优化生成流程

backend/app/services/attachments.py

@@ -4,8 +4,8 @@ import hashlib
 from pathlib import Path
 
 
-IMAGE_SUFFIXES = {".png", ".jpg", ".jpeg", ".webp"}
-DOCUMENT_SUFFIXES = {".txt", ".md", ".csv", ".json"}
+IMAGE_SUFFIXES = {".png", ".jpg", ".jpeg", ".webp", ".svg"}
+DOCUMENT_SUFFIXES = {".txt", ".md", ".csv", ".json", ".pdf", ".dxf", ".step", ".stp"}
 MAX_IMAGE_BYTES = 10 * 1024 * 1024
 MAX_DOCUMENT_BYTES = 2 * 1024 * 1024
 MAX_EXTRACTED_CHARS = 30_000
@@ -13,8 +13,6 @@ MAX_EXTRACTED_CHARS = 30_000
 
 def classify_upload(filename: str, mime: str, size: int) -> str:
     suffix = Path(filename).suffix.lower()
-    if suffix in {".step", ".stp"}:
-        raise ValueError("STEP/STP upload is not supported in this version")
     if suffix in IMAGE_SUFFIXES or mime.startswith("image/"):
         if size > MAX_IMAGE_BYTES:
             raise ValueError("Image upload exceeds the 10 MB limit")
@@ -23,14 +21,24 @@ def classify_upload(filename: str, mime: str, size: int) -> str:
         if size > MAX_DOCUMENT_BYTES:
             raise ValueError("Document upload exceeds the 2 MB limit")
         return "document"
-    raise ValueError("Only PNG, JPG, WEBP, TXT, MD, CSV, and JSON uploads are supported")
+    raise ValueError("Only PNG, JPG, WEBP, SVG, TXT, MD, CSV, JSON, PDF, DXF, and STEP/STP uploads are supported")
 
 
 def extract_document_text(data: bytes) -> str:
+    # PDF extraction is optional so the API can still accept a reference when
+    # the local PDF dependency is unavailable; the binary remains available as
+    # the primary attachment artifact.
+    if data.startswith(b"%PDF"):
+        try:
+            from pypdf import PdfReader
+            import io
+            return "\n".join(page.extract_text() or "" for page in PdfReader(io.BytesIO(data)).pages)[:MAX_EXTRACTED_CHARS]
+        except Exception:
+            return "[PDF reference uploaded; text extraction unavailable]"
     try:
         text = data.decode("utf-8")
     except UnicodeDecodeError as error:
-        raise ValueError("Documents must be UTF-8 text") from error
+        return "[Binary CAD reference uploaded; structured geometry inspection is deferred to the CAD reference analyzer]"
     return text[:MAX_EXTRACTED_CHARS]