Tools / Modules /_engines.py
alyxsis's picture
Web search: own DuckDuckGo/Bing/Yahoo/Wikipedia text engines, relevance check, timeout cooldown
82f841e verified
Raw History Blame Contribute Delete
10.8 kB
"""Text search engines scraped directly: DuckDuckGo (HTML), Bing, Yahoo, plus the Wikipedia API.
ddgs 9.x disables its Bing engine, and its Yahoo parser runs nested result blocks
together (one title made of every title below it). These are small, tested
replacements for text search only; news, images, videos and books still use ddgs.
Each engine returns a list of {"title", "href", "body"} dicts, or raises
NoResults (nothing on the page) or Blocked (a captcha or challenge page).
"""
from __future__ import annotations
import base64
import re
import time
from typing import Any, Callable
from urllib.parse import parse_qs, unquote, unquote_plus, urlparse
import primp
from lxml import html as lxml_html
class NoResults(Exception):
pass
class Blocked(Exception):
pass
def _client(timeout: float) -> primp.Client:
# A fresh client per request: a new random browser fingerprint each time.
return primp.Client(impersonate="random", impersonate_os="random", timeout=max(1.0, timeout), follow_redirects=True)
def _decode(content: bytes) -> str:
try:
return content.decode("utf-8")
except UnicodeDecodeError:
return content.decode("cp1252", errors="replace")
def _text(node: Any) -> str:
return " ".join(node.text_content().split()) if node is not None else ""
def _first(nodes: list[Any]) -> Any:
return nodes[0] if nodes else None
def _tree(content: bytes) -> Any:
return lxml_html.fromstring(_decode(content))
def _has_class(name: str) -> str:
return f"contains(concat(' ', normalize-space(@class), ' '), ' {name} ')"
# --- DuckDuckGo (html.duckduckgo.com) -------------------------------------------------
def _ddg_href(raw: str) -> str:
if raw.startswith("//"):
raw = "https:" + raw
parsed = urlparse(raw)
if parsed.netloc.endswith("duckduckgo.com") and parsed.path.startswith("/l/"):
target = parse_qs(parsed.query).get("uddg", [""])[0]
return unquote(target)
return raw
def duckduckgo(query: str, page: int, timelimit: str | None, deadline: float) -> list[dict[str, str]]:
payload = {"q": query, "b": "", "kl": "us-en"}
if page > 1:
payload["s"] = str(10 + (page - 2) * 15)
payload["dc"] = str(11 + (page - 2) * 15)
if timelimit:
payload["df"] = timelimit
# DuckDuckGo challenges some browser fingerprints (HTTP 202) and not others,
# so a challenge gets one retry with a fresh fingerprint while time allows.
for attempt in range(2):
remaining = deadline - time.monotonic()
if remaining < 0.5:
break
resp = _client(remaining).post("https://html.duckduckgo.com/html/", data=payload)
if resp.status_code == 200:
tree = _tree(resp.content)
results = []
for item in tree.xpath(f"//div[{_has_class('result')}]"):
if "result--ad" in (item.get("class") or ""):
continue
link = _first(item.xpath(f".//a[{_has_class('result__a')}]"))
if link is None:
continue
href = _ddg_href(link.get("href") or "")
if not href.startswith("http") or "duckduckgo.com/y.js" in href:
continue
snippet = _first(item.xpath(f".//*[{_has_class('result__snippet')}]"))
results.append({"title": _text(link), "href": href, "body": _text(snippet)})
if results:
return results
if b"anomaly" in resp.content or b"challenge" in resp.content:
continue
raise NoResults("duckduckgo returned no results")
if resp.status_code not in (202, 403, 429):
raise Blocked(f"duckduckgo HTTP {resp.status_code}")
raise Blocked("duckduckgo challenged the request")
# --- Bing -------------------------------------------------------------------------------
def _bing_href(raw: str) -> str:
parsed = urlparse(raw)
if parsed.netloc.endswith("bing.com") and parsed.path.startswith("/ck/"):
u = parse_qs(parsed.query).get("u", [""])[0]
if len(u) > 2:
b64 = u[2:]
try:
return base64.urlsafe_b64decode(b64 + "=" * (-len(b64) % 4)).decode()
except Exception:
return ""
return raw
_BING_TIME = {"d": 'ex1:"ez1"', "w": 'ex1:"ez2"', "m": 'ex1:"ez3"'}
def bing(query: str, page: int, timelimit: str | None, deadline: float) -> list[dict[str, str]]:
if page > 1:
# Bing ignores "first=" without its session cookies and repeats page one.
raise NoResults("bing serves only one page of results here")
params = {"q": query, "setlang": "en", "cc": "us", "mkt": "en-US"}
if timelimit == "y":
day = int(time.time() // 86400)
params["filters"] = f'ex1:"ez5_{day - 365}_{day}"'
elif timelimit in _BING_TIME:
params["filters"] = _BING_TIME[timelimit]
resp = _client(deadline - time.monotonic()).get("https://www.bing.com/search", params=params)
if resp.status_code != 200:
raise Blocked(f"bing HTTP {resp.status_code}")
tree = _tree(resp.content)
results = []
for item in tree.xpath(f"//li[{_has_class('b_algo')}]"):
title = _text(_first(item.xpath(".//h2")))
href = ""
# The title link wraps the h2 in some layouts and sits beside it in others;
# the site-icon link (a.tilk) always points at the page.
for candidate in item.xpath(".//h2//a/@href | .//h2/ancestor::a[1]/@href | .//a[contains(@class, 'tilk')]/@href"):
href = _bing_href(candidate)
if href.startswith("http") and "bing.com/" not in href.split("?")[0]:
break
href = ""
if not title or not href:
continue
body = _text(_first(item.xpath(f".//p | .//div[{_has_class('b_caption')}]")))
results.append({"title": title.rstrip("…").strip(), "href": href, "body": body})
if results:
return results
if b"captcha" in resp.content.lower() or b"challenge" in resp.content.lower():
raise Blocked("bing challenged the request")
raise NoResults("bing returned no results")
# --- Yahoo ------------------------------------------------------------------------------
def _yahoo_href(raw: str) -> str:
if "/RU=" in raw:
target = raw.split("/RU=", 1)[1]
return unquote_plus(re.split(r"/R[KS]=", target, maxsplit=1)[0])
return raw
def yahoo(query: str, page: int, timelimit: str | None, deadline: float) -> list[dict[str, str]]:
params = {"p": query, "ei": "UTF-8"}
if page > 1:
params["b"] = str((page - 1) * 7 + 1)
if timelimit:
params["btf"] = timelimit
resp = _client(deadline - time.monotonic()).get("https://search.yahoo.com/search", params=params)
if resp.status_code != 200:
raise Blocked(f"yahoo HTTP {resp.status_code}")
tree = _tree(resp.content)
results = []
# Yahoo's markup nests later results inside earlier ones, so take only the
# first title, link and text of each block.
for item in tree.xpath(f"//div[{_has_class('algo-sr')}]"):
title_box = _first(item.xpath(".//div[contains(@class, 'compTitle')]"))
if title_box is None:
continue
title = _text(_first(title_box.xpath(".//h3")))
href = _yahoo_href(_first(title_box.xpath(".//a/@href")) or "")
if not title or not href.startswith("http") or "bing.com/aclick" in href:
continue
body = _text(_first(item.xpath(".//div[contains(@class, 'compText')]")))
results.append({"title": title, "href": href, "body": body})
if results:
return results
raise NoResults("yahoo returned no results")
# --- Wikipedia (API) ---------------------------------------------------------------------
def wikipedia(query: str, page: int, timelimit: str | None, deadline: float) -> list[dict[str, str]]: # noqa: ARG001
params = {
"action": "query", "list": "search", "srsearch": query, "format": "json",
"srlimit": "10", "sroffset": str((page - 1) * 10), "utf8": "1",
}
resp = _client(deadline - time.monotonic()).get("https://en.wikipedia.org/w/api.php", params=params)
if resp.status_code != 200:
raise Blocked(f"wikipedia HTTP {resp.status_code}")
hits = (resp.json().get("query") or {}).get("search") or []
results = []
for hit in hits:
title = str(hit.get("title") or "")
snippet = re.sub(r"<[^>]+>", "", str(hit.get("snippet") or ""))
results.append({
"title": title,
"href": "https://en.wikipedia.org/wiki/" + title.replace(" ", "_"),
"body": " ".join(lxml_html.fromstring(f"<p>{snippet}</p>").text_content().split()) if snippet else "",
})
if results:
return results
raise NoResults("wikipedia returned no results")
TEXT_ENGINES: dict[str, Callable[[str, int, str | None, float], list[dict[str, str]]]] = {
"duckduckgo": duckduckgo,
"bing": bing,
"yahoo": yahoo,
"wikipedia": wikipedia,
}
# --- Relevance ----------------------------------------------------------------------------
_STOP = {
"a", "an", "the", "in", "on", "of", "for", "to", "and", "with", "at", "by", "from", "is", "are",
"what", "how", "why", "when", "where", "who", "which", "does", "do", "can", "near", "about", "best",
"vs", "or", "my", "your", "i", "you", "it", "be", "this", "that",
}
def query_terms(query: str) -> list[str]:
words = []
for word in query.lower().split():
if re.match(r"^-|^\w+:", word):
continue
word = re.sub(r"[^\w]", "", word)
if len(word) >= 3 and word not in _STOP:
words.append(word)
return words
def looks_relevant(query: str, results: list[dict[str, Any]], sample: int = 5) -> bool:
"""False when the first results do not look like answers to the query.
Throttled engines sometimes answer with a page of unrelated results (a
makeup search returning bank login pages, a wig search returning Best Buy).
A query word matches on its first five letters, so "walking" matches "walk".
A result counts when it mentions one query word (two, for queries of three
or more words), and at least two of the first results must count. Queries
with no usable words pass.
"""
stems = {t[:5] for t in query_terms(query)}
if not stems:
return True
need = 1 if len(stems) <= 2 else 2
top = results[:sample]
good = 0
for item in top:
hay = " ".join(str(item.get(k) or "") for k in ("title", "body", "href", "url")).lower()
if sum(1 for stem in stems if stem in hay) >= need:
good += 1
return good >= min(2, len(top))