admin/mcp-nextcloud
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #891 from cbcoutinho/fix/309-dav-encoding-exceptiongroup

Browse Source 
fix(vector): URL-encode DAV paths and unwrap TaskGroup exceptions
This commit is contained in:
Chris Coutinho
2026-06-11 10:39:20 +02:00
committed by GitHub
parent 4767312ad7 801bf108fa
commit 0c73fb7bff
7 changed files with 347 additions and 38 deletions
Download Patch File Download Diff File Expand all files Collapse all files
nextcloud_mcp_server/client/webdav.py
+50 -22
View File
@@ -5,7 +5,7 @@ import mimetypes
import xml.etree.ElementTree as ET
from email.utils import parsedate_to_datetime
from typing import Any, Dict, List, Optional, Tuple
from urllib.parse import unquote
from urllib.parse import quote, unquote
from xml.sax.saxutils import escape as xml_escape
from httpx import HTTPStatusError
@@ -26,11 +26,42 @@ WEBDAV_SEARCH_PAGE_SIZE = 500
WEBDAV_SEARCH_MAX_RESULTS = 50000
def _encode_dav_path(path: str) -> str:
"""Percent-encode a *decoded* DAV path for use in a request URL/header.
Paths flow through this client already URL-decoded (e.g. ``unquote`` on the
``<d:href>`` of a PROPFIND/REPORT response, or raw user-supplied paths from
MCP tools), so characters like ``#``, ``,`` and spaces reach httpx verbatim.
An unencoded ``#`` is parsed as a URL fragment and silently truncates the
request path → spurious 404 on otherwise-valid files (issue: OHR-Bench
ingest, card 309). ``quote`` with ``safe="/"`` encodes the unsafe characters
while preserving the path separators; ASCII-only paths are unchanged.
Encode exactly once: the input is decoded, so a literal ``%`` becomes
``%25`` (correct) rather than being mistaken for an existing escape.
"""
return quote(path, safe="/")
class WebDAVClient(BaseNextcloudClient):
"""Client for Nextcloud WebDAV operations."""
app_name = "webdav"
def _webdav_path(self, path: str) -> str:
"""Build the request path for ``path`` under the user's DAV root.
Percent-encodes the caller-supplied portion (see ``_encode_dav_path``)
so names with ``#``, commas, or spaces don't truncate/404; the base
``/remote.php/dav/files/<user>`` segment is left as-is.
Precondition: ``path`` is a **decoded** path (the convention everywhere
in this client — PROPFIND/REPORT hrefs are ``unquote``d before storage,
and MCP-tool inputs are raw). It is encoded exactly once, so passing an
already-encoded path would double-encode it (``%20`` → ``%2520``).
"""
return f"{self._get_webdav_base_path()}/{_encode_dav_path(path.lstrip('/'))}"
async def delete_resource(self, path: str) -> Dict[str, Any]:
"""Delete a resource (file or directory) via WebDAV DELETE."""
# Ensure path ends with a slash if it's a directory
@@ -39,7 +70,7 @@ class WebDAVClient(BaseNextcloudClient):
else:
path_with_slash = path
webdav_path = f"{self._get_webdav_base_path()}/{path_with_slash.lstrip('/')}"
webdav_path = self._webdav_path(path_with_slash)
logger.debug("Deleting WebDAV resource: %s", webdav_path)
headers = {"OCS-APIRequest": "true"}
@@ -124,15 +155,15 @@ class WebDAVClient(BaseNextcloudClient):
mime_type: Optional[str] = None,
) -> Dict[str, Any]:
"""Add/Update an attachment to a note via WebDAV PUT."""
# Construct paths based on provided category
webdav_base = self._get_webdav_base_path()
# Construct paths based on provided category. Encode via _webdav_path so
# categories/filenames with '#', commas or spaces don't truncate/404.
category_path_part = f"{category}/" if category else ""
attachment_dir_segment = f".attachments.{note_id}"
parent_dir_webdav_rel_path = (
f"Notes/{category_path_part}{attachment_dir_segment}"
)
parent_dir_path = f"{webdav_base}/{parent_dir_webdav_rel_path}"
attachment_path = f"{parent_dir_path}/{filename}"
parent_dir_path = self._webdav_path(parent_dir_webdav_rel_path)
attachment_path = self._webdav_path(f"{parent_dir_webdav_rel_path}/{filename}")
logger.debug("Uploading attachment '%s' for note %s", filename, note_id)
@@ -144,7 +175,7 @@ class WebDAVClient(BaseNextcloudClient):
headers = {"Content-Type": mime_type, "OCS-APIRequest": "true"}
try:
# First check if we can access WebDAV at all
notes_dir_path = f"{webdav_base}/Notes"
notes_dir_path = self._webdav_path("Notes")
propfind_headers = {"Depth": "0", "OCS-APIRequest": "true"}
notes_dir_response = await self._make_request(
"PROPFIND", notes_dir_path, headers=propfind_headers
@@ -209,10 +240,11 @@ class WebDAVClient(BaseNextcloudClient):
self, note_id: int, filename: str, category: Optional[str] = None
) -> Tuple[bytes, str]:
"""Fetch a specific attachment from a note via WebDAV GET."""
webdav_base = self._get_webdav_base_path()
category_path_part = f"{category}/" if category else ""
attachment_dir_segment = f".attachments.{note_id}"
attachment_path = f"{webdav_base}/Notes/{category_path_part}{attachment_dir_segment}/{filename}"
attachment_path = self._webdav_path(
f"Notes/{category_path_part}{attachment_dir_segment}/{filename}"
)
logger.debug("Fetching attachment '%s' for note %s", filename, note_id)
@@ -252,7 +284,7 @@ class WebDAVClient(BaseNextcloudClient):
async def list_directory(self, path: str = "") -> List[Dict[str, Any]]:
"""List files and directories in the specified path via WebDAV PROPFIND."""
webdav_path = f"{self._get_webdav_base_path()}/{path.lstrip('/')}"
webdav_path = self._webdav_path(path)
if not webdav_path.endswith("/"):
webdav_path += "/"
@@ -352,7 +384,7 @@ class WebDAVClient(BaseNextcloudClient):
async def read_file(self, path: str) -> Tuple[bytes, str]:
"""Read a file's content via WebDAV GET."""
webdav_path = f"{self._get_webdav_base_path()}/{path.lstrip('/')}"
webdav_path = self._webdav_path(path)
logger.debug("Reading file: %s", path)
@@ -379,7 +411,7 @@ class WebDAVClient(BaseNextcloudClient):
self, path: str, content: bytes, content_type: Optional[str] = None
) -> Dict[str, Any]:
"""Write content to a file via WebDAV PUT."""
webdav_path = f"{self._get_webdav_base_path()}/{path.lstrip('/')}"
webdav_path = self._webdav_path(path)
logger.debug("Writing file: %s", path)
@@ -410,7 +442,7 @@ class WebDAVClient(BaseNextcloudClient):
self, path: str, recursive: bool = False
) -> Dict[str, Any]:
"""Create a directory via WebDAV MKCOL."""
webdav_path = f"{self._get_webdav_base_path()}/{path.lstrip('/')}"
webdav_path = self._webdav_path(path)
if not webdav_path.endswith("/"):
webdav_path += "/"
@@ -468,10 +500,8 @@ class WebDAVClient(BaseNextcloudClient):
Returns:
Dict with status_code and optional message
"""
source_webdav_path = f"{self._get_webdav_base_path()}/{source_path.lstrip('/')}"
destination_webdav_path = (
f"{self._get_webdav_base_path()}/{destination_path.lstrip('/')}"
)
source_webdav_path = self._webdav_path(source_path)
destination_webdav_path = self._webdav_path(destination_path)
# Ensure paths have consistent trailing slashes for directories
if source_path.endswith("/") and not destination_path.endswith("/"):
@@ -552,10 +582,8 @@ class WebDAVClient(BaseNextcloudClient):
Returns:
Dict with status_code and optional message
"""
source_webdav_path = f"{self._get_webdav_base_path()}/{source_path.lstrip('/')}"
destination_webdav_path = (
f"{self._get_webdav_base_path()}/{destination_path.lstrip('/')}"
)
source_webdav_path = self._webdav_path(source_path)
destination_webdav_path = self._webdav_path(destination_path)
# Ensure paths have consistent trailing slashes for directories
if source_path.endswith("/") and not destination_path.endswith("/"):
@@ -1678,7 +1706,7 @@ class WebDAVClient(BaseNextcloudClient):
distinguish a definitive absence (HTTP 404) from a
brittle response (None).
"""
webdav_path = f"{self._get_webdav_base_path()}/{path.lstrip('/')}"
webdav_path = self._webdav_path(path)
propfind_body = """<?xml version="1.0"?>
<d:propfind xmlns:d="DAV:" xmlns:oc="http://owncloud.org/ns">
nextcloud_mcp_server/vector/_errors.py
+35
View File
@@ -0,0 +1,35 @@
"""Error-formatting helpers for the vector-sync pipeline.
Vector-sync work runs inside anyio task groups, so a failure in a child task
surfaces as a ``BaseExceptionGroup`` whose default ``str()`` is the useless
``"unhandled errors in a TaskGroup (N sub-exception)"`` -- it hides the real
``ConnectError`` / ``APIConnectionError`` that operators need to triage embed
drops (card 309). ``format_exception_group`` flattens the group to the leaf
exceptions so log lines name the actual cause; pair it with ``exc_info=True`` to
keep the full traceback.
"""
def format_exception_group(exc: BaseException) -> str:
"""Return a concise, leaf-naming string for ``exc``.
For a (possibly nested) ``BaseExceptionGroup`` this joins the ``repr`` of
each leaf exception; for an ordinary exception it returns its ``repr``. The
result is meant for the human-readable portion of a log message, not for
parsing.
"""
if not isinstance(exc, BaseExceptionGroup):
return repr(exc)
leaves = _flatten(exc)
noun = "sub-exception" if len(leaves) == 1 else "sub-exceptions"
return f"{len(leaves)} {noun}: " + "; ".join(repr(e) for e in leaves)
def _flatten(exc: BaseException) -> list[BaseException]:
"""Depth-first list of the leaf exceptions within ``exc``."""
if isinstance(exc, BaseExceptionGroup):
leaves: list[BaseException] = []
for sub in exc.exceptions:
leaves.extend(_flatten(sub))
return leaves
return [exc]
nextcloud_mcp_server/vector/oauth_sync.py
+14 -6
View File
@@ -30,6 +30,7 @@ from httpx import BasicAuth, HTTPStatusError
from nextcloud_mcp_server.auth.storage import RefreshTokenStorage
from nextcloud_mcp_server.client import NextcloudClient
from nextcloud_mcp_server.config import get_settings
from nextcloud_mcp_server.vector._errors import format_exception_group
from nextcloud_mcp_server.vector.processor import process_document
from nextcloud_mcp_server.vector.queue.ports import TaskProducer
from nextcloud_mcp_server.vector.scanner import DocumentTask, scan_user_documents
@@ -257,7 +258,7 @@ async def user_scanner_task(
logger.error(
"[BasicAuth] Scanner error for %s: %s (%s/%s)",
user_id,
e,
format_exception_group(e),
consecutive_errors,
max_consecutive_errors,
exc_info=True,
@@ -331,7 +332,7 @@ async def multi_user_processor_task(
break
except NotProvisionedError:
if doc_task:
if doc_task is not None:
logger.warning(
"[BasicAuth] User %s not provisioned, skipping %s_%s",
doc_task.user_id,
@@ -341,18 +342,21 @@ async def multi_user_processor_task(
continue
except Exception as e:
if doc_task:
if doc_task is not None:
logger.error(
"[BasicAuth] Processor %s error processing %s_%s: %s",
worker_id,
doc_task.doc_type,
doc_task.doc_id,
e,
format_exception_group(e),
exc_info=True,
)
else:
logger.error(
"[BasicAuth] Processor %s error: %s", worker_id, e, exc_info=True
"[BasicAuth] Processor %s error: %s",
worker_id,
format_exception_group(e),
exc_info=True,
)
finally:
@@ -481,7 +485,11 @@ async def user_manager_task(
logger.info("[BasicAuth] Stopped %s scanner(s)", len(revoked_users))
except Exception as e:
logger.error("[BasicAuth] User manager error: %s", e, exc_info=True)
logger.error(
"[BasicAuth] User manager error: %s",
format_exception_group(e),
exc_info=True,
)
# Sleep until next poll
try:
nextcloud_mcp_server/vector/processor.py
+24 -10
View File
@@ -31,6 +31,7 @@ from nextcloud_mcp_server.observability.tracing import trace_operation
from nextcloud_mcp_server.search.pdf_highlighter import PDFHighlighter
from nextcloud_mcp_server.usage import UsageEventStore
from nextcloud_mcp_server.vector import payload_keys
from nextcloud_mcp_server.vector._errors import format_exception_group
from nextcloud_mcp_server.vector.document_chunker import (
DocumentChunker,
PageAwareChunker,
@@ -235,6 +236,11 @@ async def processor_task(
# Signal that the task has started and is ready
task_status.started()
# Initialised before the loop so the broad except handler below can't hit an
# unbound name if receive() itself raises a non-TimeoutError/EndOfStream
# exception on the very first iteration (mirrors multi_user_processor_task).
doc_task: DocumentTask | None = None
while not shutdown_event.is_set():
try:
# Get document with timeout (allows checking shutdown)
@@ -264,14 +270,22 @@ async def processor_task(
break
except Exception as e:
logger.error(
"Processor %s error processing %s_%s: %s",
worker_id,
doc_task.doc_type,
doc_task.doc_id,
e,
exc_info=True,
)
if doc_task is not None:
logger.error(
"Processor %s error processing %s_%s: %s",
worker_id,
doc_task.doc_type,
doc_task.doc_id,
format_exception_group(e),
exc_info=True,
)
else:
logger.error(
"Processor %s error: %s",
worker_id,
format_exception_group(e),
exc_info=True,
)
# Continue to next document (no task_done() needed with streams)
logger.info("Processor %s stopped", worker_id)
@@ -443,7 +457,7 @@ async def process_document(
max_retries,
doc_task.doc_type,
doc_task.doc_id,
e,
format_exception_group(e),
extra={
"doc_id": doc_task.doc_id,
"doc_type": doc_task.doc_type,
@@ -460,7 +474,7 @@ async def process_document(
doc_task.doc_type,
doc_task.doc_id,
max_retries,
e,
format_exception_group(e),
extra={
"doc_id": doc_task.doc_id,
"doc_type": doc_task.doc_type,