Merge pull request 'feat(config): AppConfig + cached get_config()' (#19) from feat/config into main

Typed pydantic-settings view over every IX_* env var, defaults matching
spec §9 exactly. @lru_cache-wrapped accessor so parsing/validation happens
once per process; tests clear the cache via get_config.cache_clear().

extra="ignore" keeps the container robust against typo'd env vars in
production .env files. engine.py's URL resolver now goes through
get_config() when ix.config is importable (bootstrap fallback remains so
hypothetical early-import callers don't crash).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

src/ix/config.py

@@ -0,0 +1,75 @@
+"""Application configuration — loaded from ``IX_*`` env vars via pydantic-settings.
+
+Spec §9 lists every tunable. This module is the single read-point for them;
+callers that need runtime config should go through :func:`get_config` rather
+than ``os.environ``. The LRU cache makes the first call materialise + validate
+the full config and every subsequent call return the same instance.
+
+Cache-clearing is public (``get_config.cache_clear()``) because tests need to
+re-read after ``monkeypatch.setenv``. Production code never clears the cache.
+"""
+
+from __future__ import annotations
+
+from functools import lru_cache
+
+from pydantic_settings import BaseSettings, SettingsConfigDict
+
+
+class AppConfig(BaseSettings):
+    """Typed view over the ``IX_*`` environment.
+
+    Field names drop the ``IX_`` prefix — pydantic-settings puts it back via
+    ``env_prefix``. Defaults match the spec exactly; do not change a default
+    here without updating spec §9 in the same commit.
+    """
+
+    model_config = SettingsConfigDict(
+        env_prefix="IX_",
+        env_file=".env",
+        env_file_encoding="utf-8",
+        extra="ignore",
+    )
+
+    # --- Job store ---
+    postgres_url: str = (
+        "postgresql+asyncpg://infoxtractor:<password>"
+        "@host.docker.internal:5431/infoxtractor"
+    )
+
+    # --- LLM backend ---
+    ollama_url: str = "http://host.docker.internal:11434"
+    default_model: str = "gpt-oss:20b"
+
+    # --- OCR ---
+    ocr_engine: str = "surya"
+
+    # --- Pipeline behavior ---
+    pipeline_worker_concurrency: int = 1
+    pipeline_request_timeout_seconds: int = 2700
+    genai_call_timeout_seconds: int = 1500
+    render_max_pixels_per_page: int = 75_000_000
+
+    # --- File fetching ---
+    tmp_dir: str = "/tmp/ix"
+    file_max_bytes: int = 52_428_800
+    file_connect_timeout_seconds: int = 10
+    file_read_timeout_seconds: int = 30
+
+    # --- Transport / callbacks ---
+    callback_timeout_seconds: int = 10
+
+    # --- Observability ---
+    log_level: str = "INFO"
+
+
+@lru_cache(maxsize=1)
+def get_config() -> AppConfig:
+    """Return the process-wide :class:`AppConfig` (materialise on first call).
+
+    Wrapped in ``lru_cache`` so config is parsed + validated once per process.
+    Tests call ``get_config.cache_clear()`` between scenarios; nothing in
+    production should touch the cache.
+    """
+
+    return AppConfig()

tests/unit/test_config.py

@@ -0,0 +1,131 @@
+"""Tests for :mod:`ix.config` — the pydantic-settings ``AppConfig``.
+
+Guardrails we care about:
+
+1. Every env var in spec §9 round-trips with the right type.
+2. Defaults match the spec exactly when no env is set.
+3. Unknown IX_ vars are ignored (``extra="ignore"``) so a typo doesn't crash
+   the container at startup.
+4. ``get_config()`` is cached — same instance per process — and
+   ``get_config.cache_clear()`` rebuilds from the current environment (used by
+   every test here to keep them independent of process state).
+"""
+
+from __future__ import annotations
+
+import pytest
+
+from ix.config import AppConfig, get_config
+
+
+@pytest.fixture(autouse=True)
+def _reset_config_cache() -> None:
+    """Flush the LRU cache around every test.
+
+    Without this, tests that set env vars would see stale data from earlier
+    runs because ``get_config()`` caches the first materialised instance.
+    """
+
+    get_config.cache_clear()
+
+
+def _clear_ix_env(monkeypatch: pytest.MonkeyPatch) -> None:
+    """Scrub every IX_* var so defaults surface predictably.
+
+    Tests that exercise env-based overrides still call ``monkeypatch.setenv``
+    after this to dial in specific values; tests for defaults rely on this
+    scrubbing so a developer's local ``.env`` can't contaminate the assertion.
+    """
+
+    import os
+
+    for key in list(os.environ):
+        if key.startswith("IX_"):
+            monkeypatch.delenv(key, raising=False)
+
+
+def test_defaults_match_spec(monkeypatch: pytest.MonkeyPatch) -> None:
+    _clear_ix_env(monkeypatch)
+    # Don't let pydantic-settings pick up the repo's .env.example.
+    cfg = AppConfig(_env_file=None)  # type: ignore[call-arg]
+
+    assert cfg.postgres_url == (
+        "postgresql+asyncpg://infoxtractor:<password>"
+        "@host.docker.internal:5431/infoxtractor"
+    )
+    assert cfg.ollama_url == "http://host.docker.internal:11434"
+    assert cfg.default_model == "gpt-oss:20b"
+    assert cfg.ocr_engine == "surya"
+    assert cfg.tmp_dir == "/tmp/ix"
+    assert cfg.pipeline_worker_concurrency == 1
+    assert cfg.pipeline_request_timeout_seconds == 2700
+    assert cfg.genai_call_timeout_seconds == 1500
+    assert cfg.file_max_bytes == 52428800
+    assert cfg.file_connect_timeout_seconds == 10
+    assert cfg.file_read_timeout_seconds == 30
+    assert cfg.render_max_pixels_per_page == 75000000
+    assert cfg.log_level == "INFO"
+    assert cfg.callback_timeout_seconds == 10
+
+
+def test_env_overrides(monkeypatch: pytest.MonkeyPatch) -> None:
+    _clear_ix_env(monkeypatch)
+    monkeypatch.setenv("IX_POSTGRES_URL", "postgresql+asyncpg://u:p@db:5432/x")
+    monkeypatch.setenv("IX_OLLAMA_URL", "http://llm:11434")
+    monkeypatch.setenv("IX_DEFAULT_MODEL", "llama3:8b")
+    monkeypatch.setenv("IX_PIPELINE_WORKER_CONCURRENCY", "4")
+    monkeypatch.setenv("IX_GENAI_CALL_TIMEOUT_SECONDS", "60")
+    monkeypatch.setenv("IX_LOG_LEVEL", "DEBUG")
+    monkeypatch.setenv("IX_CALLBACK_TIMEOUT_SECONDS", "30")
+
+    cfg = AppConfig(_env_file=None)  # type: ignore[call-arg]
+
+    assert cfg.postgres_url == "postgresql+asyncpg://u:p@db:5432/x"
+    assert cfg.ollama_url == "http://llm:11434"
+    assert cfg.default_model == "llama3:8b"
+    assert cfg.pipeline_worker_concurrency == 4
+    assert cfg.genai_call_timeout_seconds == 60
+    assert cfg.log_level == "DEBUG"
+    assert cfg.callback_timeout_seconds == 30
+
+
+def test_get_config_is_cached(monkeypatch: pytest.MonkeyPatch) -> None:
+    _clear_ix_env(monkeypatch)
+    monkeypatch.setenv("IX_POSTGRES_URL", "postgresql+asyncpg://a:b@c:5432/d1")
+    first = get_config()
+    # Later mutation must NOT be seen until cache_clear — this is a feature,
+    # not a bug: config is process-level state, not per-call.
+    monkeypatch.setenv("IX_POSTGRES_URL", "postgresql+asyncpg://a:b@c:5432/d2")
+    second = get_config()
+    assert first is second
+    assert second.postgres_url.endswith("/d1")
+
+    get_config.cache_clear()
+    third = get_config()
+    assert third is not first
+    assert third.postgres_url.endswith("/d2")
+
+
+def test_extra_env_keys_are_ignored(monkeypatch: pytest.MonkeyPatch) -> None:
+    """A typo'd IX_FOOBAR should not raise ValidationError at startup."""
+    _clear_ix_env(monkeypatch)
+    monkeypatch.setenv("IX_FOOBAR", "whatever")
+    # Should not raise.
+    cfg = AppConfig(_env_file=None)  # type: ignore[call-arg]
+    assert cfg.ollama_url.startswith("http://")
+
+
+def test_engine_uses_config_url(monkeypatch: pytest.MonkeyPatch) -> None:
+    """``ix.store.engine`` reads the URL through ``AppConfig``.
+
+    Task 3.2 refactors engine.py to go through ``get_config()`` instead of
+    reading ``os.environ`` directly. We can't actually construct an async
+    engine in a unit test (would need the DB), so we verify the resolution
+    function exists and returns the configured URL.
+    """
+    _clear_ix_env(monkeypatch)
+    monkeypatch.setenv("IX_POSTGRES_URL", "postgresql+asyncpg://a:b@c:5432/d")
+
+    from ix.store.engine import _resolve_url
+
+    assert _resolve_url() == "postgresql+asyncpg://a:b@c:5432/d"