litellm/litellm-rust/crates/python-interop/AGENTS.md
yujonglee 347b642bdd
refactor(ocr): complete native lifecycle and preserve Azure auth (#40734)
* refactor(ocr): extract call completion boundary

* fix(ocr): release completion state after dispatch

* test(ocr): prove wrapper completion handoff

* test(ocr): narrow mapped failure assertion

* fix(ocr): preserve wrapper invocation kwargs

* fix(ocr): retain completion through finalization

* fix(ocr): make completion ownership explicit

* refactor(ocr): resolve logging executor explicitly

* fix(callbacks): preserve completion lifecycle behavior

* refactor(ocr): move public OCR into native lifecycle

* refactor(ocr): remove unused rust bridge capability

* wip

* wip

* refactor

* wip

* fix(ocr): preserve reducto native compatibility

* wip

* fix(ocr): document native callable casts

* perf(ocr): bound responses and reduce native scheduling overhead

* refactor(python-bridge): organize placeholder routes

* refactor test

* fix(ocr): normalize DeepSeek document content

* perf(ocr): skip unused callback work and benchmark callback overhead

* fix(ocr): align conversion contracts

* test(ocr): cover official provider response shapes

* fix(ocr): restore Python fallback and honor Rust opt-out

* fixes and refactor

* fix(ocr): preserve Azure Document Intelligence authentication

* fix(rust): enforce OCR response limits and lint contracts

* test(rust): align native OCR contract coverage

* test(ocr): isolate Azure auth precedence coverage
2026-09-12 11:56:49 -07:00

2.1 KiB

  • Target invariants; implementation and runtime validation may lag these rules
  • Keep this crate a small, domain-neutral foundation: Python/Serde conversion and interpreter-boundary utilities
    • No LiteLLM domain dependencies, route types, callback policy, public API registration or cdylib build features
    • Generic code alone does not justify extraction: runtime integration stays in python-bridge/src/execution.rs, host adaptation in its lifecycle.rs
  • Use standard PyO3 ownership and conversion APIs
    • Prefer Bound<'py, T> for attached operations/results, Py<T> for retention; binding/unbinding does not copy payloads
    • Use pythonize for selected Serde data, never a JSON-text round trip; share conversion with Pythonized<T>
    • Preserve PythonizeError's standard conversion into PyErr; do not stringify original Python exceptions into new ValueErrors
    • Keep serializer-panic containment in Pythonized<T>: async output conversion can run in an unjoined blocking task and otherwise strand delivery
  • Use Python::detach for Rust-only work; Python operations require attachment
    • Keep diagnostic counters in the consumer; wrapper invocations do not measure every interpreter release
    • Release exclusive class borrows/locks before Python calls or decrements that can invoke finalizers; expose retained Python edges to GC without calling Python during traversal
  • Keep coroutine driving in the shared Python driver and native adapter
    • Driver: litellm/rust_bridge/lifecycle.py; handle: python-bridge/src/lifecycle.rs; native-backed behavior tests: python-bridge/tests/lifecycle.py
  • References: ownership, conversions, pythonize errors