app.py

import os
os.system("playwright install")
import os
import re
import urllib.parse
import asyncio
from typing import Dict

import gradio as gr
from bs4 import BeautifulSoup, NavigableString
from playwright.async_api import async_playwright

# --- 1. GLOBAL RESOURCES & CONFIGURATION ---

# This dictionary will hold the long-lived Playwright and Browser objects.
# It starts empty and browsers are added on-demand.
PLAYWRIGHT_STATE: Dict = {}

# A comprehensive list of search engines
SEARCH_ENGINES = {
    "DuckDuckGo": "https://duckduckgo.com/html/?q={query}", "Google": "https://www.google.com/search?q={query}",
    "Bing": "https://www.bing.com/search?q={query}", "Brave": "https://search.brave.com/search?q={query}",
    "Ecosia": "https://www.ecosia.org/search?q={query}", "Yahoo": "https://search.yahoo.com/search?p={query}",
    "Startpage": "https://www.startpage.com/sp/search?q={query}", "Qwant": "https://www.qwant.com/?q={query}",
    "Swisscows": "https://swisscows.com/web?query={query}", "You.com": "https://you.com/search?q={query}",
    "SearXNG": "https://searx.be/search?q={query}", "MetaGer": "https://metager.org/meta/meta.ger-en?eingabe={query}",
    "Yandex": "https://yandex.com/search/?text={query}", "Baidu": "https://www.baidu.com/s?wd={query}",
    "Perplexity": "https://www.perplexity.ai/search?q={query}"
}

# --- 2. ADVANCED HTML-TO-MARKDOWN CONVERTER (Unchanged) ---
class HTML_TO_MARKDOWN_CONVERTER:
    # ... [The class code is identical and correct] ...
    def __init__(self, soup: BeautifulSoup, base_url: str):
        self.soup = soup
        self.base_url = base_url

    def _cleanup_html(self):
        selectors_to_remove = [
            'nav', 'footer', 'header', 'aside', 'form', 'script', 'style', 'svg', 'button', 'input', 'textarea',
            '[role="navigation"]', '[role="search"]', '[id*="comment"]', '[class*="comment-"]',
            '[id*="sidebar"]', '[class*="sidebar"]', '[id*="related"]', '[class*="related"]',
            '[id*="share"]', '[class*="share"]', '[id*="social"]', '[class*="social"]',
            '[id*="cookie"]', '[class*="cookie"]'
        ]
        for selector in selectors_to_remove:
            for element in self.soup.select(selector):
                element.decompose()

    def convert(self):
        self._cleanup_html()
        content_node = self.soup.find('main') or self.soup.find('article') or self.soup.find('body')
        if not content_node:
            return "Could not find main content."
        md = self._process_node(content_node)
        return re.sub(r'\n{3,}', '\n\n', md).strip()

    def _process_node(self, element):
        if isinstance(element, NavigableString): return re.sub(r'\s+', ' ', element.strip())
        if element.name is None or not element.name: return ''
        inner_md = " ".join(self._process_node(child) for child in element.children).strip()
        if element.name in ['p', 'div', 'section']: return f"\n\n{inner_md}\n\n"
        if element.name == 'h1': return f"\n\n# {inner_md}\n\n"
        if element.name == 'h2': return f"\n\n## {inner_md}\n\n"
        if element.name == 'h3': return f"\n\n### {inner_md}\n\n"
        if element.name in ['h4', 'h5', 'h6']: return f"\n\n#### {inner_md}\n\n"
        if element.name == 'li': return f"* {inner_md}\n"
        if element.name in ['ul', 'ol']: return f"\n{inner_md}\n"
        if element.name == 'blockquote': return f"> {inner_md.replace(chr(10), chr(10) + '> ')}\n\n"
        if element.name == 'hr': return "\n\n---\n\n"
        if element.name == 'table':
            header = " | ".join(f"**{th.get_text(strip=True)}**" for th in element.select('thead th, tr th'))
            separator = " | ".join(['---'] * len(header.split('|')))
            rows = [" | ".join(td.get_text(strip=True) for td in tr.find_all('td')) for tr in element.select('tbody tr')]
            return f"\n\n{header}\n{separator}\n" + "\n".join(rows) + "\n\n"
        if element.name == 'pre': return f"\n```\n{element.get_text(strip=True)}\n```\n\n"
        if element.name == 'code': return f"`{inner_md}`"
        if element.name in ['strong', 'b']: return f"**{inner_md}**"
        if element.name in ['em', 'i']: return f"*{inner_md}*"
        if element.name == 'a':
            href = element.get('href', '')
            full_href = urllib.parse.urljoin(self.base_url, href)
            return f"[{inner_md}]({full_href})"
        if element.name == 'img':
            src = element.get('src', '')
            alt = element.get('alt', 'Image').strip()
            full_src = urllib.parse.urljoin(self.base_url, src)
            return f"\n\n![{alt}]({full_src})\n\n"
        return inner_md


# --- 3. CORE API FUNCTION (WITH LAZY LOADING) ---

async def perform_web_browse(query: str, browser_name: str, search_engine: str):
    """
    A stateless function that takes a query, browser, and search engine,
    then returns the parsed content of the resulting page.
    It launches and caches browsers on-demand.
    """
    # Step 1: Initialize Playwright process itself if not already running.
    if "playwright" not in PLAYWRIGHT_STATE:
        print("🚀 First request received, starting Playwright process...")
        PLAYWRIGHT_STATE["playwright"] = await async_playwright().start()
        print("✅ Playwright process is running.")

    # Step 2: Check if the *specific browser requested* has been launched.
    browser_key = browser_name.lower()
    if browser_key not in PLAYWRIGHT_STATE:
        print(f"🚀 Launching '{browser_key}' for the first time...")
        try:
            p = PLAYWRIGHT_STATE["playwright"]
            if browser_key == 'firefox':
                browser_instance = await p.firefox.launch(headless=True)
            elif browser_key == 'chromium':
                browser_instance = await p.chromium.launch(headless=True)
            elif browser_key == 'webkit':
                browser_instance = await p.webkit.launch(headless=True)
            else:
                raise ValueError(f"Invalid browser name: {browser_name}")
            PLAYWRIGHT_STATE[browser_key] = browser_instance
            print(f"✅ '{browser_key}' is now running and cached.")
        except Exception as e:
            error_message = str(e).splitlines()[0]
            print(f"❌ Failed to launch '{browser_key}': {error_message}")
            return {"status": "error", "query": query, "error_message": f"Failed to launch browser '{browser_key}'. Your system might be missing dependencies. Error: {error_message}"}
    
    browser_instance = PLAYWRIGHT_STATE[browser_key]

    # Step 3: Determine URL
    is_url = urllib.parse.urlparse(query).scheme in ['http', 'https']
    if is_url:
        url = query
    else:
        search_url_template = SEARCH_ENGINES.get(search_engine)
        if not search_url_template:
            return {"error": f"Invalid search engine: '{search_engine}'."}
        url = search_url_template.format(query=urllib.parse.quote_plus(query))

    # Step 4: Create isolated context and browse
    context = await browser_instance.new_context(user_agent='Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36')
    page = await context.new_page()

    try:
        print(f"Navigating to: {url} using {browser_name}...")
        await page.goto(url, wait_until='domcontentloaded', timeout=30000)
        final_url, title = page.url, await page.title() or "No Title"
        print(f"Arrived at: {final_url}")
        
        soup = BeautifulSoup(await page.content(), 'lxml')
        converter = HTML_TO_MARKDOWN_CONVERTER(soup, base_url=final_url)
        markdown_text = converter.convert()
        
        print("Content parsed successfully.")
        return {"status": "success", "query": query, "final_url": final_url, "page_title": title, "markdown_content": markdown_text}
    except Exception as e:
        error_message = str(e).splitlines()[0]
        print(f"An error occurred: {error_message}")
        return {"status": "error", "query": query, "error_message": error_message}
    finally:
        if page: await page.close()
        if context: await context.close()
        print("Session context closed.")


# --- 4. GRADIO INTERFACE & API LAUNCH (Unchanged) ---
with gr.Blocks(title="Web Browse API", theme=gr.themes.Soft()) as demo:
    # ... UI definition is identical ...
    gr.Markdown("# Web Browse API")
    gr.Markdown("This interface exposes a stateless API endpoint (`/api/web_browse`) to fetch and parse web content.")
    query_input = gr.Textbox(label="URL or Search Query", placeholder="e.g., https://openai.com or 'history of artificial intelligence'")
    with gr.Row():
        browser_input = gr.Dropdown(label="Browser", choices=["firefox", "chromium", "webkit"], value="firefox", scale=1)
        search_engine_input = gr.Dropdown(label="Search Engine (for non-URL queries)", choices=sorted(list(SEARCH_ENGINES.keys())), value="DuckDuckGo", scale=2)
    submit_button = gr.Button("Browse", variant="primary")
    output_json = gr.JSON(label="API Result")
    submit_button.click(fn=perform_web_browse, inputs=[query_input, browser_input, search_engine_input], outputs=output_json, api_name="web_browse")

if __name__ == "__main__":
    demo.launch()