test plan

• python3 -m unittest -v on Python 3.8.10: 23 tests ran, OK
• Regression check: with the trace_client.py change stashed, test_report_does_not_raise_for_a_name_that_is_not_a_string errors (AttributeError) and test_a_base_url_without_a_scheme_is_logged_not_a_dead_thread fails (threading.excepthook received the ValueError). Both pass once the change is restored.
• CI Build: test (3.8), test (3.10), test (3.12)

Java parity

No Java-visible behavior changed: the wire format, the opt-out values and the disabled_reason strings are untouched. Both fixes cover inputs specific to Python (untyped name, and URL parsing that happens lazily on the sender thread). Nothing has to be done in trace-client-java.

Notes

• Only ubuntu-latest is covered by CI. Nothing in this diff depends on the platform or on Pyodide.
• No other open issues existed at triage time, so nothing was deferred.

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 (checked against CI run 35967784767 on the PR head):

• Scope: PASS. Only trace_client/trace_client.py (the two fixes) and tests/test_trace_client.py (two regression tests) were modified: +34/−8.
• Tests-new: PASS. No new public names were added.
• Tests-fix: PASS. Confirmed by stash-and-run on Python 3.8.10. With the source change stashed, test_report_does_not_raise_for_a_name_that_is_not_a_string errored (AttributeError) and test_a_base_url_without_a_scheme_is_logged_not_a_dead_thread failed (threading.excepthook received ValueError: unknown url type). With the change restored, all 23 tests pass.
• Sibling structure / renames: PASS. No new files and no renames.
• Docs: PASS. The README's "Never raises" row and the class docstring already describe the behavior this PR delivers. No wording needed to change.
• Issue resolution: PASS. report() raises AttributeError for a non-string name, breaking the never-raises promise #2 (report() name check) and A base_url without a scheme kills the sender thread and prints a traceback to stderr #3 (Request built outside the try) are both changed at the named lines.
• CI: PASS. test (3.8), test (3.10) and test (3.12) are green.
• Stdlib-only: PASS. The diff adds no import lines, and pyproject.toml is untouched.
• Py3.8 floor: PASS. test (3.8) is green. threading.excepthook, which the new test patches, has existed since 3.8.
• Never-raises / never-blocks: PASS. No new blocking call was added. Both moved statements now sit inside existing try blocks that log at DEBUG.
• Log level: PASS. The diff adds no _LOG. calls, and the existing [trace] DEBUG messages are reused.
• Version agreement: PASS. The version is untouched, and 0.2.0 still appears consistently.
• Exports: PASS. No public names changed.
• Java parity noted: PASS. The PR body states that no Java-visible behavior changed.

Observations outside the diff:

• trace_client/trace_client.py:176 (_drain) still has no catch-all around self._send(body). After this PR, every known raising statement in _send is inside its try, so nothing currently reaches it. An outer guard would be defense in depth for future edits to _send. It is a candidate for a separate issue and was left out here to keep this PR scoped.
• Do-not-auto-merge list: no protected path is touched (no pyproject.toml, no version string, no wire-format or opt-out change, no workflow). The merge still needs a human, because this dispatch is not authorized to merge.

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

---

drafted by Claude on behalf of Daniel Stephenson