mirror of
https://github.com/hwchase17/langchain
synced 2024-11-10 01:10:59 +00:00
d3db83abe3
* Lint for usage of standard xml library * Add forced opt-in for quip client * Actual security issue is with underlying QuipClient not LangChain integration (since the client is doing the parsing), but adding enforcement at the LangChain level.
250 lines
9.0 KiB
Python
250 lines
9.0 KiB
Python
import logging
|
|
import re
|
|
import xml.etree.cElementTree # OK: user-must-opt-in
|
|
from io import BytesIO
|
|
from typing import List, Optional, Sequence
|
|
from xml.etree.ElementTree import ElementTree # OK: user-must-opt-in
|
|
|
|
from langchain_core.documents import Document
|
|
|
|
from langchain_community.document_loaders.base import BaseLoader
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
_MAXIMUM_TITLE_LENGTH = 64
|
|
|
|
|
|
class QuipLoader(BaseLoader):
|
|
"""Load `Quip` pages.
|
|
|
|
Port of https://github.com/quip/quip-api/tree/master/samples/baqup
|
|
"""
|
|
|
|
def __init__(
|
|
self,
|
|
api_url: str,
|
|
access_token: str,
|
|
request_timeout: Optional[int] = 60,
|
|
*,
|
|
allow_dangerous_xml_parsing: bool = False,
|
|
):
|
|
"""
|
|
Args:
|
|
api_url: https://platform.quip.com
|
|
access_token: token of access quip API. Please refer:
|
|
https://quip.com/dev/automation/documentation/current#section/Authentication/Get-Access-to-Quip's-APIs
|
|
request_timeout: timeout of request, default 60s.
|
|
allow_dangerous_xml_parsing: Allow dangerous XML parsing, defaults to False
|
|
"""
|
|
try:
|
|
from quip_api.quip import QuipClient
|
|
except ImportError:
|
|
raise ImportError(
|
|
"`quip_api` package not found, please run " "`pip install quip_api`"
|
|
)
|
|
|
|
self.quip_client = QuipClient(
|
|
access_token=access_token, base_url=api_url, request_timeout=request_timeout
|
|
)
|
|
|
|
if not allow_dangerous_xml_parsing:
|
|
raise ValueError(
|
|
"The quip client uses the built-in XML parser which may cause"
|
|
"security issues when parsing XML data in some cases. "
|
|
"Please see "
|
|
"https://docs.python.org/3/library/xml.html#xml-vulnerabilities "
|
|
"For more information, set `allow_dangerous_xml_parsing` as True "
|
|
"if you are sure that your distribution of the standard library "
|
|
"is not vulnerable to XML vulnerabilities."
|
|
)
|
|
|
|
def load(
|
|
self,
|
|
folder_ids: Optional[List[str]] = None,
|
|
thread_ids: Optional[List[str]] = None,
|
|
max_docs: Optional[int] = 1000,
|
|
include_all_folders: bool = False,
|
|
include_comments: bool = False,
|
|
include_images: bool = False,
|
|
) -> List[Document]:
|
|
"""
|
|
Args:
|
|
:param folder_ids: List of specific folder IDs to load, defaults to None
|
|
:param thread_ids: List of specific thread IDs to load, defaults to None
|
|
:param max_docs: Maximum number of docs to retrieve in total, defaults 1000
|
|
:param include_all_folders: Include all folders that your access_token
|
|
can access, but doesn't include your private folder
|
|
:param include_comments: Include comments, defaults to False
|
|
:param include_images: Include images, defaults to False
|
|
"""
|
|
if not folder_ids and not thread_ids and not include_all_folders:
|
|
raise ValueError(
|
|
"Must specify at least one among `folder_ids`, `thread_ids` "
|
|
"or set `include_all`_folders as True"
|
|
)
|
|
|
|
thread_ids = thread_ids or []
|
|
|
|
if folder_ids:
|
|
for folder_id in folder_ids:
|
|
self.get_thread_ids_by_folder_id(folder_id, 0, thread_ids)
|
|
|
|
if include_all_folders:
|
|
user = self.quip_client.get_authenticated_user()
|
|
if "group_folder_ids" in user:
|
|
self.get_thread_ids_by_folder_id(
|
|
user["group_folder_ids"], 0, thread_ids
|
|
)
|
|
if "shared_folder_ids" in user:
|
|
self.get_thread_ids_by_folder_id(
|
|
user["shared_folder_ids"], 0, thread_ids
|
|
)
|
|
|
|
thread_ids = list(set(thread_ids[:max_docs]))
|
|
return self.process_threads(thread_ids, include_images, include_comments)
|
|
|
|
def get_thread_ids_by_folder_id(
|
|
self, folder_id: str, depth: int, thread_ids: List[str]
|
|
) -> None:
|
|
"""Get thread ids by folder id and update in thread_ids"""
|
|
from quip_api.quip import HTTPError, QuipError
|
|
|
|
try:
|
|
folder = self.quip_client.get_folder(folder_id)
|
|
except QuipError as e:
|
|
if e.code == 403:
|
|
logging.warning(
|
|
f"depth {depth}, Skipped over restricted folder {folder_id}, {e}"
|
|
)
|
|
else:
|
|
logging.warning(
|
|
f"depth {depth}, Skipped over folder {folder_id} "
|
|
f"due to unknown error {e.code}"
|
|
)
|
|
return
|
|
except HTTPError as e:
|
|
logging.warning(
|
|
f"depth {depth}, Skipped over folder {folder_id} "
|
|
f"due to HTTP error {e.code}"
|
|
)
|
|
return
|
|
|
|
title = folder["folder"].get("title", "Folder %s" % folder_id)
|
|
|
|
logging.info(f"depth {depth}, Processing folder {title}")
|
|
for child in folder["children"]:
|
|
if "folder_id" in child:
|
|
self.get_thread_ids_by_folder_id(
|
|
child["folder_id"], depth + 1, thread_ids
|
|
)
|
|
elif "thread_id" in child:
|
|
thread_ids.append(child["thread_id"])
|
|
|
|
def process_threads(
|
|
self, thread_ids: Sequence[str], include_images: bool, include_messages: bool
|
|
) -> List[Document]:
|
|
"""Process a list of thread into a list of documents."""
|
|
docs = []
|
|
for thread_id in thread_ids:
|
|
doc = self.process_thread(thread_id, include_images, include_messages)
|
|
if doc is not None:
|
|
docs.append(doc)
|
|
return docs
|
|
|
|
def process_thread(
|
|
self, thread_id: str, include_images: bool, include_messages: bool
|
|
) -> Optional[Document]:
|
|
thread = self.quip_client.get_thread(thread_id)
|
|
thread_id = thread["thread"]["id"]
|
|
title = thread["thread"]["title"]
|
|
link = thread["thread"]["link"]
|
|
update_ts = thread["thread"]["updated_usec"]
|
|
sanitized_title = QuipLoader._sanitize_title(title)
|
|
|
|
logger.info(
|
|
f"processing thread {thread_id} title {sanitized_title} "
|
|
f"link {link} update_ts {update_ts}"
|
|
)
|
|
|
|
if "html" in thread:
|
|
# Parse the document
|
|
try:
|
|
tree = self.quip_client.parse_document_html(thread["html"])
|
|
except xml.etree.cElementTree.ParseError as e:
|
|
logger.error(f"Error parsing thread {title} {thread_id}, skipping, {e}")
|
|
return None
|
|
|
|
metadata = {
|
|
"title": sanitized_title,
|
|
"update_ts": update_ts,
|
|
"id": thread_id,
|
|
"source": link,
|
|
}
|
|
|
|
# Download each image and replace with the new URL
|
|
text = ""
|
|
if include_images:
|
|
text = self.process_thread_images(tree)
|
|
|
|
if include_messages:
|
|
text = text + "/n" + self.process_thread_messages(thread_id)
|
|
|
|
return Document(
|
|
page_content=thread["html"] + text,
|
|
metadata=metadata,
|
|
)
|
|
return None
|
|
|
|
def process_thread_images(self, tree: ElementTree) -> str:
|
|
text = ""
|
|
|
|
try:
|
|
from PIL import Image
|
|
from pytesseract import pytesseract
|
|
except ImportError:
|
|
raise ImportError(
|
|
"`Pillow or pytesseract` package not found, "
|
|
"please run "
|
|
"`pip install Pillow` or `pip install pytesseract`"
|
|
)
|
|
|
|
for img in tree.iter("img"):
|
|
src = img.get("src")
|
|
if not src or not src.startswith("/blob"):
|
|
continue
|
|
_, _, thread_id, blob_id = src.split("/")
|
|
blob_response = self.quip_client.get_blob(thread_id, blob_id)
|
|
try:
|
|
image = Image.open(BytesIO(blob_response.read()))
|
|
text = text + "\n" + pytesseract.image_to_string(image)
|
|
except OSError as e:
|
|
logger.error(f"failed to convert image to text, {e}")
|
|
raise e
|
|
return text
|
|
|
|
def process_thread_messages(self, thread_id: str) -> str:
|
|
max_created_usec = None
|
|
messages = []
|
|
while True:
|
|
chunk = self.quip_client.get_messages(
|
|
thread_id, max_created_usec=max_created_usec, count=100
|
|
)
|
|
messages.extend(chunk)
|
|
if chunk:
|
|
max_created_usec = chunk[-1]["created_usec"] - 1
|
|
else:
|
|
break
|
|
messages.reverse()
|
|
|
|
texts = [message["text"] for message in messages]
|
|
|
|
return "\n".join(texts)
|
|
|
|
@staticmethod
|
|
def _sanitize_title(title: str) -> str:
|
|
sanitized_title = re.sub(r"\s", " ", title)
|
|
sanitized_title = re.sub(r"(?u)[^- \w.]", "", sanitized_title)
|
|
if len(sanitized_title) > _MAXIMUM_TITLE_LENGTH:
|
|
sanitized_title = sanitized_title[:_MAXIMUM_TITLE_LENGTH]
|
|
return sanitized_title
|