Files
Priceflow/app/services/parsers/action_parser.py
T

75 lines
2.5 KiB
Python

from bs4 import BeautifulSoup
from app.services.parsers.base_parser import BaseParser, ProductResult
import logging
import re
logger = logging.getLogger(__name__)
class ActionParser(BaseParser):
def __init__(self):
super().__init__("action.com", "https://www.action.com")
def parse_search_results(self, html: str, query: str, search_url: str) -> list[ProductResult]:
soup = BeautifulSoup(html, "html.parser")
results = []
# Action products
# Config: a.group[href^='/fr-fr/p/']
links = soup.select("a[href^='/fr-fr/p/']")
seen_urls = set()
for link in links:
try:
href = link.get("href")
if not href or href in seen_urls:
continue
seen_urls.add(href)
url = self.make_absolute_url(href)
# Action usually has the card content inside the link
title = None
title_el = link.select_one("[class*='title'], h3, h4")
if title_el:
title = title_el.get_text(strip=True)
else:
# Fallback to specific spans or remove price info from text
# Often Action cards have a .title span
title = link.get_text(" ", strip=True)
# Remove price-like patterns from title if we had to get whole link text
title = re.sub(r"\d+,\d+\s*€.*$", "", title)
title = re.sub(r"\d+\s*€.*$", "", title)
if not title:
continue
img_url = self.extract_image_url(link)
price = None
price_el = link.select_one("[class*='price']")
if price_el:
price = self.parse_price_text(price_el.get_text())
results.append(
ProductResult(
title=title,
url=url,
source="Action",
price=price,
currency="EUR",
in_stock=True,
image_url=img_url,
snippet=f"Product from Action",
)
)
except Exception as e:
logger.error(f"Error parsing Action product: {e}")
continue
logger.info(f"ActionParser found {len(results)} results")
return results