Merge pull request #41 from modern-python/feat/multi-decoder-routing

feat!: multi-decoder routing (decoders=[...])

Author: lesnik512

README.md

@@ -15,12 +15,13 @@
 
 ```bash
 pip install httpware                # core only — no decoder
-pip install httpware[pydantic]      # + PydanticDecoder (the default-decoder path)
-pip install httpware[msgspec]       # + MsgspecDecoder
+pip install httpware[pydantic]      # + PydanticDecoder — handles BaseModel + dataclasses + primitives + generics
+pip install httpware[msgspec]       # + MsgspecDecoder — handles Struct + dataclasses + primitives + generics
+pip install httpware[pydantic,msgspec]   # both extras — both decoders register; BaseModel routes to pydantic, Struct to msgspec
 pip install httpware[all]           # everything declared above (pydantic, msgspec, otel)
 ```
 
-`AsyncClient()` with no `decoder=` argument defaults to constructing a `PydanticDecoder`; that path requires the `pydantic` extra and raises `ImportError` at `AsyncClient.__init__` if it is missing.
+`AsyncClient()` resolves `decoders=None` against installed extras: pydantic if installed (first), msgspec if installed (second), or an empty tuple if neither. `AsyncClient()` never raises on missing extras — failure is deferred to the first `response_model=` call, where `MissingDecoderError` fires *before* the HTTP request if no registered decoder claims the model.
 
 ## Quickstart
 

docs/errors.md

@@ -27,7 +27,8 @@ ClientError                          (catch-all for anything httpware raises)
 │       └── ServiceUnavailableError (503)
 ├── RetryBudgetExhaustedError       (a retry was needed but the budget refused)
 ├── BulkheadFullError                (acquire_timeout elapsed before a slot opened)
-└── DecodeError                      (response_model= decoder failed; HTTP call itself succeeded)
+├── DecodeError                      (response_model= decoder failed; HTTP call itself succeeded)
+└── MissingDecoderError              (no registered decoder claims response_model=; fires before the HTTP call)
 ```
 
 ## Status-to-exception mapping
@@ -155,6 +156,20 @@ except DecodeError as exc:
     raise
 ```
 
+## `MissingDecoderError`
+
+Raised by `send()` / `send_with_response()` / verb methods when `response_model=` is set but no registered decoder claims the model. Carries:
+
+- `model: type` — the `response_model=` value that wasn't claimed.
+- `registered_names: tuple[str, ...]` — class names of the registered decoders that all rejected the model. Empty tuple means no decoders were registered.
+
+Corrective action depends on the message hint:
+
+- `no decoders registered. Install pip install httpware[pydantic] or pip install httpware[msgspec], or pass decoders=[...] explicitly.` — install an extra or pass an explicit decoder list.
+- `registered decoders (PydanticDecoder + MsgspecDecoder) all rejected it.` — your `response_model` type is exotic enough that neither built-in claims it. Pass a custom `ResponseDecoder` via `decoders=[...]`.
+
+Unlike `DecodeError`, this error fires *before* the HTTP request — no traffic is sent.
+
 ## See also
 
 - **[Resilience reference](resilience.md)** — `AsyncRetry`, `RetryBudget`, `AsyncBulkhead` parameter tables.

docs/index.md

@@ -13,8 +13,9 @@ pip install httpware
 Optional extras:
 
 ```bash
-pip install httpware[pydantic]   # PydanticDecoder (the default decoder path)
-pip install httpware[msgspec]    # MsgspecDecoder
+pip install httpware[pydantic]   # PydanticDecoder — handles BaseModel + dataclasses + primitives + generics
+pip install httpware[msgspec]    # MsgspecDecoder — handles Struct + dataclasses + primitives + generics
+pip install httpware[pydantic,msgspec]   # both extras — both decoders register; BaseModel routes to pydantic, Struct to msgspec
 ```
 
 ## First request
@@ -64,6 +65,31 @@ async def main() -> None:
 
 Need the raw response **and** a decoded body from the same call (e.g., for header-based pagination)? See [Link header pagination](recipes/link-header-pagination.md) — it uses `send_with_response`.
 
+### Decoder dispatch
+
+When `response_model=` is set, the client walks `decoders` in order and picks
+the first decoder whose `can_decode(model)` returns `True`. Both built-in
+decoders claim broadly within their library; the ordering encodes your
+preference for shared shapes (`dict`, `list[Foo]`, dataclasses, primitives):
+
+```python
+# pydantic-first (the default when both extras are installed):
+# - BaseModel  -> pydantic
+# - Struct     -> msgspec
+# - dict, list -> pydantic (first in list)
+AsyncClient(decoders=[PydanticDecoder(), MsgspecDecoder()])
+
+# msgspec-first — same native routing, but shared shapes go to msgspec:
+# - BaseModel  -> pydantic
+# - Struct     -> msgspec
+# - dict, list -> msgspec
+AsyncClient(decoders=[MsgspecDecoder(), PydanticDecoder()])
+```
+
+If no registered decoder claims your `response_model`, the call raises
+`MissingDecoderError` *before* the HTTP request — see the
+[Errors reference](errors.md#missingdecodererror).
+
 ### With resilience middleware
 
 Compose resilience middleware at construction; `AsyncBulkhead` goes outside `AsyncRetry` so one slot covers all retry attempts.
@@ -109,7 +135,7 @@ All errors inherit `httpware.ClientError`. The categories:
 - **Status errors** (4xx/5xx responses) — raised automatically, no `raise_for_status()` needed: `NotFoundError`, `RateLimitedError`, `ServiceUnavailableError`, and the rest. All subclass `StatusError`.
 - **Transport errors** — connection / network / protocol failures before a response arrived. `NetworkError` (transient) subclasses `TransportError`.
 - **Resilience refusals** — `RetryBudgetExhaustedError` and `BulkheadFullError`, raised by the resilience middleware.
-- **Decode errors** — `DecodeError`, raised when `response_model=` decoding fails (HTTP call itself succeeded).
+- **Decode errors** — `DecodeError`, raised when `response_model=` decoding fails (HTTP call itself succeeded). `MissingDecoderError`, raised when no registered decoder claims the `response_model=` type — fires *before* the HTTP call.
 
 See the [Errors reference](errors.md) for the full tree and catching strategies.
 

planning/engineering.md

@@ -4,7 +4,7 @@ This doc is the single distilled reference for `httpware` design rationale, prot
 
 ## 1. Project intent
 
-`httpware` is a thin opinionated wrapper around `httpx2`. It re-exports `httpx2.Request` and `httpx2.Response` as the public request/response surface and adds three things on top: typed response decoding (via a `ResponseDecoder` protocol; pydantic and msgspec are both opt-in extras as of 0.3.0), a middleware chain composed at client construction, and a status-keyed exception tree raised automatically on 4xx and 5xx. `AsyncClient(decoder=None)` defaults to constructing a `PydanticDecoder` and so requires the `pydantic` extra; callers can supply an explicit `decoder=` argument to escape the default. As of 0.4.0, the package ships a small resilience suite under `httpware.middleware.resilience` — a `Retry` middleware with a Finagle-style `RetryBudget`, plus a `Bulkhead` concurrency limiter — composed via the standard middleware chain. As of 0.5.0, `AsyncClient.stream()` provides a context-manager API for chunked response bodies; it bypasses the middleware chain by design (see planning/archive/specs/2026-06-05-streaming-design.md). As of 0.6.0, `Retry` and `Bulkhead` emit operational events via stdlib `logging` records (`httpware.retry` / `httpware.bulkhead` loggers) and — when `opentelemetry-api` is installed — OpenTelemetry span events on the active span. As of 0.7.0, the first-cut user-docs surface is live at <https://httpware.readthedocs.io/> (Middleware, Resilience, Errors, Testing guides) and Epic 3 is closed.
+`httpware` is a thin opinionated wrapper around `httpx2`. It re-exports `httpx2.Request` and `httpx2.Response` as the public request/response surface and adds three things on top: typed response decoding (via a `ResponseDecoder` protocol; pydantic and msgspec are both opt-in extras as of 0.3.0), a middleware chain composed at client construction, and a status-keyed exception tree raised automatically on 4xx and 5xx. As of 0.9.0, both clients take `decoders: Sequence[ResponseDecoder] | None = None` (a *list*, not a single instance) and dispatch via each decoder's `can_decode(model)` predicate; the default resolves against installed extras (pydantic-first when both present) and `AsyncClient()` / `Client()` no longer raise on missing extras. A new `MissingDecoderError` (sibling of `DecodeError` under `ClientError`) fires before the HTTP call when `response_model=` is set but no registered decoder claims the model. As of 0.4.0, the package ships a small resilience suite under `httpware.middleware.resilience` — a `Retry` middleware with a Finagle-style `RetryBudget`, plus a `Bulkhead` concurrency limiter — composed via the standard middleware chain. As of 0.5.0, `AsyncClient.stream()` provides a context-manager API for chunked response bodies; it bypasses the middleware chain by design (see planning/archive/specs/2026-06-05-streaming-design.md). As of 0.6.0, `Retry` and `Bulkhead` emit operational events via stdlib `logging` records (`httpware.retry` / `httpware.bulkhead` loggers) and — when `opentelemetry-api` is installed — OpenTelemetry span events on the active span. As of 0.7.0, the first-cut user-docs surface is live at <https://httpware.readthedocs.io/> (Middleware, Resilience, Errors, Testing guides) and Epic 3 is closed.
 
 As of 0.8.0 the async middleware surface uses the `Async*`/`async_*` prefix (aligning with httpx2's convention); the `attempt_timeout=` kwarg was removed from `AsyncRetry` in the same release — see `planning/specs/2026-06-07-sync-client-design.md` for the rationale.
 
@@ -36,11 +36,15 @@ The 0.1.0 seams numbered 1 (Middleware↔Transport) and 4 (Transport↔httpx2) h
 - **Contract:** the middleware chain is composed once at client construction and frozen for the client's lifetime. Both worlds follow the same contract; the only difference is the per-world type: `AsyncClient` composes `AsyncMiddleware` via `compose_async` (the continuation type is `AsyncNext`), and `Client` composes `Middleware` via `compose` (the continuation type is `Next`). Both `compose` and `compose_async` live in `src/httpware/middleware/chain.py`. The chain bottom (the "terminal") is internal: it calls `self._httpx2_client.send(request)`, maps `httpx2` errors to `httpware` errors, and raises a `StatusError` subclass on 4xx/5xx. Same lifecycle rules in both worlds.
 - **Rule:** mutating the chain after construction is not supported. Per-request behavior goes through `httpx2.Request.extensions` or through `extensions=` kwargs at call sites.
 
-### Seam B: `Client`/`AsyncClient` ↔ `ResponseDecoder`
+### Seam B: `Client`/`AsyncClient` ↔ `ResponseDecoder` list
 
 - **Where:** `src/httpware/client.py` ↔ `src/httpware/decoders/`.
-- **Contract:** the decoder is invoked when the caller passes `response_model=`. The protocol is `decode(content: bytes, model: type[T]) -> T`. Any exception raised by `decode` is wrapped by the call sites in `client.py` — `Client.send` / `AsyncClient.send` (when `response_model=` is set) and `Client.send_with_response` / `AsyncClient.send_with_response` — into `httpware.DecodeError` (a `ClientError` subclass carrying `response`, `model`, `original`). Decoder implementers do not need to raise `DecodeError` directly.
-- **Rule:** the decoder must operate on raw bytes in a single parse pass. Two-pass decoding (`json.loads` then `validate_python`) is rejected: a single bytes-in / typed-object-out pass avoids the redundant intermediate `dict` allocation and parses faster. The Pydantic adapter implements this as `TypeAdapter(model).validate_json(content)`, with the `TypeAdapter` itself memoized via `@functools.lru_cache(maxsize=1024)` on a module-level `_get_adapter(model)` factory (the adapter is the expensive part to build). The msgspec adapter implements it as `msgspec.json.decode(content, type=model)`.
+- **Contract:** the client holds `_decoders: tuple[ResponseDecoder, ...]` composed at `__init__` and frozen for the client's lifetime. The Protocol exposes two methods:
+    - `can_decode(model: type) -> bool` — predicate used at send-time to walk `_decoders` and pick the first claiming decoder (`_dispatch_decoder` on both classes). Built-in decoders claim broadly (pydantic via `TypeAdapter(model)` probe, msgspec via `msgspec.inspect.type_info(model)` + `CustomType` filter); list ordering decides ambiguous shared shapes (dataclass, primitive, generic). Native types of another library MUST be rejected.
+    - `decode(content: bytes, model: type[T]) -> T` — the decode itself. Any exception is wrapped by `Client.send` / `AsyncClient.send` (when `response_model=` is set) and `Client.send_with_response` / `AsyncClient.send_with_response` into `httpware.DecodeError` (a `ClientError` subclass carrying `response`, `model`, `original`). Decoder implementers do not need to raise `DecodeError` directly.
+- **Pre-flight check:** when `response_model=` is set and no decoder claims it, `send` / `send_with_response` raise `MissingDecoderError(model=..., registered_names=...)` BEFORE the HTTP call. Distinct from `DecodeError` (which means the decoder ran and the payload was malformed); distinct corrective actions (install an extra or pass `decoders=[...]`).
+- **Default list:** `decoders=None` resolves via `client.py:_build_default_decoders()` against installed extras — pydantic-first when both are present, either-only when only one is installed, empty tuple when neither. `AsyncClient()` / `Client()` never raise on missing extras; failure surfaces only at the first `response_model=` use site.
+- **Rule:** the decoder must operate on raw bytes in a single parse pass. Two-pass decoding (`json.loads` then `validate_python`) is rejected: a single bytes-in / typed-object-out pass avoids the redundant intermediate `dict` allocation and parses faster. The Pydantic adapter implements this as `TypeAdapter(model).validate_json(content)`, with the `TypeAdapter` itself memoized via `@functools.lru_cache(maxsize=1024)` on a module-level `_get_adapter(model)` factory; the msgspec adapter mirrors the pattern with a cached `msgspec.json.Decoder(model)`.
 
 ### Seam C: `httpware ↔ optional extras`
 
@@ -118,7 +122,7 @@ Each extra's code lives in a single dedicated module (`decoders/pydantic.py`, `d
 
 New extras are added at the same time as the code that uses them — never preemptively. (An `otel` extra existed pre-0.4 but was removed once we noticed it was advertising functionality that didn't exist. 0.6.0 reintroduces it paired with the code that uses it — `Retry` and `Bulkhead` add events to the active OpenTelemetry span via `trace.get_current_span().add_event(...)`.)
 
-Caller-facing pattern: consumers select the implementation by passing it explicitly, e.g., `AsyncClient(decoder=PydanticDecoder())`. There is no auto-detection or implicit registry.
+Caller-facing pattern: as of 0.9.0, `AsyncClient()` / `Client()` default `decoders=None` resolves via `_build_default_decoders()` against installed extras (pydantic-first when both are present; empty tuple when neither). Consumers override by passing `decoders=[...]` explicitly; `decoders=[]` is honored as an opt-out. The auto-resolution is a snapshot of `import_checker.is_<extra>_installed` flags at `__init__` time — there is no runtime re-detection or implicit registry beyond the two built-in decoders.
 
 ## 8. Remaining roadmap