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

hyw_core/crawling/completeness.py

@@ -0,0 +1,348 @@
+"""
+Page Completeness Checker
+
+Multi-signal page readiness detection with focus on image loading guarantees.
+Implements Crawl4AI's wait_for_images and networkidle concepts.
+"""
+
+import time
+from typing import Any, Optional, Dict
+from dataclasses import dataclass
+from loguru import logger
+
+from .models import CrawlConfig, CompletenessResult
+
+
+# CSP-compliant JavaScript for image loading verification
+# Based on Crawl4AI's wait_for_images logic
+IMAGE_CHECK_JS = """
+(() => {
+    const results = {
+        total: 0,
+        loaded: 0,
+        failed: 0,
+        placeholders: 0,
+        details: []
+    };
+    
+    const imgs = Array.from(document.querySelectorAll('img'));
+    const viewportHeight = window.innerHeight;
+    const minSize = %d;  // Injected from Python
+    
+    for (const img of imgs) {
+        // Skip tiny images (icons, tracking pixels)
+        if (img.clientWidth < minSize && img.clientHeight < minSize) {
+            continue;
+        }
+        
+        // Skip images outside viewport (unless scan_full_page)
+        const rect = img.getBoundingClientRect();
+        const inViewport = rect.top < viewportHeight * 2 && rect.bottom > -viewportHeight;
+        
+        if (!inViewport && !%s) {  // scan_full_page injected
+            continue;
+        }
+        
+        results.total++;
+        
+        const src = img.src || img.getAttribute('data-src') || '';
+        const isPlaceholder = (
+            (img.getAttribute('data-src') && img.src !== img.getAttribute('data-src')) ||
+            (img.naturalWidth < 50 && img.clientWidth > 100) ||
+            src.includes('placeholder') ||
+            src.includes('loading') ||
+            src.startsWith('data:image/svg+xml') ||
+            (img.naturalWidth === 1 && img.naturalHeight === 1)
+        );
+        
+        if (isPlaceholder) {
+            results.placeholders++;
+            results.details.push({
+                src: src.substring(0, 100),
+                status: 'placeholder',
+                natural: [img.naturalWidth, img.naturalHeight],
+                display: [img.clientWidth, img.clientHeight]
+            });
+            continue;
+        }
+        
+        // Check if fully loaded
+        if (img.complete && img.naturalWidth > 0 && img.naturalHeight > 0) {
+            results.loaded++;
+            results.details.push({
+                src: src.substring(0, 100),
+                status: 'loaded',
+                natural: [img.naturalWidth, img.naturalHeight]
+            });
+        } else {
+            results.failed++;
+            results.details.push({
+                src: src.substring(0, 100),
+                status: 'pending',
+                complete: img.complete,
+                natural: [img.naturalWidth, img.naturalHeight]
+            });
+        }
+    }
+    
+    return results;
+})()
+"""
+
+# JavaScript for height stability check
+HEIGHT_CHECK_JS = """
+(() => {
+    return {
+        height: Math.max(
+            document.body.scrollHeight || 0,
+            document.documentElement.scrollHeight || 0,
+            document.body.offsetHeight || 0,
+            document.documentElement.offsetHeight || 0
+        ),
+        ready: document.readyState === 'complete'
+    };
+})()
+"""
+
+
+class CompletenessChecker:
+    """
+    Multi-signal page completeness verification.
+    
+    Signals checked:
+    1. Image Loading (naturalWidth > 0, not placeholder)
+    2. Height Stability (no layout shifts)
+    3. DOM Ready State
+    
+    Usage:
+        checker = CompletenessChecker(CrawlConfig())
+        result = checker.check(tab)
+        
+        # Or wait until complete
+        result = checker.wait_for_complete(tab)
+    """
+    
+    def __init__(self, config: Optional[CrawlConfig] = None):
+        self._config = config or CrawlConfig()
+    
+    def check(self, tab: Any) -> CompletenessResult:
+        """
+        Perform a single completeness check.
+        
+        Args:
+            tab: DrissionPage tab object
+            
+        Returns:
+            CompletenessResult with all signals
+        """
+        start = time.time()
+        
+        # Check images
+        img_result = self._check_images(tab)
+        
+        # Check height
+        height_result = self._check_height(tab)
+        
+        # Determine overall completeness
+        # Complete if: all real images loaded AND height is stable
+        total_pending = img_result.get('failed', 0) + img_result.get('placeholders', 0)
+        all_images_loaded = total_pending == 0 or img_result.get('total', 0) == 0
+        
+        return CompletenessResult(
+            is_complete=all_images_loaded and height_result.get('ready', False),
+            total_images=img_result.get('total', 0),
+            loaded_images=img_result.get('loaded', 0),
+            failed_images=img_result.get('failed', 0),
+            placeholder_images=img_result.get('placeholders', 0),
+            height=height_result.get('height', 0),
+            height_stable=True,  # Single check can't determine stability
+            network_idle=True,   # TODO: network monitoring
+            check_duration=time.time() - start
+        )
+    
+    def wait_for_complete(
+        self, 
+        tab: Any, 
+        timeout: Optional[float] = None
+    ) -> CompletenessResult:
+        """
+        Wait for page to be fully loaded with image guarantees.
+        
+        Uses multi-signal approach:
+        1. Poll image loading status
+        2. Track height stability
+        3. Respect timeout
+        
+        Args:
+            tab: DrissionPage tab object
+            timeout: Max wait time (default from config)
+            
+        Returns:
+            CompletenessResult (may not be complete if timeout)
+        """
+        timeout = timeout or self._config.image_load_timeout
+        start = time.time()
+        
+        height_history = []
+        stable_count = 0
+        last_result = None
+        
+        logger.debug(f"CompletenessChecker: Starting wait (timeout={timeout}s)")
+        
+        while time.time() - start < timeout:
+            result = self.check(tab)
+            last_result = result
+            
+            # Track height stability
+            height_history.append(result.height)
+            if len(height_history) > self._config.height_stability_checks:
+                height_history.pop(0)
+            
+            # Check if height is stable
+            height_stable = False
+            if len(height_history) >= self._config.height_stability_checks:
+                max_h = max(height_history)
+                min_h = min(height_history)
+                height_stable = (max_h - min_h) <= self._config.height_stability_threshold
+            
+            # Log progress
+            elapsed = time.time() - start
+            logger.debug(
+                f"CompletenessChecker: [{elapsed:.1f}s] "
+                f"images={result.loaded_images}/{result.total_images} "
+                f"pending={result.failed_images} "
+                f"placeholders={result.placeholder_images} "
+                f"height={result.height} stable={height_stable}"
+            )
+            
+            # Check if all images loaded AND height stable
+            all_loaded = (
+                result.failed_images == 0 and 
+                result.placeholder_images == 0
+            )
+            
+            if all_loaded and height_stable:
+                stable_count += 1
+                if stable_count >= self._config.image_stability_checks:
+                    logger.info(
+                        f"CompletenessChecker: Page complete! "
+                        f"{result.loaded_images} images loaded, "
+                        f"height={result.height}, "
+                        f"took {elapsed:.2f}s"
+                    )
+                    return CompletenessResult(
+                        is_complete=True,
+                        total_images=result.total_images,
+                        loaded_images=result.loaded_images,
+                        failed_images=0,
+                        placeholder_images=0,
+                        height=result.height,
+                        height_stable=True,
+                        network_idle=True,
+                        check_duration=elapsed
+                    )
+            else:
+                stable_count = 0
+            
+            # Wait before next check
+            time.sleep(self._config.image_check_interval)
+        
+        # Timeout reached
+        elapsed = time.time() - start
+        logger.warning(
+            f"CompletenessChecker: Timeout after {elapsed:.1f}s! "
+            f"images={last_result.loaded_images}/{last_result.total_images} "
+            f"pending={last_result.failed_images}"
+        )
+        
+        # Return last result with is_complete=False
+        if last_result:
+            return CompletenessResult(
+                is_complete=False,
+                total_images=last_result.total_images,
+                loaded_images=last_result.loaded_images,
+                failed_images=last_result.failed_images,
+                placeholder_images=last_result.placeholder_images,
+                height=last_result.height,
+                height_stable=len(set(height_history)) == 1,
+                network_idle=True,
+                check_duration=elapsed
+            )
+        
+        return CompletenessResult(
+            is_complete=False,
+            total_images=0, loaded_images=0, failed_images=0, placeholder_images=0,
+            height=0, height_stable=False, network_idle=False,
+            check_duration=elapsed
+        )
+    
+    def _check_images(self, tab: Any) -> Dict[str, Any]:
+        """Run image check JavaScript and return results."""
+        try:
+            js = IMAGE_CHECK_JS % (
+                self._config.min_image_size,
+                'true' if self._config.scan_full_page else 'false'
+            )
+            result = tab.run_js(js, as_expr=True)
+            return result or {'total': 0, 'loaded': 0, 'failed': 0, 'placeholders': 0}
+        except Exception as e:
+            logger.warning(f"CompletenessChecker: Image check failed: {e}")
+            return {'total': 0, 'loaded': 0, 'failed': 0, 'placeholders': 0}
+    
+    def _check_height(self, tab: Any) -> Dict[str, Any]:
+        """Run height check JavaScript and return results."""
+        try:
+            result = tab.run_js(HEIGHT_CHECK_JS, as_expr=True)
+            return result or {'height': 0, 'ready': False}
+        except Exception as e:
+            logger.warning(f"CompletenessChecker: Height check failed: {e}")
+            return {'height': 0, 'ready': False}
+
+
+def trigger_lazy_load(tab: Any, config: Optional[CrawlConfig] = None) -> None:
+    """
+    Scroll through page to trigger lazy-loaded images.
+    
+    Implements Crawl4AI's scan_full_page behavior.
+    
+    Args:
+        tab: DrissionPage tab object
+        config: Crawl configuration
+    """
+    config = config or CrawlConfig()
+    if not config.scan_full_page:
+        return
+    
+    start = time.time()
+    current_pos = 0
+    
+    logger.debug("CompletenessChecker: Starting lazy load trigger scroll")
+    
+    try:
+        while time.time() - start < config.scroll_timeout:
+            # Scroll down
+            current_pos += config.scroll_step
+            tab.run_js(f"window.scrollTo(0, {current_pos});")
+            
+            # Wait for lazy load triggers
+            time.sleep(config.scroll_delay)
+            
+            # Check if reached bottom
+            height = tab.run_js("""
+                return Math.max(
+                    document.body.scrollHeight || 0,
+                    document.documentElement.scrollHeight || 0
+                );
+            """, as_expr=True) or 0
+            
+            if current_pos >= height:
+                break
+        
+        # Scroll back to top
+        tab.run_js("window.scrollTo(0, 0);")
+        
+        elapsed = time.time() - start
+        logger.debug(f"CompletenessChecker: Lazy load scroll complete in {elapsed:.1f}s")
+        
+    except Exception as e:
+        logger.warning(f"CompletenessChecker: Lazy load scroll failed: {e}")

hyw_core/crawling/models.py

@@ -0,0 +1,88 @@
+"""
+Crawling Data Models
+
+Core data structures for the intelligent crawling system.
+"""
+
+from dataclasses import dataclass, field
+from typing import List, Optional, Set
+
+
+@dataclass
+class CrawlConfig:
+    """
+    Crawling configuration with Crawl4AI-style parameters.
+    
+    Focuses on page completeness and image loading guarantees.
+    """
+    # Wait Strategy (Priority: Image Loading)
+    wait_for_images: bool = True
+    wait_until: str = "networkidle"  # domcontentloaded | networkidle | load
+    delay_before_return: float = 0.1
+    page_timeout: float = 30.0
+    
+    # Image Loading Specific
+    image_load_timeout: float = 10.0  # Max wait for images
+    image_stability_checks: int = 3   # Consecutive stable checks needed
+    image_check_interval: float = 0.2 # Interval between checks
+    min_image_size: int = 50          # Ignore images smaller than this
+    
+    # Scroll for lazy loading
+    scan_full_page: bool = True
+    scroll_step: int = 800
+    scroll_delay: float = 0.5
+    scroll_timeout: float = 15.0
+    
+    # Height Stability
+    height_stability_checks: int = 3
+    height_stability_threshold: int = 10  # pixels
+    
+    # Future: Adaptive Stop Logic
+    confidence_threshold: float = 0.75
+    min_gain_threshold: float = 0.1
+    max_pages: int = 20
+
+
+@dataclass
+class CompletenessResult:
+    """Result from completeness check."""
+    is_complete: bool
+    total_images: int
+    loaded_images: int
+    failed_images: int
+    placeholder_images: int
+    height: int
+    height_stable: bool
+    network_idle: bool
+    check_duration: float
+    
+    @property
+    def image_load_ratio(self) -> float:
+        if self.total_images == 0:
+            return 1.0
+        return self.loaded_images / self.total_images
+
+
+@dataclass
+class PageResult:
+    """Result from fetching a single page."""
+    url: str
+    final_url: str
+    title: str
+    html: str
+    content: str  # Extracted markdown
+    images: List[str] = field(default_factory=list)  # base64 images
+    screenshot: Optional[str] = None
+    
+    # Quality Signals
+    load_time: float = 0.0
+    completeness: Optional[CompletenessResult] = None
+    
+    # Error handling
+    error: Optional[str] = None
+    
+    @property
+    def is_complete(self) -> bool:
+        if self.completeness is None:
+            return False
+        return self.completeness.is_complete

hyw_core/search.py

@@ -8,7 +8,6 @@ from loguru import logger
 from .browser_control.service import get_screenshot_service
 # Search engines from browser_control subpackage
 from .browser_control.engines.duckduckgo import DuckDuckGoEngine
-from .browser_control.engines.google import GoogleEngine
 from .browser_control.engines.default import DefaultEngine
 
 class SearchService:
@@ -21,15 +20,14 @@ class SearchService:
         # Domain blocking
         self._blocked_domains = getattr(config, "blocked_domains", []) or []
         
-        # Select Engine - DefaultEngine when not specified
+        # Select Engine - DuckDuckGo is the default and only engine
         self._engine_name = getattr(config, "search_engine", None)
         if self._engine_name:
             self._engine_name = self._engine_name.lower()
         
-        if self._engine_name == "google":
-            self._engine = GoogleEngine()
-        elif self._engine_name == "default_address_bar": # Explicitly requested address bar capability if needed
-             self._engine = DefaultEngine()
+        if self._engine_name == "default_address_bar":
+            # Explicitly requested address bar capability if needed
+            self._engine = DefaultEngine()
         else:
             # Default: use DuckDuckGo
             self._engine = DuckDuckGoEngine()

hyw_core/browser_control/engines/google.py

@@ -1,155 +0,0 @@
-
-import urllib.parse
-import re
-from typing import List, Dict, Any
-from loguru import logger
-from .base import SearchEngine
-
-
-class GoogleEngine(SearchEngine):
-    """
-    Search engine implementation for Google.
-    Parses Google Search HTML results.
-    """
-    
-    def build_url(self, query: str, limit: int = 10) -> str:
-        encoded_query = urllib.parse.quote(query)
-        return f"https://www.google.com/search?q={encoded_query}&udm=14"
-
-    def parse(self, content: str) -> List[Dict[str, Any]]:
-        results = []
-        seen_urls = set()
-        
-        # Google search results are in blocks with class="MjjYud" or similar containers
-        # Split by result blocks first for more accurate extraction
-        
-        # Method 1: Split by common result block classes
-        block_patterns = [
-            r'<div class="MjjYud"[^>]*>',
-            r'<div class="tF2Cxc"[^>]*>',
-            r'<div class="g Ww4FFb"[^>]*>',
-        ]
-        
-        blocks = [content]
-        for bp in block_patterns:
-            new_blocks = []
-            for block in blocks:
-                parts = re.split(bp, block)
-                new_blocks.extend(parts)
-            blocks = new_blocks
-        
-        for block in blocks:
-            if len(block) < 100:
-                continue
-                
-            # Find URL in this block - prefer links with h3 nearby
-            url_match = re.search(r'<a[^>]+href="(https?://(?!www\.google\.|google\.|webcache\.googleusercontent\.)[^"]+)"[^>]*>', block)
-            if not url_match:
-                continue
-                
-            url = url_match.group(1)
-            if url in seen_urls or self._should_skip_url(url):
-                continue
-            
-            # Find h3 title in this block
-            h3_match = re.search(r'<h3[^>]*>(.*?)</h3>', block, re.IGNORECASE | re.DOTALL)
-            if not h3_match:
-                continue
-                
-            title = re.sub(r'<[^>]+>', '', h3_match.group(1)).strip()
-            if not title or len(title) < 2:
-                continue
-            
-            seen_urls.add(url)
-            
-            # Extract snippet from VwiC3b class (Google's snippet container)
-            snippet = ""
-            snippet_match = re.search(r'<div[^>]*class="[^"]*VwiC3b[^"]*"[^>]*>(.*?)</div>', block, re.IGNORECASE | re.DOTALL)
-            if snippet_match:
-                snippet = re.sub(r'<[^>]+>', ' ', snippet_match.group(1)).strip()
-                snippet = re.sub(r'\s+', ' ', snippet).strip()
-            
-            # Fallback: look for any text after h3
-            if not snippet:
-                # Try other common snippet patterns
-                alt_patterns = [
-                    r'<span[^>]*class="[^"]*aCOpRe[^"]*"[^>]*>(.*?)</span>',
-                    r'<div[^>]*data-snc[^>]*>(.*?)</div>',
-                ]
-                for ap in alt_patterns:
-                    am = re.search(ap, block, re.IGNORECASE | re.DOTALL)
-                    if am:
-                        snippet = re.sub(r'<[^>]+>', ' ', am.group(1)).strip()
-                        snippet = re.sub(r'\s+', ' ', snippet).strip()
-                        break
-            
-            # Extract images from this block
-            images = []
-            # Pattern 1: Regular img src (excluding data: and tracking pixels)
-            # Note: gstatic.com/images/branding is logo, but encrypted-tbn*.gstatic.com are thumbnails
-            img_matches = re.findall(r'<img[^>]+src="(https?://[^"]+)"', block)
-            for img_url in img_matches:
-                # Decode HTML entities
-                img_url = img_url.replace('&amp;', '&')
-                # Skip tracking/icon/small images (but allow encrypted-tbn which are valid thumbnails)
-                if any(x in img_url.lower() for x in ['favicon', 'icon', 'tracking', 'pixel', 'logo', 'gstatic.com/images/branding', '1x1', 'transparent', 'gstatic.com/images/icons']):
-                    continue
-                if img_url not in images:
-                    images.append(img_url)
-            
-            # Pattern 2: data-src (lazy loaded images)
-            data_src_matches = re.findall(r'data-src="(https?://[^"]+)"', block)
-            for img_url in data_src_matches:
-                img_url = img_url.replace('&amp;', '&')
-                if any(x in img_url.lower() for x in ['favicon', 'icon', 'tracking', 'pixel', 'logo']):
-                    continue
-                if img_url not in images:
-                    images.append(img_url)
-            
-            results.append({
-                "title": title,
-                "url": url,
-                "domain": urllib.parse.urlparse(url).hostname or "",
-                "content": snippet[:1000],
-                "images": images[:3]  # Limit to 3 images per result
-            })
-            
-            if len(results) >= 15:
-                break
-        
-        total_images = sum(len(r.get("images", [])) for r in results)
-        logger.info(f"GoogleEngine parsed {len(results)} results with {total_images} images total.")
-        return results
-    
-    def _should_skip_url(self, url: str) -> bool:
-        """Check if URL should be skipped."""
-        skip_patterns = [
-            "google.com",
-            "googleusercontent.com",
-            "gstatic.com",
-            "youtube.com/watch",  # Keep channel/playlist but skip individual videos
-            "maps.google",
-            "translate.google",
-            "accounts.google",
-            "support.google",
-            "policies.google",
-            "schema.org",
-            "javascript:",
-            "data:",
-            "#",
-        ]
-        
-        for pattern in skip_patterns:
-            if pattern in url.lower():
-                return True
-        
-        # Skip very short URLs (likely invalid)
-        if len(url) < 20:
-            return True
-        
-        # Skip URLs that are just root domains without path
-        parsed = urllib.parse.urlparse(url)
-        if not parsed.path or parsed.path == "/":
-            return True
-            
-        return False