coze-coding-utils 0.2.4__tar.gz → 0.2.5__tar.gz

{coze_coding_utils-0.2.4 → coze_coding_utils-0.2.5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: coze-coding-utils
-Version: 0.2.4
+Version: 0.2.5
 Summary: Utilities for Coze coding client runtime context and helpers.
 Project-URL: Homepage, https://code.byted.org/stone/coze-coding-client
 Author: Bytedance Stone Team

{coze_coding_utils-0.2.4 → coze_coding_utils-0.2.5}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "coze-coding-utils"
-version = "0.2.4"
+version = "0.2.5"
 description = "Utilities for Coze coding client runtime context and helpers."
 readme = "README.md"
 requires-python = ">=3.10"

{coze_coding_utils-0.2.4 → coze_coding_utils-0.2.5}/src/coze_coding_utils/helper/agent_helper.py

@@ -1,7 +1,8 @@
 import uuid
 import json
-import os
-from typing import Any, Dict, List, Tuple, Iterator
+import requests
+from urllib.parse import urlparse, urlencode, parse_qs, urljoin
+from typing import Any, Dict, List, Tuple, Iterator, Optional
 import time
 from coze_coding_utils.file.file import File, FileOps, infer_file_category
 from coze_coding_utils.error import classify_error
@@ -31,6 +32,78 @@ from coze_coding_utils.messages.server import (
 )
 
 
+_VIDEO_MIME_MAP: Dict[str, str] = {
+    "mp4": "video/mp4",
+    "avi": "video/x-msvideo",
+    "mov": "video/quicktime",
+    "mkv": "video/x-matroska",
+    "flv": "video/x-flv",
+    "wmv": "video/x-ms-wmv",
+    "webm": "video/webm",
+    "m4v": "video/x-m4v",
+    "3gp": "video/3gpp",
+}
+
+# 下游 LLM API 实际支持的视频 MIME 类型白名单
+_SUPPORTED_VIDEO_MIMES = {
+    "video/mp4",
+    "video/quicktime",
+    "video/webm",
+    "video/3gpp",
+    "video/x-m4v",
+}
+
+
+def _probe_remote_mime(url: str) -> Optional[str]:
+    """HEAD 请求探测远程文件的真实 Content-Type，失败时返回 None"""
+    try:
+        resp = requests.head(url, timeout=5, allow_redirects=True)
+        ct = resp.headers.get("Content-Type", "")
+        return ct.split(";")[0].strip() or None
+    except Exception:
+        return None
+
+
+def _ensure_video_content_type(url: str, mime_type: str) -> str:
+    """
+    对象存储（TOS/S3）的 presigned URL 可能返回 application/octet-stream。
+    通过追加 response-content-type 参数让服务端在响应中覆盖 Content-Type，
+    从而使 LLM API fetch URL 时能拿到正确的 video MIME，而不会因 octet-stream 被拒。
+    已有签名不受影响（TOS/S3 response-* 参数不计入签名哈希）。
+    """
+    parsed = urlparse(url)
+    # 已经有正确的 response-content-type，不重复添加
+    qs = parse_qs(parsed.query, keep_blank_values=True)
+    if qs.get("response-content-type") == [mime_type]:
+        return url
+    sep = "&" if parsed.query else "?"
+    return url + sep + "response-content-type=" + requests.utils.quote(mime_type, safe="")
+
+
+def _build_video_parts(
+    file_data: File, file_info: Any, file_ext: str, mime_override: Optional[str] = None
+) -> List[Dict[str, Any]]:
+    """将视频文件转换为 content_parts，不支持的格式退化为文本描述。
+    mime_override: 外部已探测好的 MIME，传入时跳过内部探测，避免重复 HEAD 请求。
+    """
+    ext = file_ext.lstrip(".").lower()
+    mime_type = mime_override or _VIDEO_MIME_MAP.get(ext)
+    if not mime_type and file_data.is_remote:
+        mime_type = _probe_remote_mime(file_info.url)
+
+    if mime_type and mime_type in _SUPPORTED_VIDEO_MIMES:
+        # 追加 response-content-type 参数，确保对象存储返回正确的 Content-Type
+        # 防止 LLM API fetch URL 时因拿到 application/octet-stream 而报错
+        video_url = _ensure_video_content_type(file_info.url, mime_type) if file_data.is_remote else file_info.url
+        return [
+            {"type": "text", "text": f'{file_data.url}'},
+            {"type": "video_url", "video_url": {"url": video_url, "media_type": mime_type}},
+        ]
+    else:
+        fmt_hint = mime_type or (f".{ext}" if ext else "unknown")
+        return [{"type": "text", "text": f"video url: {file_info.url} (format: {fmt_hint})"}]
+
+
 def to_stream_input(msg: ClientMessage) -> Dict[str, Any]:
     content_parts = []
     if msg and msg.content and msg.content.query and msg.content.query.prompt:
@@ -43,7 +116,23 @@ def to_stream_input(msg: ClientMessage) -> Dict[str, Any]:
                     and block.content.upload_file
             ):
                 file_info = block.content.upload_file
-                file_type, _ = infer_file_category(file_info.url)
+                # 第一轮：从 URL 后缀识别类型
+                file_type, file_ext = infer_file_category(file_info.url)
+                # 第二轮：URL 无后缀时用 file_name 兜底
+                if file_type == "default" and file_info.file_name:
+                    file_type, file_ext = infer_file_category(file_info.file_name)
+                # 第三轮：两轮扩展名都无法识别时，HEAD 探测真实 Content-Type
+                probed_mime: Optional[str] = None
+                if file_type == "default" and file_info.url.startswith(("http://", "https://")):
+                    probed_mime = _probe_remote_mime(file_info.url)
+                    if probed_mime:
+                        top = probed_mime.split("/")[0]
+                        if top == "video":
+                            file_type = "video"
+                        elif top == "image":
+                            file_type = "image"
+                        elif top == "audio":
+                            file_type = "audio"
                 file_data = File(url=file_info.url, file_type=file_type)
                 # check is image
                 if file_data.file_type == "image":
@@ -61,18 +150,7 @@ def to_stream_input(msg: ClientMessage) -> Dict[str, Any]:
                     )
                 # check is video
                 elif file_data.file_type == "video":
-                    content_parts.append(
-                        {
-                            "type": "text",
-                            "text": f'{file_data.url}'
-                        }
-                    )
-                    content_parts.append(
-                        {
-                            "type": "video_url",
-                            "video_url": {"url": file_info.url},
-                        }
-                    )
+                    content_parts.extend(_build_video_parts(file_data, file_info, file_ext, probed_mime))
                 # check is audio
                 elif file_data.file_type == "audio":
                     content_parts.append(

{coze_coding_utils-0.2.4 → coze_coding_utils-0.2.5}/src/coze_coding_utils/openai/converter/request_converter.py

@@ -6,6 +6,10 @@ from coze_coding_utils.openai.types.request import (
     ChatMessage,
 )
 from coze_coding_utils.file.file import File, FileOps, infer_file_category
+from coze_coding_utils.helper.agent_helper import (
+    _VIDEO_MIME_MAP,
+    _ensure_video_content_type,
+)
 
 
 class RequestConverter:
@@ -110,6 +114,15 @@ class RequestConverter:
             video_url_data = part.get("video_url", {})
             url = video_url_data.get("url", "")
             if url:
+                _, file_ext = infer_file_category(url)
+                ext = file_ext.lstrip(".").lower()
+                mime_type = _VIDEO_MIME_MAP.get(ext)
+                if mime_type:
+                    fixed_url = _ensure_video_content_type(url, mime_type)
+                    return [
+                        {"type": "text", "text": url},
+                        {"type": "video_url", "video_url": {"url": fixed_url, "media_type": mime_type}},
+                    ]
                 return [
                     {"type": "text", "text": url},
                     {"type": "video_url", "video_url": {"url": url}},
@@ -147,6 +160,15 @@ class RequestConverter:
                     {"type": "image_url", "image_url": {"url": url}},
                 ]
             elif file_type == "video":
+                _, file_ext = infer_file_category(url)
+                ext = file_ext.lstrip(".").lower()
+                mime_type = _VIDEO_MIME_MAP.get(ext)
+                if mime_type:
+                    fixed_url = _ensure_video_content_type(url, mime_type)
+                    return [
+                        {"type": "text", "text": url},
+                        {"type": "video_url", "video_url": {"url": fixed_url, "media_type": mime_type}},
+                    ]
                 return [
                     {"type": "text", "text": url},
                     {"type": "video_url", "video_url": {"url": url}},