Merge pull request #437 from bd-mkt/invalid-attachments

fix for issue with parsing pdf attachment such as *.joboptions files

CHANGELOG.md

@@ -1,6 +1,10 @@
 # Changelog
 ## [Unreleased]
 
+### Fixed
+
+- Embedded PDF attachment extension is derived from the attachment filename, not the parent's synthetic `...#attachment=<name>` URI; non-PDF attachments (e.g. `.joboptions`) are no longer misrouted to docling's PDF backend, and unsupported extensions are skipped.
+
 ## [0.57.0] - 2026-06-11
 
 ### Added

haiku_rag_slim/haiku/rag/client/documents.py

@@ -332,16 +332,25 @@ async def _ingest_fetch_result(
     stored_uri: str,
     existing_doc: Document | None,
     depth: int = 0,
+    filename: str | None = None,
 ) -> Document:
     """Convert / chunk / embed / store a fetched document. Replaces an
     existing document if one is supplied. ``depth`` tracks position in an
-    attachment chain so the reconciliation step can bound recursion."""
+    attachment chain so the reconciliation step can bound recursion.
+
+    ``filename``, when given, makes its suffix authoritative for the file
+    extension (and thus the docling format), overriding the URI/content-type
+    fallback. Callers pass it when ``result.uri`` cannot yield the right
+    extension, e.g. embedded attachments whose name lives in a URI fragment."""
     from haiku.rag.embeddings import embed_chunks
 
     converter = get_converter(client._config)
-    file_extension = get_extension_from_content_type_or_url(
-        result.uri, result.content_type
-    )
+    if filename is not None:
+        file_extension = Path(filename).suffix.lower()
+    else:
+        file_extension = get_extension_from_content_type_or_url(
+            result.uri, result.content_type
+        )
     if file_extension not in converter.supported_extensions:
         raise UnsupportedSourceError(
             f"Unsupported content type/extension: {result.content_type}/{file_extension}"
@@ -508,12 +517,15 @@ async def _reconcile_pdf_attachments(
                 stored_uri=child_uri,
                 existing_doc=existing_child,
                 depth=depth + 1,
+                filename=name,
             )
         except UnsupportedSourceError:
             logger.warning(
-                "Skipping attachment %r in %s: unsupported content type %r",
+                "Skipping attachment %r in %s: unsupported extension %r "
+                "(content type %r)",
                 name,
                 parent_doc.uri,
+                Path(name).suffix.lower(),
                 content_type,
             )
 

tests/test_pdf_attachments.py

@@ -33,6 +33,7 @@ async def fake_ingest_fetch_result(
     stored_uri,
     existing_doc,
     depth=0,
+    filename=None,
 ):
     """A stand-in for ``_ingest_fetch_result`` that skips docling/embedder
     entirely: it writes the document with content_type/md5/parent_uri set
@@ -306,6 +307,7 @@ async def test_unsupported_attachment_continues_loop(temp_db_path, monkeypatch):
         stored_uri,
         existing_doc,
         depth=0,
+        filename=None,
     ):
         if stored_uri.endswith("unsupported.xyz"):
             raise UnsupportedSourceError("nope")
@@ -329,6 +331,24 @@ async def test_unsupported_attachment_continues_loop(temp_db_path, monkeypatch):
         assert {c.uri for c in children} == {f"{parent_uri}#attachment=ok.txt"}
 
 
+async def test_joboptions_attachment_skipped_not_routed_as_pdf(temp_db_path, caplog):
+    """An attachment's own name drives its extension, not the synthetic
+    ``...#attachment=<name>`` URI (whose fragment the URL-suffix fallback drops,
+    inheriting the parent's ``.pdf``). ``.joboptions`` is unsupported, so the
+    child is skipped before any converter/embedder call (real ingest, no fake)."""
+    import logging
+
+    pdf_bytes = build_pdf([("Press Quality.joboptions", b"/CompressObjects /Tags\n")])
+    async with HaikuRAG(temp_db_path, create=True) as client:
+        parent_uri = "file:///fixtures/brochure.pdf"
+        parent = await _make_parent(client, parent_uri, pdf_bytes)
+
+        with caplog.at_level(logging.WARNING, logger="haiku.rag.client.documents"):
+            await _reconcile_pdf_attachments(client, parent, pdf_bytes, depth=0)
+
+        assert await client.list_documents(filter=parent_uri_filter(parent_uri)) == []
+
+
 async def test_cascade_delete_removes_reconciled_children(temp_db_path, monkeypatch):
     monkeypatch.setattr(
         "haiku.rag.client.documents._ingest_fetch_result",