phenny-1/modules/search.py

201 lines
6.2 KiB
Python
Raw Normal View History

#!/usr/bin/env python
"""
search.py - Phenny Web Search Module
2009-06-07 05:08:49 -04:00
Copyright 2008-9, Sean B. Palmer, inamidst.com
Licensed under the Eiffel Forum License 2.
http://inamidst.com/phenny/
"""
import re
import web
2011-06-17 11:49:37 -04:00
class Grab(web.urllib.URLopener):
def __init__(self, *args):
self.version = 'Mozilla/5.0 (Phenny)'
web.urllib.URLopener.__init__(self, *args)
self.addheader('Referer', 'https://github.com/sbp/phenny')
def http_error_default(self, url, fp, errcode, errmsg, headers):
return web.urllib.addinfourl(fp, [headers, errcode], "http:" + url)
def google_ajax(query):
2008-11-20 05:18:21 -05:00
"""Search using AjaxSearch, and return its JSON."""
uri = 'http://ajax.googleapis.com/ajax/services/search/web'
args = '?v=1.0&safe=off&q=' + web.urllib.quote(query)
2011-06-17 11:49:37 -04:00
handler = web.urllib._urlopener
web.urllib._urlopener = Grab()
2008-11-20 05:18:21 -05:00
bytes = web.get(uri + args)
2011-06-17 11:49:37 -04:00
web.urllib._urlopener = handler
return web.json(bytes)
def google_search(query):
results = google_ajax(query)
try: return results['responseData']['results'][0]['unescapedUrl']
except IndexError: return None
2011-06-17 11:49:37 -04:00
except TypeError:
print results
return False
def google_count(query):
results = google_ajax(query)
2008-11-20 05:25:03 -05:00
if not results.has_key('responseData'): return '0'
if not results['responseData'].has_key('cursor'): return '0'
if not results['responseData']['cursor'].has_key('estimatedResultCount'):
return '0'
2008-11-20 05:18:21 -05:00
return results['responseData']['cursor']['estimatedResultCount']
def formatnumber(n):
"""Format a number with beautiful commas."""
parts = list(str(n))
for i in range((len(parts) - 3), 0, -3):
parts.insert(i, ',')
return ''.join(parts)
def g(phenny, input):
2008-03-10 15:58:28 -04:00
"""Queries Google for the specified input."""
query = input.group(2)
if not query:
return phenny.reply('.g what?')
query = query.encode('utf-8')
uri = google_search(query)
if uri:
phenny.reply(uri)
if not hasattr(phenny.bot, 'last_seen_uri'):
phenny.bot.last_seen_uri = {}
phenny.bot.last_seen_uri[input.sender] = uri
2011-06-17 11:49:37 -04:00
elif uri is False: phenny.reply("Problem getting data from Google.")
else: phenny.reply("No results found for '%s'." % query)
g.commands = ['g']
g.priority = 'high'
2008-03-10 15:58:28 -04:00
g.example = '.g swhack'
def gc(phenny, input):
2008-03-10 15:58:28 -04:00
"""Returns the number of Google results for the specified input."""
query = input.group(2)
if not query:
return phenny.reply('.gc what?')
query = query.encode('utf-8')
num = formatnumber(google_count(query))
phenny.say(query + ': ' + num)
gc.commands = ['gc']
gc.priority = 'high'
2008-03-10 15:58:28 -04:00
gc.example = '.gc extrapolate'
r_query = re.compile(
r'\+?"[^"\\]*(?:\\.[^"\\]*)*"|\[[^]\\]*(?:\\.[^]\\]*)*\]|\S+'
)
def gcs(phenny, input):
if not input.group(2):
return phenny.reply("Nothing to compare.")
queries = r_query.findall(input.group(2))
if len(queries) > 6:
return phenny.reply('Sorry, can only compare up to six things.')
results = []
for i, query in enumerate(queries):
query = query.strip('[]')
query = query.encode('utf-8')
n = int((formatnumber(google_count(query)) or '0').replace(',', ''))
results.append((n, query))
if i >= 2: __import__('time').sleep(0.25)
if i >= 4: __import__('time').sleep(0.25)
results = [(term, n) for (n, term) in reversed(sorted(results))]
reply = ', '.join('%s (%s)' % (t, formatnumber(n)) for (t, n) in results)
phenny.say(reply)
gcs.commands = ['gcs', 'comp']
r_bing = re.compile(r'<h3><a href="([^"]+)"')
def bing_search(query, lang='en-GB'):
query = web.urllib.quote(query)
base = 'http://www.bing.com/search?mkt=%s&q=' % lang
bytes = web.get(base + query)
m = r_bing.search(bytes)
if m: return m.group(1)
def bing(phenny, input):
"""Queries Bing for the specified input."""
query = input.group(2)
if query.startswith(':'):
lang, query = query.split(' ', 1)
lang = lang[1:]
else: lang = 'en-GB'
if not query:
return phenny.reply('.bing what?')
query = query.encode('utf-8')
uri = bing_search(query, lang)
if uri:
phenny.reply(uri)
if not hasattr(phenny.bot, 'last_seen_uri'):
phenny.bot.last_seen_uri = {}
phenny.bot.last_seen_uri[input.sender] = uri
else: phenny.reply("No results found for '%s'." % query)
bing.commands = ['bing']
bing.example = '.bing swhack'
r_duck = re.compile(r'nofollow" class="[^"]+" href="(http.*?)">')
def duck_search(query):
query = query.replace('!', '')
query = web.urllib.quote(query)
uri = 'http://duckduckgo.com/html/?q=%s&kl=uk-en' % query
bytes = web.get(uri)
m = r_duck.search(bytes)
if m: return web.decode(m.group(1))
def duck(phenny, input):
query = input.group(2)
if not query: return phenny.reply('.ddg what?')
query = query.encode('utf-8')
uri = duck_search(query)
if uri:
phenny.reply(uri)
if not hasattr(phenny.bot, 'last_seen_uri'):
phenny.bot.last_seen_uri = {}
phenny.bot.last_seen_uri[input.sender] = uri
else: phenny.reply("No results found for '%s'." % query)
duck.commands = ['duck', 'ddg']
def search(phenny, input):
if not input.group(2):
return phenny.reply('.search for what?')
query = input.group(2).encode('utf-8')
gu = google_search(query) or '-'
bu = bing_search(query) or '-'
du = duck_search(query) or '-'
if (gu == bu) and (bu == du):
result = '%s (g, b, d)' % gu
elif (gu == bu):
result = '%s (g, b), %s (d)' % (gu, du)
elif (bu == du):
result = '%s (b, d), %s (g)' % (bu, gu)
elif (gu == du):
result = '%s (g, d), %s (b)' % (gu, bu)
else:
if len(gu) > 250: gu = '(extremely long link)'
if len(bu) > 150: bu = '(extremely long link)'
if len(du) > 150: du = '(extremely long link)'
result = '%s (g), %s (b), %s (d)' % (gu, bu, du)
phenny.reply(result)
search.commands = ['search']
def suggest(phenny, input):
if not input.group(2):
return phenny.reply("No query term.")
query = input.group(2).encode('utf-8')
uri = 'http://websitedev.de/temp-bin/suggest.pl?q='
answer = web.get(uri + web.urllib.quote(query).replace('+', '%2B'))
if answer:
phenny.say(answer)
else: phenny.reply('Sorry, no result.')
suggest.commands = ['suggest']
if __name__ == '__main__':
print __doc__.strip()