Geoff Youngs 173dd2607a Fix/rotation recalc compound transforms (#1083) ...

* Fix rotation recalculation corrupting compound transforms

Two bugs in the rotation center recalculation triggered by attribute
changes on rotated elements:

1. The recalculation fires for ALL attribute changes (stroke-width,
   fill, opacity, etc.) even though only geometric attributes (x, y,
   width, height, d, points, etc.) can change the bbox center. This
   causes unnecessary and incorrect rotation center updates.

2. In undo.js, the center is computed through ALL remaining transforms
   via transformListToTransform(tlist).matrix, but for compound
   transform lists like [translate(tx,ty), rotate(angle)], the
   pre-rotation translate leaks into the center calculation, producing
   rotate(angle, bcx+tx, bcy+ty) instead of the correct
   rotate(angle, bcx, bcy).

3. In history.js (undo/redo), the code replaces the ENTIRE transform
   attribute with just rotate(...), completely destroying any other
   transforms (translate, scale, etc.) in the list.

Fixes:
- Guard recalculation with a BBOX_AFFECTING_ATTRS set so non-geometric
  attribute changes skip the rotation block entirely.
- In undo.js, use transformListToTransform(tlist, n, max) to compute
  the center through only post-rotation transforms.
- In history.js, replace string-based transform replacement with
  transform list API that finds and updates only the rotation entry,
  preserving all other transforms. Extract shared logic into
  relocateRotationCenter() helper.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Add tests for rotation recalculation with compound transforms

Tests demonstrate and verify fixes for three bugs:

1. Non-geometric attributes (stroke-width, fill, opacity) no longer
   trigger rotation center recalculation — the transform list is
   left untouched.

2. Geometric attribute changes on elements with compound transforms
   (e.g. translate + rotate) compute the rotation center using only
   post-rotation transforms, preventing the translate from leaking
   into the center calculation.

3. ChangeElementCommand.apply/unapply preserve compound transform
   structure instead of replacing the entire transform attribute
   with just rotate(...).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Add visual demo and screenshot for rotation recalculation bug

Includes an HTML demo showing both bugs side-by-side:
- Bug 1: Non-geometric attribute changes (stroke-width) on compound
  transforms corrupt character positions (e.g., word art on curves)
- Bug 2: Rotation center computed through ALL transforms instead of
  only post-rotation transforms, causing translate to leak into center

Each panel shows Original / Buggy / Fixed comparisons with the actual
SVG transform math applied. A Playwright script generates screenshots.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Add sample SVG demonstrating compound transform corruption bug

A pinwheel of 6 colored rectangles using translate(200,200) rotate(N).
Loading this into SVG-Edit and changing stroke-width on the group
triggers the rotation recalculation bug on unfixed builds.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Fix standard linting: object properties on separate lines

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* Centralize BBOX_AFFECTING_ATTRS and fix bbox-before-remove safety

Address two code review comments:

1. Export BBOX_AFFECTING_ATTRS from history.js and import it in undo.js
   so the attribute list is defined in one place.

2. Compute getBBox() before calling tlist.removeItem() so that a falsy
   bbox causes an early return/break without having already destroyed
   the rotation transform.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

2026-03-15 22:28:53 +01:00

1.1 KiB

Raw Blame History