fix: improve PDF layout generation for Direct track

Key fixes:
- Skip large vector_graphics charts (>50% page coverage) that cover text
- Fix font fallback to use NotoSansSC for CJK support instead of Helvetica
- Improve translated table rendering with dynamic font sizing
- Add merged cell (row_span/col_span) support for reflow tables
- Skip text elements inside table bboxes to avoid duplication

Archive openspec proposal: fix-pdf-table-rendering

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>
This commit is contained in:
egg
2025-12-03 14:55:00 +08:00
parent 08adf3d01d
commit 1b5c7f39a8
5 changed files with 405 additions and 111 deletions

View File

@@ -0,0 +1,68 @@
# Design: Fix PDF Table Rendering
## Context
OCR track produces tables with:
- `cell_boxes`: Accurate pixel coordinates for each cell border
- `cells`: Content with row/col indices and row_span/col_span
- `embedded_images`: Images within table cells
Current implementations fail to use these correctly:
- **Reflow PDF**: Ignores merged cells, misaligns content
- **Translated Layout PDF**: Creates new Table object instead of using cell_boxes
## Goals / Non-Goals
**Goals:**
- Translated Layout PDF tables match untranslated Layout PDF quality
- Reflow PDF tables are readable and correctly structured
- Embedded images appear in both formats
**Non-Goals:**
- Perfect pixel-level replication of original table styling
- Support for complex nested tables
## Decisions
### Decision 1: Translated Layout PDF uses Layered Rendering
**What**: Draw cell borders using `cell_boxes`, then render translated text in each cell separately
**Why**: This matches the working approach in `_draw_table_with_cell_boxes()` for untranslated PDFs
```python
# Step 1: Draw borders using cell_boxes
for cell_box in cell_boxes:
pdf_canvas.rect(x, y, width, height)
# Step 2: Render text for each cell
for cell in cells:
cell_bbox = find_matching_cell_box(cell, cell_boxes)
draw_text_in_bbox(translated_content, cell_bbox)
```
### Decision 2: Reflow PDF uses ReportLab SPAN for merged cells
**What**: Apply `('SPAN', (col1, row1), (col2, row2))` style for merged cells
**Why**: ReportLab's Table natively supports merged cells via TableStyle
```python
# Build span commands from cell data
for cell in cells:
if cell.row_span > 1 or cell.col_span > 1:
spans.append(('SPAN',
(cell.col, cell.row),
(cell.col + cell.col_span - 1, cell.row + cell.row_span - 1)))
```
### Decision 3: Column widths from cell_boxes ratio
**What**: Calculate column widths proportionally from cell_boxes
**Why**: Preserves original table structure in reflow mode
## Risks / Trade-offs
| Risk | Mitigation |
|------|------------|
| Text overflow in translated cells | Shrink font (min 8pt) or truncate with ellipsis |
| cell_boxes not matching cells count | Fall back to equal-width columns |
| Complex merged cell patterns | Handle simple spans, skip complex patterns |
## Open Questions
- Should reflow PDF preserve exact column width ratios or allow ReportLab auto-sizing?
- How to handle cells with both text and images?