feat(v2.0.1): align audio with current speech models and protocols (2.0.1)

.pre-commit-config.yaml

@@ -30,17 +30,23 @@ repos:
     rev: v3.1.0
     hooks:
       - id: add-trailing-comma
-        exclude: ^tests/legacy/
+        # examples/01_atomic_capabilities/audio is ported verbatim from the sibling repo's
+        # samples/ tree, whose hooks exempt samples/.
+        exclude: ^(tests/legacy|examples/01_atomic_capabilities/audio)/
   - repo: https://github.com/pre-commit/mirrors-mypy
     rev: v1.7.0
     hooks:
       - id: mypy
+        # examples/01_atomic_capabilities/audio: mypy cannot map the
+        # digit-leading 01_atomic_capabilities segment to a module name
+        # (reports duplicate _common).
         exclude:
           (?x)(
           pb2\.py$
           | grpc\.py$
           | ^docs
           | ^tests/legacy/
+          | ^examples/01_atomic_capabilities/audio/
           | \.html$
           | \.pyi$
           )
@@ -61,13 +67,13 @@ repos:
     hooks:
       - id: black
         args: [ --line-length=79 ]
-        exclude: ^tests/legacy/
+        exclude: ^(tests/legacy|examples/01_atomic_capabilities/audio)/
   - repo: https://github.com/PyCQA/flake8
     rev: 6.1.0
     hooks:
       - id: flake8
         args: [ "--extend-ignore=E203"]
-        exclude: ^tests/legacy/
+        exclude: ^(tests/legacy|examples/01_atomic_capabilities/audio)/
   - repo: https://github.com/pylint-dev/pylint
     rev: v3.0.2
     hooks:
@@ -76,6 +82,7 @@ repos:
           (?x)(
           ^docs
           | ^tests/
+          | ^examples/01_atomic_capabilities/audio/
           | pb2\.py$
           | grpc\.py$
           | \.demo$
@@ -168,6 +175,15 @@ repos:
           # Long sequential arrange/act scripts are normal in integration
           # tests; splitting them for R0915 hurts readability.
           --disable=R0915,
+          # Table-driven tests are branchy by nature (R0912); test doubles
+          # deliberately hold resources open to assert lifecycle order
+          # (R1732); passthrough lambda seams read clearer than direct
+          # method refs (W0108); explicit trailing ``return`` marks the
+          # end of scripted callbacks (R1711).
+          --disable=R0912,
+          --disable=R1732,
+          --disable=W0108,
+          --disable=R1711,
         ]
   - repo: https://github.com/pre-commit/mirrors-eslint
     rev: v7.32.0

dashscope/__init__.pyi

@@ -11,6 +11,15 @@ from typing import (
     Union,
 )
 
+# ---------------------------------------------------------------------------
+# dashscope 2.0
+# ---------------------------------------------------------------------------
+from dashscope.v2 import AsyncClient as AsyncClient
+from dashscope.v2 import Client as Client
+from dashscope.v2 import WsConnection as WsConnection
+from dashscope.v2 import chat as chat
+from dashscope.v2 import settings as settings
+
 # ---------------------------------------------------------------------------
 # Config
 # ---------------------------------------------------------------------------

dashscope/v2/client.py

@@ -10,7 +10,6 @@
 
 from pathlib import Path
 from typing import Any, Dict, Mapping, Optional, Union
-import warnings
 
 import httpx
 
@@ -117,33 +116,14 @@ def __init__(
         self.models = Models(self)
         self.images = Images(self)
         self.videos = Videos(self)
-        self._audio = None
+        self.audio = Audio(self)
         self.fine_tunes = FineTunes(self)
         self.deployments = Deployments(self)
         self.rl = Rl(self)
         self.apps = Apps(self)
         self.tingwu = TingWu(self)
         self._agentstudio = None
 
-    @property
-    def audio(self) -> "Audio":
-        """Audio capabilities (lazily mounted).
-
-        Preview: the 2.0 audio API surface is not final and may change
-        in a future release.
-        """
-        if self._audio is None:
-            warnings.warn(
-                "dashscope 2.0 audio API is a preview and may change "
-                "in a future release.",
-                FutureWarning,
-                stacklevel=2,
-            )
-            self._audio = Audio(self)
-        audio = self._audio
-        assert audio is not None
-        return audio
-
     @property
     def reinforcement_learning(self) -> "Rl":
         """Long-form alias for ``client.rl``."""
@@ -266,11 +246,11 @@ def ws_connection(
         Credentials, region and timeout come from this client's config.
         Pass the connection as ``connection=`` to the realtime audio
         sessions (``audio.asr.recognize`` / ``audio.tts.connect`` /
-        ``audio.translation.recognize`` / ``tingwu.realtime``) to run
-        tasks concurrently or back-to-back on one socket, saving the
-        handshake per task -- the 1.x Java-side connection-reuse pattern
-        for latency-sensitive audio traffic. The caller owns the
-        connection and closes it via ``WsConnection.close``.
+        ``tingwu.realtime``) to run tasks concurrently or back-to-back
+        on one socket, saving the handshake per task -- the 1.x
+        Java-side connection-reuse pattern for latency-sensitive audio
+        traffic. The caller owns the connection and closes it via
+        ``WsConnection.close``.
         """
         config = self._config
         return WsConnection(
@@ -358,33 +338,14 @@ def __init__(
         self.models = AsyncModels(self)
         self.images = AsyncImages(self)
         self.videos = AsyncVideos(self)
-        self._audio = None
+        self.audio = AsyncAudio(self)
         self.fine_tunes = AsyncFineTunes(self)
         self.deployments = AsyncDeployments(self)
         self.rl = AsyncRl(self)
         self.apps = AsyncApps(self)
         self.tingwu = AsyncTingWu(self)
         self._agentstudio = None
 
-    @property
-    def audio(self) -> "AsyncAudio":
-        """Audio capabilities (lazily mounted).
-
-        Preview: the 2.0 audio API surface is not final and may change
-        in a future release.
-        """
-        if self._audio is None:
-            warnings.warn(
-                "dashscope 2.0 audio API is a preview and may change "
-                "in a future release.",
-                FutureWarning,
-                stacklevel=2,
-            )
-            self._audio = AsyncAudio(self)
-        audio = self._audio
-        assert audio is not None
-        return audio
-
     @property
     def reinforcement_learning(self) -> "AsyncRl":
         """Long-form alias for ``client.rl``."""

dashscope/v2/errors.py

@@ -374,28 +374,52 @@ def error_from_sdk_exception(
 
 
 def _sse_error_body(text: str) -> Any:
-    """Decode an SSE error body into its first ``data:`` JSON payload.
-
-    A streaming request the gateway rejects before any output arrives
-    as a non-2xx response whose body is the error frame itself --
-    ``event:error`` plus a ``data:`` JSON line -- instead of a plain
-    JSON document. 1.x parses the same shape in
-    ``common.utils._handle_http_failed_response``; like
-    ``resources/_sse.py``, the frame is detected by line shape rather
-    than content-type, so a mislabeled header still maps. Bodies with
-    no decodable frame yield None, leaving the raw-text fallback
-    intact.
-    """
-    for line in text.splitlines():
-        if not line.startswith("data:"):
-            continue
+    """Decode an SSE error payload, preferring codes over message-only frames."""
+    data_lines = []
+    fallback = None
+
+    def flush() -> Any:
+        nonlocal fallback
+        if not data_lines:
+            return None
         try:
-            body = json.loads(line[len("data:") :])
+            bodies = [json.loads("\n".join(data_lines))]
         except json.JSONDecodeError:
-            continue
-        if isinstance(body, dict):
-            return body
-    return None
+            # Older gateways omit blank lines between complete JSON frames.
+            bodies = []
+            for data in data_lines:
+                try:
+                    bodies.append(json.loads(data))
+                except json.JSONDecodeError:
+                    continue
+        for body in bodies:
+            if not isinstance(body, dict):
+                continue
+            error = _extract(body.get("error"))
+            if body.get("code") or any(
+                error.get(key) for key in ("code", "type", "message")
+            ):
+                return body
+            if (
+                fallback is None
+                and isinstance(body.get("message"), str)
+                and body["message"]
+            ):
+                fallback = body
+        return None
+
+    for line in text.splitlines():
+        if line.startswith("data:"):
+            data_lines.append(line[len("data:") :])
+        elif line == "data":
+            data_lines.append("")
+        elif not line:
+            body = flush()
+            data_lines.clear()
+            if body is not None:
+                return body
+    body = flush()
+    return body if body is not None else fallback
 
 
 def _parse_body(response: httpx.Response) -> Any:

dashscope/v2/protocols/anthropic.py

@@ -605,7 +605,10 @@ def build_chat_request(
         # unchanged, so this converges both entry points and keeps the
         # hatch-wins precedence. Copy first: never mutate the caller's
         # dict.
-        if extra_parameters and "tool_choice" in extra_parameters:
+        if (
+            extra_parameters
+            and extra_parameters.get("tool_choice") is not None
+        ):
             extra_parameters = dict(extra_parameters)
             extra_parameters["tool_choice"] = _translate_tool_choice(
                 extra_parameters["tool_choice"],