from bs4 import BeautifulSoup from app.services.parsers.base_parser import BaseParser, ProductResult import logging logger = logging.getLogger(__name__) class LIncroyableParser(BaseParser): def __init__(self): super().__init__("lincroyable.fr", "https://www.lincroyable.fr") def parse_search_results(self, html: str, query: str, search_url: str) -> list[ProductResult]: soup = BeautifulSoup(html, "html.parser") results = [] # L'Incroyable products # Config: div.tailleBlocProdNew cards = soup.select("div.tailleBlocProdNew") for card in cards: try: # Link is usually in an 'a' tag inside, or the card itself might be clickable (but here we see multiple links) # We look for the main link to the product link_el = card.select_one("a[href*='/p']") if not link_el: continue href = link_el.get('href') url = self.make_absolute_url(href) # Title title_el = card.select_one("h3.nomCoupDeCoeurNew, .nomCoupDeCoeurNew") title = title_el.get_text(strip=True) if title_el else link_el.get_text(strip=True) if not title: continue # Image img_url = None img_el = card.select_one("img.imgCoup2coeur, img") if img_el: img_url = self._get_image_src(img_el) if img_url: img_url = self.make_absolute_url(img_url) # Price price = None price_el = card.select_one("p.prixCoupDeCoeurNew, .prixCoupDeCoeurNew") if price_el: price = self.parse_price_text(price_el.get_text()) # Fallback price search in text if not price: text = card.get_text(" ", strip=True) price = self.parse_price_text(text) results.append(ProductResult( title=title, url=url, source="L'Incroyable", price=price, currency="EUR", in_stock=True, image_url=img_url, snippet=f"Product from L'Incroyable" )) except Exception as e: logger.error(f"Error parsing L'Incroyable product: {e}") continue logger.info(f"LIncroyableParser found {len(results)} results") return results