[chivector]

Describe your changes

Fix S3 output with both default and explicit client options. On current main, to_json(filepath="s3://...") fails because the writer requires s3_options; supplying that keyword instead sends it to the JSON encoder and raises another TypeError.

Remove the transport-only keyword before encoding and normalize omitted/None options to an empty mapping at S3 dispatch, matching the input path. Serializer options such as indent and sort_keys still reach the serializer, and the caller's options are preserved. The change adds no dependency or public method.

Related issue

Fixes #649.

Validation

• Final new regressions on unchanged Git base e910141: 13 errors across two test methods; fixed source: all 13 subcases pass.
• Standard full coverage/unittest run on Windows / CPython 3.12.14: 992 tests passed, aggregate coverage 99%. After the submit hook added a postponed-annotations import to the new test only, the final 13 subcases were rerun successfully; production code is unchanged.
• Independent real-boto3 localhost HTTP checks: original source six TypeErrors and zero uploads; fixed source six successful PUTs, verifying exact bytes, default/None/empty/custom configuration, JSON/YAML/TOML output, and existing success-path cleanup. The public API was also rerun from the final Git checkout.
• All configured pinned pre-commit hooks pass, including Ruff 0.17.0 and mypy 2.4.0 (66 files in the hook). Separate repository-wide Ruff lint/format and mypy on all 65 package source files also pass.
• git diff --check and independent agent review pass.

No live AWS operations or full Python/OS matrix were run. Existing credential-gated S3 tests did not exercise AWS; the localhost upload checks use an isolated environment and disposable test values. The initial full test attempt needed the missing orjson test dependency; the completed 992-test run includes it.

Checklist before requesting a review

• I have performed a self-review of my code.
• I have added tests for the proposed changes.
• I have run the tests and there are no errors.

AI assistance: OpenAI Codex investigated, implemented and tested this user-authorized contribution, including an independent agent review. No separate human review or personal production incident is claimed.