fix: partition_email raises KeyError on a multipart/* attachment

CHANGELOG.md

@@ -17,6 +17,8 @@
 ### Fixes
 
 - **Reject an empty `tokenizer` when chunking by `max_tokens`**: `""` is not `None`, so it slipped past the "tokenizer is required" check while still leaving the chunkers without a token counter — the window was then silently measured in characters, making `max_tokens=20` mean 20 characters. It now raises the same `ValueError` as omitting `tokenizer` altogether.
+- **`partition_email()` no longer crashes on a `multipart/*` attachment**: a multipart sub-part surfaced as an "attachment" (e.g. a PGP/MIME-signed forwarded message) previously raised an uncaught `KeyError` from `email.contentmanager`, which has no `get_content()` handler for any `multipart/*` content-type. `_AttachmentPartitioner._file_bytes` now falls back to the part's raw serialized bytes for this case instead.
+- **`partition_email()` no longer drops a `multipart/*` attachment's own body**: serializing such an attachment verbatim left its own `Content-Disposition: attachment` header intact, so re-parsing it as the root of a new message made `EmailMessage.get_body()` skip it entirely instead of finding its body part. The attachment's raw bytes are now serialized from a copy with that header stripped.
 
 ## 0.25.1
 

example-docs/eml/mime-attach-multipart-signed.eml

@@ -0,0 +1,36 @@
+From: sender@example.com
+To: recipient@example.com
+Subject: Test with PGP-signed forwarded message attachment
+MIME-Version: 1.0
+Content-Type: multipart/mixed; boundary="===============6075316962646898834=="
+
+--===============6075316962646898834==
+Content-Type: text/plain; charset="utf-8"
+Content-Transfer-Encoding: 7bit
+
+This is the main email body.
+
+--===============6075316962646898834==
+Content-Type: multipart/signed; protocol="application/pgp-signature";
+ micalg="pgp-sha256"; boundary="===============3567714482309737971=="
+Content-Disposition: attachment; filename="signed-message.eml"
+MIME-Version: 1.0
+
+--===============3567714482309737971==
+Content-Type: text/plain; charset="utf-8"
+Content-Transfer-Encoding: 7bit
+MIME-Version: 1.0
+
+This is the signed forwarded message body.
+
+--===============3567714482309737971==
+MIME-Version: 1.0
+Content-Type: application/pgp-signature
+
+-----BEGIN PGP SIGNATURE-----
+fakefakefake
+-----END PGP SIGNATURE-----
+
+--===============3567714482309737971==--
+
+--===============6075316962646898834==--

test_unstructured/partition/test_email.py

@@ -389,6 +389,32 @@ def test_partition_email_silently_skips_attachments_it_cannot_partition():
     ]
 
 
+def test_partition_email_does_not_raise_on_multipart_attachment():
+    """A multipart/* "attachment" (e.g. a PGP/MIME-signed forwarded message) must not crash.
+
+    `email.contentmanager` has no `get_content()` handler for any `multipart/*` content-type, so
+    without special-casing this in `_AttachmentPartitioner._file_bytes`, resolving such an
+    attachment's bytes raises `KeyError` -- crashing the whole `partition_email()` call instead
+    of gracefully processing or skipping just that one attachment. Regression test for #3922.
+
+    The fixture's attachment is itself a genuine boundary-bearing `multipart/signed` MIME part
+    (a PGP-signed forwarded message), so this also guards a second, subtler bug: serializing
+    that part verbatim (with its own `Content-Disposition: attachment` header intact) and
+    re-parsing it as a new root message made `EmailMessage.get_body()` skip it entirely, i.e.
+    the attachment partitioned to zero elements instead of surfacing its signed body text.
+    """
+    elements = partition_email(
+        example_doc_path("eml/mime-attach-multipart-signed.eml"), process_attachments=True
+    )
+
+    # -- No exception; the outer email body is partitioned ... --
+    assert elements[0] == NarrativeText("This is the main email body.")
+    # -- ... and the signed attachment's own body is also partitioned, not silently dropped. --
+    assert elements[1] == NarrativeText("This is the signed forwarded message body.")
+    assert elements[1].metadata.filename == "signed-message.eml"
+    assert elements[1].metadata.attached_to_filename == "mime-attach-multipart-signed.eml"
+
+
 # ================================================================================================
 # ISOLATED UNIT TESTS
 # ================================================================================================

unstructured/partition/email.py

@@ -5,6 +5,7 @@
 
 from __future__ import annotations
 
+import copy
 import datetime as dt
 import email
 import email.policy
@@ -432,10 +433,36 @@ def _attachment_file_name(self) -> str | None:
     @cached_property
     def _file_bytes(self) -> bytes:
         """The bytes of the attached file."""
+        # -- `email.contentmanager` has no `get_content()` handler registered for any
+        # -- `multipart/*` content-type (handlers exist for text/*, application/*, image/*,
+        # -- message/rfc822, etc., but multipart sub-parts are normally consumed via
+        # -- `.iter_parts()`, not `.get_content()`). A multipart sub-part can still appear as an
+        # -- "attachment" though, e.g. a PGP/MIME-signed forwarded message (`multipart/signed`)
+        # -- nested inside a `multipart/mixed` envelope. Fall back to the part's raw serialized
+        # -- bytes in that case rather than letting `get_content()` raise `KeyError`.
+        if self._attachment.get_content_type().startswith("multipart/"):
+            return self._serialized_multipart_bytes
+
         content = self._attachment.get_content()
 
         if isinstance(content, str):
             return content.encode("utf-8")
 
         assert isinstance(content, bytes)
         return content
+
+    @cached_property
+    def _serialized_multipart_bytes(self) -> bytes:
+        """Raw serialized bytes of a `multipart/*` attachment, re-parseable as a document.
+
+        `EmailMessage.get_body()` skips any candidate part -- including a `multipart/*` root
+        message -- whose own `Content-Disposition` is "attachment". This MIME part carries
+        exactly that header (that's how it surfaced as an attachment via `iter_attachments()`
+        in the first place), so serializing it verbatim and re-parsing it as the root of a new
+        message would make `partition_email()` find no body at all. Strip that header on a
+        *copy* before serializing -- `self._attachment` itself is left untouched since
+        `_attachment_file_name` still needs its original `Content-Disposition` filename param.
+        """
+        part = copy.deepcopy(self._attachment)
+        del part["Content-Disposition"]
+        return part.as_bytes()