fix: Use alpha_composite for proper RGBA highlight blending

Drawing directly with ImageDraw on RGBA mode doesn't blend alpha properly. Use Image.alpha_composite() with a transparent overlay to achieve correct semi-transparent highlight fills. 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
fix: Remove pymupdf.layout.activate() to fix page_chunks behavior
2025-11-22 17:04:29 +01:00 · 2025-11-22 16:58:35 +01:00
3 changed files with 18 additions and 11 deletions
@@ -18,6 +18,7 @@ RUN uv sync --locked --no-dev --no-editable --no-cache

 ENV PYTHONUNBUFFERED=1
 ENV VIRTUAL_ENV=/app/.venv
+ENV PATH=/app/.vnev/bin:$PATH
 ENV TESSDATA_PREFIX=/usr/share/tesseract-ocr/5/tessdata

 ENTRYPOINT ["/app/.venv/bin/nextcloud-mcp-server", "--host", "0.0.0.0"]
@@ -6,15 +6,15 @@ import tempfile
 from collections.abc import Awaitable, Callable
 from typing import Any, Optional

+# NOTE: Do NOT call pymupdf.layout.activate() here!
+# It changes the behavior of pymupdf4llm.to_markdown() when page_chunks=True,
+# causing it to return a string instead of a list[dict].
+# See: https://github.com/pymupdf/pymupdf4llm/issues/323
 import pymupdf
-import pymupdf.layout
+import pymupdf4llm

 from .base import DocumentProcessor, ProcessingResult, ProcessorError

-# Activate layout analysis for better text extraction
-pymupdf.layout.activate()
-import pymupdf4llm  # noqa
-
 logger = logging.getLogger(__name__)


@@ -845,9 +845,8 @@ class PDFHighlighter:
                            logger.warning(f"Chunk {chunk_index}: could not find bbox")
                            continue

-                        # Copy base image and draw highlight using PIL (fast!)
+                        # Copy base image for this chunk
                        chunk_image = base_image.copy()
-                        draw = ImageDraw.Draw(chunk_image, "RGBA")

                        # Scale bbox coordinates to pixmap coordinates
                        scale_x = base_pix.width / page_rect.width
@@ -859,10 +858,17 @@ class PDFHighlighter:
                            int(bbox[3] * scale_y),
                        )

-                        # Draw semi-transparent fill
-                        draw.rectangle(pil_bbox, fill=fill_color)
-                        # Draw dashed border (PIL doesn't support dashes, use solid)
-                        draw.rectangle(pil_bbox, outline=pil_color, width=3)
+                        # Create transparent overlay for fill (proper alpha blending)
+                        overlay = Image.new("RGBA", chunk_image.size, (0, 0, 0, 0))
+                        overlay_draw = ImageDraw.Draw(overlay)
+                        overlay_draw.rectangle(pil_bbox, fill=fill_color)
+
+                        # Alpha composite the overlay onto the chunk image
+                        chunk_image = Image.alpha_composite(chunk_image, overlay)
+
+                        # Draw border on top (solid, not transparent)
+                        border_draw = ImageDraw.Draw(chunk_image)
+                        border_draw.rectangle(pil_bbox, outline=pil_color, width=3)

                        # Convert back to PNG bytes
                        output = BytesIO()