searxngRebrandZaclys/searx/webutils.py

149 lines
4.4 KiB
Python
Raw Normal View History

# -*- coding: utf-8 -*-
import os
import csv
import hashlib
import hmac
import re
import inspect
from io import StringIO
from codecs import getincrementalencoder
from searx import logger
VALID_LANGUAGE_CODE = re.compile(r'^[a-z]{2,3}(-[a-zA-Z]{2})?$')
logger = logger.getChild('webutils')
class UnicodeWriter:
"""
A CSV writer which will write rows to CSV file "f",
which is encoded in the given encoding.
"""
def __init__(self, f, dialect=csv.excel, encoding="utf-8", **kwds):
# Redirect output to a queue
self.queue = StringIO()
self.writer = csv.writer(self.queue, dialect=dialect, **kwds)
self.stream = f
self.encoder = getincrementalencoder(encoding)()
def writerow(self, row):
self.writer.writerow(row)
# Fetch UTF-8 output from the queue ...
data = self.queue.getvalue()
data = data.strip('\x00')
# ... and reencode it into the target encoding
data = self.encoder.encode(data)
# write to the target stream
self.stream.write(data.decode())
# empty queue
self.queue.truncate(0)
def writerows(self, rows):
for row in rows:
self.writerow(row)
def get_resources_directory(searx_directory, subdirectory, resources_directory):
if not resources_directory:
resources_directory = os.path.join(searx_directory, subdirectory)
if not os.path.isdir(resources_directory):
raise Exception(resources_directory + " is not a directory")
return resources_directory
def get_themes(templates_path):
"""Returns available themes list."""
themes = os.listdir(templates_path)
if '__common__' in themes:
themes.remove('__common__')
return themes
def get_static_files(static_path):
static_files = set()
static_path_length = len(static_path) + 1
for directory, _, files in os.walk(static_path):
for filename in files:
f = os.path.join(directory[static_path_length:], filename)
static_files.add(f)
return static_files
def get_result_templates(templates_path):
result_templates = set()
templates_path_length = len(templates_path) + 1
for directory, _, files in os.walk(templates_path):
if directory.endswith('result_templates'):
for filename in files:
f = os.path.join(directory[templates_path_length:], filename)
result_templates.add(f)
return result_templates
def new_hmac(secret_key, url):
try:
secret_key_bytes = bytes(secret_key, 'utf-8')
except TypeError as err:
if isinstance(secret_key, bytes):
secret_key_bytes = secret_key
else:
raise err
return hmac.new(secret_key_bytes, url, hashlib.sha256).hexdigest()
def prettify_url(url, max_length=74):
if len(url) > max_length:
chunk_len = int(max_length / 2 + 1)
return '{0}[...]{1}'.format(url[:chunk_len], url[-chunk_len:])
else:
return url
def highlight_content(content, query):
if not content:
return None
# ignoring html contents
# TODO better html content detection
if content.find('<') != -1:
return content
if content.lower().find(query.lower()) > -1:
query_regex = '({0})'.format(re.escape(query))
content = re.sub(query_regex, '<span class="highlight">\\1</span>',
content, flags=re.I | re.U)
else:
regex_parts = []
for chunk in query.split():
chunk = chunk.replace('"', '')
if len(chunk) == 0:
continue
elif len(chunk) == 1:
regex_parts.append('\\W+{0}\\W+'.format(re.escape(chunk)))
else:
regex_parts.append('{0}'.format(re.escape(chunk)))
query_regex = '({0})'.format('|'.join(regex_parts))
content = re.sub(query_regex, '<span class="highlight">\\1</span>',
content, flags=re.I | re.U)
return content
def is_flask_run_cmdline():
"""Check if the application was started using "flask run" command line
Inspect the callstack.
See https://github.com/pallets/flask/blob/master/src/flask/__main__.py
Returns:
bool: True if the application was started using "flask run".
"""
frames = inspect.stack()
if len(frames) < 2:
return False
return frames[-2].filename.endswith('flask/cli.py')