Overlay v6 annotations on the original PDF (preserve A4/source page size)

[williamjameshandley]

Summary

On the v6_with_rmscene branch, v6 (.rm software 3+) annotations are placed on the rendered PDF using a scale derived from the stroke bbox extent and an x-offset of RM_WIDTH / 2. For overlays on real source PDFs, this puts margin annotations off the page — the rm-scene canvas is referenced against rm screen height (1872), not width, and the natural offset is RM_HEIGHT / 2.

Output is also stretched to rm-canvas dimensions, distorting non-rM-aspect PDFs (e.g. A4 portrait, 595 × 842).

This PR keeps the v5 path unchanged and adds a v6 branch that overlays annotations on the source PDF at its native size using the existing rescale_parsed_data helper.

Changes (single file, +12 / -17)

remarks/remarks.py:

• Page is created at pdf_src[page_idx].rect (source PDF size), not the rm canvas. The source page is copied 1:1.
• In the rescale block, V6 takes a new branch:

  scale = dims.width / RM_HEIGHT
  offset_x = (RM_HEIGHT / 2) * scale
  ann_data = rescale_parsed_data(ann_data, scale, offset_x, offset_y)

• V5 logic (the existing extent-derived scale branch) is wrapped unchanged in else:.

drawing.py is untouched.

Why this transform

rm-scene points have x centred on 0 with the canonical canvas extending ±RM_HEIGHT/2 (the rm screen-height units in which the device represents portrait pages). The linear map back to source PDF coords is out = in * (pdf_w / RM_HEIGHT) + (RM_HEIGHT / 2) * (pdf_w / RM_HEIGHT) for x, and out = in * (pdf_w / RM_HEIGHT) for y, which is exactly the scale * point + offset shape that rescale_parsed_data already implements.

Test plan

• python -m remarks <dir> <out> on a v6 .rmdoc whose source PDF is portrait A4 → output is A4 (pdfinfo shows 595 × 842).
• Annotations land within the A4 page bounds, aligned with the content they were drawn next to on the device.
• Verify against landscape PDF and rM-native-aspect notebook cases.

🤖 Generated with Claude Code