entari-plugin-hyw 4.0.0rc8__py3-none-any.whl → 4.0.0rc10__py3-none-any.whl

hyw_core/browser_control/engines/__init__.py

@@ -5,13 +5,11 @@ Provides search engine adapters for different search providers.
 """
 
 from .base import SearchEngine
-from .google import GoogleEngine
 from .duckduckgo import DuckDuckGoEngine
 from .default import DefaultEngine
 
 __all__ = [
     "SearchEngine",
-    "GoogleEngine", 
     "DuckDuckGoEngine",
     "DefaultEngine",
 ]

hyw_core/browser_control/manager.py

@@ -124,15 +124,28 @@ class SharedBrowserManager:
     @staticmethod
     def hide_scrollbars(page: ChromiumPage):
         """
-        Robustly hide scrollbars using CDP commands.
-        This eliminates the reserved space/gutter that standard CSS might miss.
+        Robustly hide scrollbars using CDP commands AND CSS injection.
+        This provides double protection against scrollbar gutters.
         """
         try:
-            # Emulation.setScrollbarsHidden is a CDP command
+            # 1. CDP Command
             page.run_cdp('Emulation.setScrollbarsHidden', hidden=True)
-            logger.debug("SharedBrowserManager: CDP scrollbars hidden.")
+            
+            # 2. CSS Injection (Standard + Webkit)
+            css = """
+                ::-webkit-scrollbar { display: none !important; width: 0 !important; height: 0 !important; }
+                * { -ms-overflow-style: none !important; scrollbar-width: none !important; }
+            """
+            # Inject into current page
+            page.run_js(f"""
+                const style = document.createElement('style');
+                style.textContent = `{css}`;
+                document.head.appendChild(style);
+            """)
+            
+            logger.debug("SharedBrowserManager: Scrollbars hidden via CDP + CSS.")
         except Exception as e:
-            logger.warning(f"SharedBrowserManager: Failed to hide scrollbars via CDP: {e}")
+            logger.warning(f"SharedBrowserManager: Failed to hide scrollbars: {e}")
 
 
 # Module-level singleton accessor

hyw_core/browser_control/renderer.py

@@ -55,16 +55,43 @@ class ContentRenderer:
         loop = asyncio.get_running_loop()
         return await loop.run_in_executor(self._executor, self._prepare_tab_sync)
 
-    def _wait_for_render_finished(self, tab, timeout: float = 3.0):
+    def _wait_for_render_finished(self, tab, timeout: float = 12.0, context: str = ""):
         """Wait for window.RENDER_FINISHED to be true in the tab."""
         import time as pytime
         start = pytime.time()
+        
+        # Check initial state
+        initial_state = tab.run_js("return window.RENDER_FINISHED")
+        logger.debug(f"ContentRenderer[{context}]: Starting wait, initial RENDER_FINISHED={initial_state}")
+        
+        # If already true, it's stale from previous render - need to wait for JS to reset it
+        if initial_state:
+            logger.debug(f"ContentRenderer[{context}]: RENDER_FINISHED was true, waiting for reset...")
+            # Wait for JS to reset it to false (updateRenderData sets it to false)
+            reset_start = pytime.time()
+            while pytime.time() - reset_start < 1.0:  # 1s max to wait for reset
+                is_reset = tab.run_js("return window.RENDER_FINISHED")
+                if not is_reset:
+                    logger.debug(f"ContentRenderer[{context}]: RENDER_FINISHED reset to false")
+                    break
+                pytime.sleep(0.05)
+            else:
+                logger.warning(f"ContentRenderer[{context}]: RENDER_FINISHED not reset, force resetting via JS")
+                tab.run_js("window.RENDER_FINISHED = false")
+        
+        # Now wait for it to become true
+        poll_count = 0
         while pytime.time() - start < timeout:
             is_finished = tab.run_js("return window.RENDER_FINISHED")
+            poll_count += 1
             if is_finished:
+                elapsed = pytime.time() - start
+                logger.debug(f"ContentRenderer[{context}]: RENDER_FINISHED=true after {elapsed:.2f}s ({poll_count} polls)")
                 return True
-            pytime.sleep(0.05) # Fast polling
-        logger.warning(f"ContentRenderer: Wait for RENDER_FINISHED timed out after {timeout}s")
+            pytime.sleep(0.1)  # Poll every 100ms
+        
+        elapsed = pytime.time() - start
+        logger.warning(f"ContentRenderer[{context}]: Wait for RENDER_FINISHED timed out after {elapsed:.2f}s ({poll_count} polls)")
         return False
 
     def _prepare_tab_sync(self) -> str:
@@ -89,8 +116,9 @@ class ContentRenderer:
                 "theme_color": "#ef4444",
             }
             
+            logger.debug(f"ContentRenderer: Calling warmup updateRenderData for tab {tab_id}")
             tab.run_js(f"window.updateRenderData({json.dumps(warmup_data)})")
-            self._wait_for_render_finished(tab, timeout=5.0)
+            self._wait_for_render_finished(tab, timeout=12.0, context=f"warmup:{tab_id}")
             
             # Wait for main-container after warmup (Vue needs to render it)
             tab.ele('#main-container', timeout=3)
@@ -203,7 +231,7 @@ class ContentRenderer:
             
             # 1. Update Data & Wait for Finished flag
             tab.run_js(f"window.updateRenderData({json.dumps(render_data)})")
-            self._wait_for_render_finished(tab)
+            self._wait_for_render_finished(tab, context=f"batch:{tab_id}")
 
             # 2. Dynamic Resize
             # Get actual content height to prevent clipping
@@ -330,10 +358,12 @@ class ContentRenderer:
                 "theme_color": theme_color,
             }
             
+            actual_tab_id = getattr(tab, 'tab_id', 'unknown')
+            logger.info(f"ContentRenderer: Calling updateRenderData for tab {actual_tab_id}, markdown length={len(markdown_content)}")
             tab.run_js(f"window.updateRenderData({json.dumps(render_data)})")
 
             # Wait for event-driven finish
-            self._wait_for_render_finished(tab, timeout=5.0)
+            self._wait_for_render_finished(tab, timeout=12.0, context=f"render:{actual_tab_id}")
             
             # Dynamic Resize
             scroll_height = tab.run_js('return Math.max(document.body.scrollHeight, document.documentElement.scrollHeight);')

hyw_core/browser_control/service.py

@@ -13,6 +13,10 @@ from typing import Optional, Dict, Any, List
 from loguru import logger
 import trafilatura
 
+# Import intelligent completeness checker
+from ..crawling.completeness import CompletenessChecker, trigger_lazy_load
+from ..crawling.models import CrawlConfig
+
 class ScreenshotService:
     """
     Browser Service using DrissionPage.
@@ -537,31 +541,83 @@ class ScreenshotService:
         """Synchronous screenshot."""
         if not url: return None
         tab = None
+        capture_width = 1024  # Standard capture width
+        
         try:
             self._ensure_ready()
             page = self._manager.page
             if not page: return None
             
-            tab = page.new_tab(url)
+            # Create blank tab first
+            tab = page.new_tab()
+            
+            # Set viewport BEFORE navigation so page renders at target width from the start
+            # This eliminates the need for post-load resize and reflow
+            try:
+                tab.run_cdp('Emulation.setDeviceMetricsOverride', 
+                            width=capture_width, height=900, deviceScaleFactor=1, mobile=False)
+            except:
+                pass
+            
+            # Now navigate to the URL - page will render at target width
+            tab.get(url)
+            
+            # Start monitoring network traffic
+            try:
+                # 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}")
+
+            # Initialize crawl config for completeness checking (defined outside try for scope)
+            crawl_config = CrawlConfig(
+                scan_full_page=True,
+                scroll_step=800,
+                scroll_delay=0.5,
+                scroll_timeout=min(timeout, 10),
+                image_load_timeout=8.0,
+                image_stability_checks=3,
+            )
+
             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. FORCED WAIT: 1.5s to allow initial redirects/rendering to start.
-                # 2. Browser ReadyState Complete
-                # 3. Height Stable for 2.0 seconds (20 checks)
-                # 4. Text > 100 chars (Crucial: Distinguishes stable content from stable spinners)
-                # 5. No Blacklist phrases
+                # 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',
@@ -571,42 +627,62 @@ class ScreenshotService:
                                     document.documentElement.scrollHeight || 0
                                 ),
                                 text: document.body.innerText.substring(0, 1000) || "",
-                                html: document.body.innerHTML.substring(0, 500) // Debug intro
+                                html_len: document.body.innerHTML.length || 0
                             };
-                        ''') or {'ready': False, 'title': "", 'height': 0, 'text': ""}
+                        ''') 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
+                        # 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
+                                        "just a moment" in title or \
+                                        "loading..." in title
                         
-                        # Stability check
-                        if current_h == last_h:
+                        # 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
-                        is_stable = stable_count >= 20  # Always require 2s stability
+                        
+                        # 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
-                        try: tab.wait.eles_loaded(timeout=0.1)
+                        # Wait timing within loop (tab.listen.steps consumed some time, so sleep less)
+                        try: time.sleep(0.05)
                         except: pass
                         
                     except Exception:
@@ -615,6 +691,10 @@ class ScreenshotService:
                         except: pass
                         continue
                 
+                # Cleanup listener
+                try: tab.listen.stop()
+                except: pass
+                
                 # DEBUG: Save HTML to inspect what happened (in data dir)
                 try:
                     import os
@@ -624,48 +704,33 @@ class ScreenshotService:
                         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))
+                # Use crawling module for lazy loading trigger (config defined above)
+                trigger_lazy_load(tab, crawl_config)
                 
             except:
                 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
+            # Scrollbar Hiding first (before any height calculation)
             from .manager import SharedBrowserManager
             SharedBrowserManager.hide_scrollbars(tab)
             
-            # Scroll back to top before screenshot
+            # Scroll back to top
             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)
+            # Use CompletenessChecker to verify all images are loaded
+            checker = CompletenessChecker(crawl_config)
+            completeness = checker.wait_for_complete(tab, timeout=crawl_config.image_load_timeout)
+            
+            logger.info(
+                f"ScreenshotService: Image completeness: "
+                f"{completeness.loaded_images}/{completeness.total_images} loaded, "
+                f"{completeness.failed_images} pending, "
+                f"{completeness.placeholder_images} placeholders, "
+                f"complete={completeness.is_complete}"
+            )
+            
+            # Now calculate final height ONCE after all content loaded
+            # CompletenessChecker already verified height stability
             try:
                 final_height = tab.run_js('''
                     return Math.max(
@@ -675,13 +740,15 @@ class ScreenshotService:
                         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)
+                h = min(int(final_height) + 50, 15000)
+                tab.run_cdp('Emulation.setDeviceMetricsOverride', 
+                            width=capture_width, height=h, deviceScaleFactor=1, mobile=False)
             except:
                 pass
             
+            # Final scroll to top
+            tab.run_js("window.scrollTo(0, 0);")
+            
             # 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)

hyw_core/crawling/__init__.py

@@ -0,0 +1,18 @@
+"""
+hyw_core.crawling - Intelligent Web Crawling Module
+
+Provides Crawl4AI-inspired adaptive crawling with:
+- Page completeness guarantees (image loading verification)
+- Content quality scoring
+- Adaptive stop logic
+"""
+
+from .models import CrawlConfig, PageResult, CompletenessResult
+from .completeness import CompletenessChecker
+
+__all__ = [
+    "CrawlConfig",
+    "PageResult", 
+    "CompletenessResult",
+    "CompletenessChecker",
+]