entari-plugin-hyw 4.0.0rc7__py3-none-any.whl → 4.0.0rc9__py3-none-any.whl

{entari_plugin_hyw/browser → hyw_core/browser_control}/service.py

@@ -22,65 +22,22 @@ class ScreenshotService:
         self.headless = headless
         self._manager = None
         self._executor = ThreadPoolExecutor(max_workers=10)
-        self._search_tab_pool = []  # List of Tab objects
-        self._pool_lock = threading.Lock()
         
         if auto_start:
             self._ensure_ready()
-            
-    def prepare_search_tabs_background(self, count: int, url: str = "https://www.google.com") -> None:
-        """
-        Pre-launch tabs for search (BACKGROUND - fire and forget).
-        Tabs are created in background thread, may not be ready immediately.
-        """
-        self._executor.submit(self._prepare_search_tabs_sync, count, url)
 
-    def _prepare_search_tabs_sync(self, count: int, url: str = "https://www.google.com"):
-        """Sync implementation of tab preparation - creates tabs in PARALLEL."""
+    def _get_tab(self, url: str) -> Any:
+        """Create a new tab and navigate to URL."""
+        self._ensure_ready()
+        return self._manager.new_tab(url)
+
+    def _release_tab(self, tab: Any):
+        """Close tab after use."""
+        if not tab: return
         try:
-            self._ensure_ready()
-            page = self._manager.page
-            if not page: return
-            
-            with self._pool_lock:
-                current_count = len(self._search_tab_pool)
-                needed = count - current_count
-            
-            if needed <= 0:
-                return
-                
-            logger.info(f"ScreenshotService: Pre-launching {needed} search tabs for {url} (parallel)...")
-            
-            # Create tabs in parallel using threads
-            created_tabs = [None] * needed
-            
-            def create_single_tab(index):
-                try:
-                    tab = page.new_tab(url)
-                    created_tabs[index] = tab
-                    logger.debug(f"ScreenshotService: Tab {index} ready")
-                except Exception as e:
-                    logger.error(f"ScreenshotService: Failed to create tab {index}: {e}")
-            
-            threads = []
-            for i in range(needed):
-                t = threading.Thread(target=create_single_tab, args=(i,))
-                t.start()
-                threads.append(t)
-            
-            # Wait for all threads to complete
-            for t in threads:
-                t.join()
-            
-            # Add successfully created tabs to pool
-            with self._pool_lock:
-                for tab in created_tabs:
-                    if tab:
-                        self._search_tab_pool.append(tab)
-                logger.info(f"ScreenshotService: Tab pool ready ({len(self._search_tab_pool)} tabs)")
-                
-        except Exception as e:
-            logger.error(f"ScreenshotService: Failed to prepare tabs: {e}")
+            tab.close()
+        except:
+            pass
 
     async def search_via_page_input_batch(self, queries: List[str], url: str, selector: str = "#input") -> List[Dict[str, Any]]:
         """
@@ -104,17 +61,13 @@ class ScreenshotService:
         
         for i in range(len(queries)):
             tab = None
-            # Try to get from pool first
-            with self._pool_lock:
-                if self._search_tab_pool:
-                    tab = self._search_tab_pool.pop(0)
-                    logger.debug(f"ScreenshotService: Got tab {i} from pool")
+            # Try to get from pool first (using shared logic now)
+            try:
+                tab = self._get_tab(target_url)
+            except Exception as e:
+                logger.warning(f"ScreenshotService: Batch search tab creation failed: {e}")
             
-            if not tab:
-                # Create new
-                self._ensure_ready()
-                tab = self._manager.page.new_tab(target_url)
-                logger.debug(f"ScreenshotService: Created tab {i} for {target_url}")
+
             
             tabs.append(tab)
         
@@ -155,7 +108,8 @@ class ScreenshotService:
                 
                 logger.debug(f"Search[{index}]: Waiting for search results...")
                 tab.wait.doc_loaded(timeout=10)
-                time.sleep(0.5)
+                # Reduced settle wait for extraction
+                time.sleep(0.1)
                 
                 logger.debug(f"Search[{index}]: Extracting content...")
                 html = tab.html
@@ -178,8 +132,7 @@ class ScreenshotService:
                 logger.error(f"ScreenshotService: Search error for '{query}': {e}")
                 results[index] = {"content": f"Error: {e}", "title": "Error", "url": "", "html": ""}
             finally:
-                try: tab.close()
-                except: pass
+                self._release_tab(tab)
 
         threads = []
         for i, (tab, query) in enumerate(zip(tabs, queries)):
@@ -248,7 +201,7 @@ class ScreenshotService:
             
             # Small delay for address bar to focus
             import time as _time
-            _time.sleep(0.1)
+            _time.sleep(0.05)
             
             # Type the query
             tab.actions.type(query)
@@ -259,8 +212,8 @@ class ScreenshotService:
             # Wait for page to load
             try:
                 tab.wait.doc_loaded(timeout=timeout)
-                # Additional wait for search results
-                _time.sleep(1)
+                # Reduced wait for initial results
+                _time.sleep(0.2)
             except:
                 pass
             
@@ -292,6 +245,89 @@ class ScreenshotService:
                 try: tab.close()
                 except: pass
 
+    def _scroll_to_bottom(self, tab, step: int = 800, delay: float = 2.0, timeout: float = 10.0):
+        """
+        Scroll down gradually to trigger lazy loading.
+        
+        Args:
+            delay: Max wait time per scroll step (seconds) if images aren't loading.
+        """
+        import time
+        start = time.time()
+        current_pos = 0
+        try:
+            while time.time() - start < timeout:
+                # Scroll down
+                current_pos += step
+                tab.run_js(f"window.scrollTo(0, {current_pos});")
+                
+                # Active Wait: Check if images in viewport are loaded
+                # Poll every 100ms, up to 'delay' seconds
+                wait_start = time.time()
+                while time.time() - wait_start < delay:
+                    all_loaded = tab.run_js("""
+                        return (async () => {
+                            const imgs = Array.from(document.querySelectorAll('img'));
+                            const viewportHeight = window.innerHeight;
+                            
+                            // 1. Identify images currently in viewport
+                            const visibleImgs = imgs.filter(img => {
+                                const rect = img.getBoundingClientRect();
+                                return (rect.top < viewportHeight && rect.bottom > 0) && (rect.width > 0 && rect.height > 0);
+                            });
+                            
+                            if (visibleImgs.length === 0) return true;
+
+                            // 2. Check loading status using decode() AND heuristic for placeholders
+                            // Some sites load a tiny blurred placeholder first. 
+                            const checks = visibleImgs.map(img => {
+                                // HEURISTIC: content is likely not ready if:
+                                // - img has 'data-src' but src is different (or src is empty)
+                                // - img has 'loading="lazy"' and is not complete
+                                // - naturalWidth is very small (placeholder) compared to display width
+                                
+                                const isPlaceholder = (
+                                    (img.getAttribute('data-src') && img.src !== img.getAttribute('data-src')) ||
+                                    (img.naturalWidth < 50 && img.clientWidth > 100) 
+                                );
+                                
+                                if (isPlaceholder) {
+                                    // If it looks like a placeholder, we return false (not loaded)
+                                    // unless it stays like this for too long (handled by outer timeout)
+                                    return Promise.resolve(false);
+                                }
+
+                                if (img.complete && img.naturalHeight > 0) return Promise.resolve(true); 
+                                
+                                return img.decode().then(() => true).catch(() => false); 
+                            });
+                            
+                            // Race against a small timeout to avoid hanging on one broken image
+                            const allDecoded = Promise.all(checks);
+                            const timeout = new Promise(resolve => setTimeout(() => resolve(false), 500));
+                            
+                            // If any check returned false (meaning placeholder or not decoded), result is false
+                            return Promise.race([allDecoded, timeout]).then(results => {
+                                if (!Array.isArray(results)) return results === true;
+                                return results.every(res => res === true);
+                            }); 
+                        })();
+                    """)
+                    if all_loaded:
+                        break
+                    time.sleep(0.1)
+
+                # Check if reached bottom
+                height = tab.run_js("return Math.max(document.body.scrollHeight, document.documentElement.scrollHeight);")
+                if current_pos >= height:
+                    break
+            
+            # Ensure final layout settle
+            time.sleep(0.2)
+            
+        except Exception as e:
+            logger.warning(f"ScreenshotService: Scroll failed: {e}")
+
     def _fetch_page_sync(self, url: str, timeout: float, include_screenshot: bool) -> Dict[str, Any]:
         """Synchronous fetch logic."""
         if not url:
@@ -304,27 +340,48 @@ class ScreenshotService:
             if not page:
                 return {"content": "Error: Browser not available", "title": "Error", "url": url}
             
-            # New Tab with URL directly
-            tab = page.new_tab(url)
+            # Get from pool
+            tab = self._get_tab(url)
             
             # Wait logic - optimized for search pages
             is_search_page = any(s in url.lower() for s in ['search', 'bing.com', 'duckduckgo', 'google.com/search', 'searx'])
             if is_search_page:
-                # Optimized waiting for search engine results
-                try:
-                    # Google uses #search or #rso
-                    # DuckDuckGo uses #react-layout
-                    # Bing uses #b_results
-                    if 'google' in url.lower():
-                        # Wait for results container (fastest possible return)
-                        tab.ele('#search', timeout=timeout)
-                    elif 'bing' in url.lower():
-                        tab.ele('#b_results', timeout=timeout)
-                    else:
-                        # Generic search fallback
-                        tab.wait.doc_loaded(timeout=timeout)
-                except:
-                    pass
+                # Optimized waiting: Rapidly poll for ACTUAL results > 0
+                start_time = time.time()
+                
+                # Special fast-path for DDG Lite (HTML only, no JS rendering needed)
+                if 'lite.duckduckgo' in url:
+                    # just wait for body, it's static HTML
+                     try:
+                         tab.wait.doc_loaded(timeout=timeout)
+                     except: pass
+                     # Sleep tiny bit to ensure render
+                     time.sleep(0.5)
+                else:
+                    while time.time() - start_time < timeout:
+                        found_results = False
+                        try:
+                            if 'google' in url.lower():
+                                # Check if we have any result items (.g, .MjjYud) or the main container (#search)
+                                # Using checks with minimal timeout to allow fast looping
+                                if tab.ele('.g', timeout=0.1) or tab.ele('.MjjYud', timeout=0.1) or tab.ele('#search', timeout=0.1):
+                                    found_results = True
+                            elif 'bing' in url.lower():
+                                if tab.ele('.b_algo', timeout=0.1) or tab.ele('#b_results', timeout=0.1):
+                                    found_results = True
+                            elif 'duckduckgo' in url.lower():
+                                if tab.ele('.result', timeout=0.1) or tab.ele('#react-layout', timeout=0.1):
+                                    found_results = True
+                            else:
+                                # Generic fallback: wait for body to be populated
+                                if tab.ele('body', timeout=0.1):
+                                    found_results = True
+                        except:
+                            pass
+                        
+                        if found_results:
+                            break
+                        time.sleep(0.05)  # Faster polling (50ms) as requested
             else:
                 # 1. Wait for document to settle (Fast Dynamic Wait)
                 try:
@@ -451,8 +508,7 @@ class ScreenshotService:
             return {"content": f"Error: fetch failed ({e})", "title": "Error", "url": url}
         finally:
             if tab:
-                try: tab.close()
-                except: pass
+               self._release_tab(tab)
 
     async def fetch_pages_batch(self, urls: List[str], timeout: float = 20.0, include_screenshot: bool = True) -> List[Dict[str, Any]]:
         """Fetch multiple pages concurrently."""
@@ -461,6 +517,13 @@ class ScreenshotService:
         tasks = [self.fetch_page(url, timeout, include_screenshot) for url in urls]
         return await asyncio.gather(*tasks, return_exceptions=True)
 
+    async def screenshot_urls_batch(self, urls: List[str], timeout: float = 15.0, full_page: bool = True) -> List[Optional[str]]:
+        """Take screenshots of multiple URLs concurrently."""
+        if not urls: return []
+        logger.info(f"ScreenshotService: Batch screenshot {len(urls)} URLs")
+        tasks = [self.screenshot_url(url, timeout=timeout, full_page=full_page) for url in urls]
+        return await asyncio.gather(*tasks, return_exceptions=True)
+
     async def screenshot_url(self, url: str, wait_load: bool = True, timeout: float = 15.0, full_page: bool = False, quality: int = 80) -> Optional[str]:
         """Screenshot URL (Async wrapper for sync)."""
         loop = asyncio.get_running_loop()
@@ -480,36 +543,201 @@ class ScreenshotService:
             if not page: return None
             
             tab = page.new_tab(url)
+            
+            # Start monitoring network traffic
             try:
-                if wait_load:
-                    tab.wait.load_complete(timeout=timeout)
-                else:
-                    tab.wait.doc_loaded(timeout=timeout)
-            except: pass
+                # Listen for data packets (XHR/Fetch/POST)
+                # Targets: xhr, fetch. POST usually falls under these or Document.
+                tab.listen.start(targets=True) # Listen to everything for now to be safe
+            except Exception as e:
+                logger.warning(f"ScreenshotService: Failed to start network listener: {e}")
+
+            try:
+                # Wait for full page load (including JS execution)
+                tab.wait.load_complete(timeout=timeout)
+                
+                # Wait for actual content to appear (for CDN verification pages)
+                # Smart Wait Logic (Final Robust):
+                # 1. Network Idle: Wait for silence in XHR/POST
+                # 2. Stability: Wait for Height/Text/DOM stability
+                
+                time.sleep(1.5)  # user request: force wait 1.5s before detection
+                
+                last_h = 0
+                last_text_len = 0
+                last_html_len = 0
+                stable_count = 0
+                
+                for i in range(200):  # Max 200 iterations (~20s)
+                    try:
+                        # 1. Check Network Activity
+                        has_recent_network = False
+                        try:
+                            # Iterate over any captured packets since last check
+                            for packet in tab.listen.steps(timeout=0.01):
+                                # Check if it's a significant request (POST or XHR/Fetch)
+                                method = packet.method.upper()
+                                r_type = packet.resourceType.upper() if getattr(packet, 'resourceType', None) else ""
+                                
+                                # Interested in: POST requests OR any XHR/Fetch response
+                                if method == 'POST' or 'XMLHTTPREQUEST' in r_type or 'FETCH' in r_type:
+                                    # Ignore some common noise? (Optional: analytics, tracking)
+                                    # For now, simplistic approach: any API traffic resets stability
+                                    has_recent_network = True
+                                    # logger.debug(f"Network Activity: {method} {packet.url[:50]}")
+                                    break
+                        except:
+                            pass
+
+                        # 2. Check DOM State
+                        state = tab.run_js('''
+                            return {
+                                ready: document.readyState === 'complete',
+                                title: document.title,
+                                height: Math.max(
+                                    document.body.scrollHeight || 0,
+                                    document.documentElement.scrollHeight || 0
+                                ),
+                                text: document.body.innerText.substring(0, 1000) || "",
+                                html_len: document.body.innerHTML.length || 0
+                            };
+                        ''') or {'ready': False, 'title': "", 'height': 0, 'text': "", 'html_len': 0}
+                        
+                        is_ready = state.get('ready', False)
+                        title = state.get('title', "").lower()
+                        current_h = int(state.get('height', 0))
+                        text_content = state.get('text', "")
+                        text_len = len(text_content)
+                        html_len = int(state.get('html_len', 0))
+                        text_lower = text_content.lower()
+                        
+                        # Blacklist check (Loading indicators)
+                        is_verification = "checking your browser" in text_lower or \
+                                        "just a moment" in text_lower or \
+                                        "please wait" in text_lower or \
+                                        "security check" in title or \
+                                        "just a moment" in title or \
+                                        "loading..." in title
+                        
+                        # Stability check (Multi-metric + Network)
+                        # We require STABILITY across Height, Text Length, DOM Size AND Network Silence
+                        is_height_stable = current_h == last_h
+                        is_text_stable = abs(text_len - last_text_len) < 5  # Allow minor fluctuations
+                        is_dom_stable = abs(html_len - last_html_len) < 20 # Allow minor fluctuations (ads/tracking)
+
+                        if is_height_stable and is_text_stable and is_dom_stable and not has_recent_network:
+                            stable_count += 1
+                        else:
+                            # Reset if ANY metric changed or NETWORK active
+                            stable_count = 0
+                            # if has_recent_network: logger.debug("Stability reset: Network Activity")
+                        
+                        # Conditions
+                        has_content = text_len > 100 # At least 100 real chars
+                        
+                        # Dynamic Stability Requirement:
+                        # If page looks like it's loading (small content), require longer stability
+                        if has_recent_network:
+                             # If we just saw network, enforce at least 1s (10 ticks) clean silence even for large pages
+                             required_stability = max(10, 40 if text_len < 500 else 25)
+                        else:
+                             required_stability = 40 if text_len < 500 else 25  # 4.0s or 2.5s
+                        
+                        is_stable = stable_count >= required_stability
+                        
+                        # Pass if all conditions met
+                        if is_ready and not is_verification and has_content and is_stable:
+                            break
+                            
+                        last_h = current_h
+                        last_text_len = text_len
+                        last_html_len = html_len
+                        
+                        # Wait timing within loop (tab.listen.steps consumed some time, so sleep less)
+                        try: time.sleep(0.05)
+                        except: pass
+                        
+                    except Exception:
+                        stable_count = 0
+                        try: time.sleep(0.1)
+                        except: pass
+                        continue
+                
+                # Cleanup listener
+                try: tab.listen.stop()
+                except: pass
+                
+                # DEBUG: Save HTML to inspect what happened (in data dir)
+                try:
+                    import os
+                    log_path = os.path.join(os.getcwd(), "data", "browser.log.html")
+                    with open(log_path, "w", encoding="utf-8") as f:
+                        f.write(f"<!-- URL: {url} -->\n")
+                        f.write(tab.html)
+                except: pass
+                
+                # Use faster scroll step (800) to ensure lazy loaded images appear
+                self._scroll_to_bottom(tab, step=800, delay=2.0, timeout=min(timeout, 10))
+                
+            except:
+                pass
             
-            # Wait for main element
-            if tab.ele("#main-container"):
-                pass 
+            # Refine calculation: Set viewport width to 1024
+
+            capture_width = 1024
             
+            # Calculate actual content height after lazy loading
+            try:
+                # Use a robust height calculation
+                content_height = tab.run_js('''
+                    return Math.max(
+                        document.body.scrollHeight || 0,
+                        document.documentElement.scrollHeight || 0,
+                        document.body.offsetHeight || 0,
+                        document.documentElement.offsetHeight || 0,
+                        document.documentElement.clientHeight || 0
+                    );
+                ''')
+                # Add a small buffer and cap at 15000px to prevent memory issues
+                h = min(int(content_height) + 50, 15000)
+            except:
+                h = 1000  # Fallback
+            
+            # Set viewport to full content size for single-shot capture
+            try:
+                tab.run_cdp('Emulation.setDeviceMetricsOverride', 
+                            width=capture_width, height=h, deviceScaleFactor=1, mobile=False)
+            except:
+                pass
+
             # Scrollbar Hiding
             from .manager import SharedBrowserManager
             SharedBrowserManager.hide_scrollbars(tab)
-            tab.run_js("""
-                const style = document.createElement('style');
-                style.textContent = `
-                    ::-webkit-scrollbar { display: none !important; }
-                    html, body { -ms-overflow-style: none !important; scrollbar-width: none !important; }
-                `;
-                document.head.appendChild(style);
-                document.documentElement.style.overflow = 'hidden';
-                document.body.style.overflow = 'hidden';
-            """)
             
-            ele = tab.ele("#main-container")
-            if ele:
-                return ele.get_screenshot(as_base64='jpg', quality=quality)
-            else:
-                return tab.get_screenshot(as_base64='jpg', full_page=full_page, quality=quality)
+            # Scroll back to top before screenshot
+            tab.run_js("window.scrollTo(0, 0);")
+            
+            # Content is already loaded by _scroll_to_bottom which waits for images
+            # Just recalculate final height (content may have grown during scrolling)
+            try:
+                final_height = tab.run_js('''
+                    return Math.max(
+                        document.body.scrollHeight || 0,
+                        document.documentElement.scrollHeight || 0,
+                        document.body.offsetHeight || 0,
+                        document.documentElement.offsetHeight || 0
+                    );
+                ''')
+                final_h = min(int(final_height) + 50, 15000)
+                if final_h != h:
+                    tab.run_cdp('Emulation.setDeviceMetricsOverride', 
+                                width=capture_width, height=final_h, deviceScaleFactor=1, mobile=False)
+            except:
+                pass
+            
+            # Use full_page=False because we manually set the viewport to the full height
+            # This avoids stitching artifacts and blank spaces
+            return tab.get_screenshot(as_base64='jpg', full_page=False)
                 
         except Exception as e:
             logger.error(f"ScreenshotService: Screenshot URL failed: {e}")