198906495f
* refactor(python-bridge): split non-streaming bridge modules * refactor(python-bridge): bring shared function tracing into route layer * feat(dev): list Python route functions and call sites * feat(dev): list Rust route functions and call sites * docs(dev): record OCR parity gaps across Python and Rust * feat(dev): list executed SDK calls with runtime tracing * feat(dev): report Python vs Rust SDK pipeline steps in one CLI * feat(dev): side-by-side pipeline step report in compare CLI * fix(dev): drop invalid Final annotations in compare cell loop * feat(dev): blue python-only and yellow rust-only steps in compare CLI * feat(dev): vertical layout with section spacing in compare CLI * fix(dev): validate SDK trace stages across sync and async routes * refactor(rust): align SDK route call structure with Python * refactor(python-bridge): share sync and async route call wrappers * refactor(dev): split compare CLI into fixtures, runtime, and report modules * fix(ci): run SDK trace tests and satisfy test lint
2.7 KiB
SDK route trace audit
Run the four native HTTP route families in both modes from the repository root:
uv run python -m tests.sdk_function_trace.compare --route all --both --check
The local fixture matrix on 2026-09-02 completed 15 successful engine invocations and one expected skip. Every successful invocation issued exactly one local HTTP request. All five comparable route/mode pairs have identical canonical steps in the same order, with no Python-only or Rust-only steps
| Route | Python async | Python sync | Rust async | Rust sync |
|---|---|---|---|---|
| Chat completions, Anthropic | Pass | Pass | Pass | Pass |
| Messages, Anthropic | Pass | Unsupported, skipped | Pass | Pass |
| OCR, Mistral | Pass | Pass | Pass | Pass |
| Audio transcription, Bedrock | Dispatch only | Dispatch only | Pass | Pass |
The same canonical step sequence ran in sync and async for each engine with both modes available. Bedrock transcription's Python SDK delegates to Rust, so its two successful calls do not establish independent provider parity. Realtime and Responses WebSockets are outside this HTTP fixture runner
Chat and OCR also have identical projected nesting in both modes. Async Messages has the same helper nesting beneath its handler, but Python starts that handler on a worker thread, so it appears as a second root. The comparison preserves this physical thread boundary and checks step order independently of absolute depth
Rust now resolves chat providers and supported parameters before entering its handler. Chat and Messages validate the environment and transform requests inside their handlers. Messages builds the final URL after transformation. OCR resolves its config and maps supported parameters during preparation, then validates credentials, builds the URL, and transforms the request inside its handler. Its during-call guardrails still run before HTTP, within the provider-call lifecycle phase
The environment hooks execute credential and header validation. Chat's supported-parameter hooks return OpenAI names paired with provider names and feed the existing request acceptance checks. The direct Rust API still accepts provider-mapped parameters, and its supported subset is smaller than Python's. Matching the pipeline does not establish identical parameter contracts
--check now fails if either comparable engine has missing, extra, or reordered canonical steps, even if its individual stage checks pass. Bedrock transcription and sync Messages report UNAVAILABLE for cross-language parity; native execution is still checked. Passing establishes step coverage and order for one non-streaming fixture per route, not complete request, response, error, or provider parity. The previously recorded OCR response gaps remain in ocr-comparison.md