dosage/scripts/creators.py
2015-11-04 23:43:31 +01:00

85 lines
2.7 KiB
Python
Executable file

#!/usr/bin/env python
# Copyright (C) 2012-2014 Bastian Kleineidam
"""
Script to get a list of creators.com comics and save the info in a JSON file for further processing.
"""
from __future__ import print_function
import re
import codecs
import sys
import os
import requests
sys.path.append(os.path.join(os.path.dirname(__file__), ".."))
from dosagelib.util import getPageContent, asciify, unescape, tagre
from dosagelib.scraper import get_scraperclasses
from scriptutil import contains_case_insensitive, capfirst, save_result, load_result, truncate_name
json_file = __file__.replace(".py", ".json")
url_matcher = re.compile(tagre("a", "href", r'/comics/([^/]+)\.html') + r'<strong>([^<]+)</strong>')
# names of comics to exclude
exclude_comics = [
]
def handle_url(url, session, res):
"""Parse one search result page."""
print("Parsing", url, file=sys.stderr)
try:
data = getPageContent(url, session)
except IOError as msg:
print("ERROR:", msg, file=sys.stderr)
return
for match in url_matcher.finditer(data):
url = match.group(1)
name = unescape(match.group(2))
name = asciify(name.replace('&', 'And').replace('@', 'At'))
name = capfirst(name)
if name in exclude_comics:
continue
if contains_case_insensitive(res, name):
# we cannot handle two comics that only differ in case
print("INFO: skipping possible duplicate", repr(name), file=sys.stderr)
continue
res[name] = url
def get_results():
"""Parse all search result pages."""
# store info in a dictionary {name -> shortname}
res = {}
session = requests.Session()
handle_url('http://www.creators.com/comics/cat-seeall.html', session, res)
save_result(res, json_file)
def has_gocomics_comic(name):
"""Test if comic name already exists."""
cname = "Gocomics/%s" % name
for scraperclass in get_scraperclasses():
lname = scraperclass.getName().lower()
if lname == cname.lower():
return True
return False
def print_results(args):
"""Print comics."""
min_comics, filename = args
with codecs.open(filename, 'a', 'utf-8') as fp:
for name, path in sorted(load_result(json_file).items()):
if name in exclude_comics:
continue
lang = 'Es' if name.lower().endswith('spanish') else ''
if has_gocomics_comic(name):
fp.write(u'# %s has a duplicate in gocomics\n' % truncate_name(name))
else:
fp.write(u"class %s(_Creators%s):\n path = %r\n\n" %
(truncate_name(name), lang, path))
if __name__ == '__main__':
if len(sys.argv) > 1:
print_results(sys.argv[1:])
else:
get_results()