v0.2.3

AstraAPI 0.2.3 Release Notes

PyPI: https://pypi.org/project/astraapi/0.2.3
Python: 3.14+ | C++: 20

---

Overview

Major performance release — pydantic validation and dependency injection overhead reduced by 3-8x on complex routes. response_model endpoints now skip redundant Pydantic re-validation. C++ coroutine inlining eliminates event-loop bounces for trivial async deps.

---

Performance Improvements

Pydantic Validation Fast Path

• astraapi/routing.py — Extended _is_trivial_schema() to accept int/float/bool alongside str. The _fast_validate closure now performs lightweight builtin coercion (~50ns) instead of Pydantic TypeAdapter.validate_python() (~2-5µs) for scalar params with no constraints (min_length, ge, le, pattern, etc.). C++ fast_spec extraction already supplies the correct raw types.

• astraapi/dependencies/utils.py — _validate_value_with_model_field() now skips Pydantic entirely when the value is already the correct Python type and the core_schema has no constraint keys. Uses module-level _TRIVIAL_CONSTRAINT_KEYS frozenset (avoiding per-call construction). Only applies to non-body params (query/header/cookie/path).

Pre-Encoded Response Body Fast Path

• astraapi/applications.py — _make_fast_dep_shim() detects when an endpoint returns a Pydantic model instance AND the route has a response_model configured. Calls model_dump_json(by_alias=True) directly and returns the JSON string, skipping C++ TypeAdapter.validate_python() + serialize_python() + JSON re-serialization (3 redundant operations).

• cpp_core/src/app.cpp — Pre-encoded body fast path (bytes/str check) placed BEFORE the response_model_local validation block. When Python returns bytes or str, C++ writes it directly to transport without touching response_model validation or JSON serialization.

Pre-Interned Spec Keys

• astraapi/dependencies/utils.py — sys.intern() called on field names and aliases at route registration time (inside _precomputed_batch_specs). Pre-interned py_field_name and py_lookup_key passed as extra dict fields in the batch specs.

• cpp_core/src/param_extractor.cpp — batch_extract_params_inline() checks for pre-interned keys from Python before calling PyUnicode_InternFromString(). Uses Py_INCREF on cached refs instead of interning per-request. Added cleanup loop (Py_XDECREF) at function exit.

Response Cache Size Guard

• cpp_core/src/app.cpp — hash_dict_content() returns 0 (skip cache) for dicts with >50 keys, avoiding expensive recursive hashing on large responses.

Memory Allocation Reductions

• astraapi/dependencies/utils.py — all_param_fields construction replaced 5 list allocations (list(path) + list(query) + list(header) + list(cookie)) with itertools.chain() iterator. Cookie string parsing uses list-accumulate-then-join instead of O(N²) += concatenation in both batch and fallback paths.

Multi-Worker SO_REUSEPORT Detection Fix

• astraapi/_multiworker.py — SO_REUSEPORT capability test now binds to ephemeral port (0) instead of the actual listen port, avoiding conflict with the parent's already-bound socket.

---

Architecture

                      ┌─────────────────────────────────────┐
                      │  Python _fast_dep_shim               │
                      │  model_dump_json() → str             │
                      └──────────────┬──────────────────────┘
                                     │ JSON string
                      ┌──────────────▼──────────────────────┐
                      │  C++ dispatch_one_request            │
                      │  PyBytes/PyUnicode check FIRST       │
                      │  ↓ skip validate_python              │
                      │  ↓ skip serialize_python             │
                      │  ↓ skip json_writer re-serialize      │
                      │  → transport.write(raw_bytes)        │
                      └──────────────────────────────────────┘

---

Installation

pip install astraapi