| 12345678910111213141516171819202122232425262728293031323334353637383940414243444546474849505152535455565758596061626364656667686970717273747576777879808182838485868788899091929394959697 | # SPDX-License-Identifier: AGPL-3.0-or-later"""Geizhals is a German website to compare the price of a product on themost common German shopping sites and find the lowest price.The sorting of the search results can be influenced by the following additionsto the search term:``asc`` or ``price``  To sort by price in ascending order.``desc``  To sort by price in descending order."""import refrom urllib.parse import urlencodefrom lxml import htmlfrom searx.utils import eval_xpath, eval_xpath_list, extract_textabout = {    'website': 'https://geizhals.de',    'wikidata_id': 'Q15977657',    'use_official_api': False,    'official_api_documentation': None,    'require_api_key': False,    'results': 'HTML',    'language': 'de',}paging = Truecategories = ['shopping']base_url = "https://geizhals.de"sort_order = 'relevance'SORT_RE = re.compile(r"sort:(\w+)")sort_order_map = {    'relevance': None,    'price': 'p',    'asc': 'p',    'desc': '-p',}def request(query, params):    sort = None    sort_order_path = SORT_RE.search(query)    if sort_order_path:        sort = sort_order_map.get(sort_order_path.group(1))        query = SORT_RE.sub("", query)        logger.debug(query)    args = {        'fs': query,        'pg': params['pageno'],        'toggle_all': 1,  # load item specs        'sort': sort,    }    params['url'] = f"{base_url}/?{urlencode(args)}"    return paramsdef response(resp):    results = []    dom = html.fromstring(resp.text)    for result in eval_xpath_list(dom, "//article[contains(@class, 'listview__item')]"):        content = []        for spec in eval_xpath_list(result, ".//div[contains(@class, 'specs-grid__item')]"):            content.append(f"{extract_text(eval_xpath(spec, './dt'))}: {extract_text(eval_xpath(spec, './dd'))}")        metadata = [            extract_text(eval_xpath(result, ".//div[contains(@class, 'stars-rating-label')]")),            extract_text(eval_xpath(result, ".//div[contains(@class, 'listview__offercount')]")),        ]        item = {            'template': 'products.html',            'url': (                base_url + "/" + extract_text(eval_xpath(result, ".//a[contains(@class, 'listview__name-link')]/@href"))            ),            'title': extract_text(eval_xpath(result, ".//h3[contains(@class, 'listview__name')]")),            'content': ' | '.join(content),            'thumbnail': extract_text(eval_xpath(result, ".//img[contains(@class, 'listview__image')]/@src")),            'metadata': ', '.join(item for item in metadata if item),        }        best_price = extract_text(eval_xpath(result, ".//a[contains(@class, 'listview__price-link')]")).split(" ")        if len(best_price) > 1:            item["price"] = f"Bestes Angebot: {best_price[1]}€"        results.append(item)    return results
 |