Review: the drag began from the stored `offset` as read element by element, so a stored
[5, 0] (shown at 4) sat at 4 until the pointer had moved a layer's width, a malformed
[0.5, null] (which the engine ignores) began at [0.5, 0], and a press on a gradient without
an offset wrote [0, 0]. The drag now begins from what the renderer shows: the stored pair when
it is two finite numbers, clamped to ±4, else the effect's own offset from its `fx` snapshot,
else the centre; and a press that moves nothing writes nothing, stored or not.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
#10642 hides the frame after a committed crop until the next press or
re-picking the tool. The #9855 and #10453 tests expected the default frame
right after the commit; they now check that it is hidden, pick the tool
again and then test the default frame as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Review: "ŠKODA" (a capital before an ASCII capital) counted for nothing and could read as Mac
Roman "äKODA"; Mac Roman "Layer 90°é" read as windows-1251 "90ЎЋ", two Cyrillic letters right
after a number. The word-initial rule now takes an ASCII letter of either case after the capital,
and `cyrillic_word` treats a digit before the run as glue (a digit after still isn't: "Слой1").
Main's new fallback test tied Mac Roman "Ändern" with windows-1251, which this branch settles for
Mac Roman; it ties "éäé" with "ЋЉЋ" instead and checks "Ändern" is certain.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Review follow-up: the failure status and notice keep the Saved file name, so with several documents or a late background save the user sees which one saved fine and whose assets failed.
A warp pulled far enough folds the picture over itself. Warp mode's
preview stacked the fold's layers by how far the warp moved them, but
applying it kept one source point per pixel, last cell in row order, so
what Enter applied could differ from what the preview showed, and where
the front layer was transparent (between brush strokes) the part behind
it was lost.
Edit > Transform > Warp and smart object rendering now order the cells
by how far the warp moved them (from where they'd be without it: the
identity, or a smart object's placement), least moved first, so the part
dragged furthest is on top in every direction, as row order mostly had
it. The preview uses the same `cell_order`. A fold's layers composite,
also when they are under a source pixel apart. The layers under the top
one are read from their own source window, within the one size cap, so
a fold joining distant parts can't make a huge read, and a top layer
over an empty part of the source still shows what's under it. The
layer's linked mask, a smart object's filter mask and the selection
stack a fold the same way, so they stay with the pixels. Other mesh maps
(lens corrections, distortions) and Puppet Warp are unchanged:
`warp_mesh_surface_with` and the new `warp_mesh_gray_with` take the
stacking as a `Fold`.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01EZbgszHyuEuWAzxaamteuU
The Eraser only clears alpha, so a layer erased completely keeps its
colour bytes and content_bounds still found content. Empty now looks
for a pixel with alpha > 0, as Reveal All does (#9379).
A control command can clear the selection while the dialog is open. OK then ran Equalize on the whole layer under a choice that names a selection. The dialog stays open until the selection is back or the dialog is canceled.
With Shift held, the box now turns in 15 degree steps from the angle it
started the drag at. A box at 25.59 degrees dragged by about 20 degrees
ends at 40.59, not at an absolute multiple of 15. From 0 degrees the
results are the same as before.
Fixes#11099
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
A click without a drag outside the box now leaves the box open and
unchanged when it lands on the image. A click on the pasteboard (outside
the document) still commits as one Free Transform step. A double-click
commits only on the box (handles included); a double-click outside it does
nothing. Dragging outside still rotates.
Fixes#11100
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
The corner scaled by the larger of the two axis factors, so a drag along
one axis did nothing until the pointer also crossed the other axis. The
factor is now the projection of the pointer offset onto the diagonal from
the anchor (the opposite corner, or the reference point with Alt) to the
dragged corner, measured in the box's own lengths. Shift (free scaling),
the legacy mapping, edges and the flip past the anchor are unchanged.
Fixes#11096
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
equalize_step_with_a_selection_opens_its_choices now confirms the paused
step's dialog with selectedAreaOnly set to false and checks that a pixel
outside the selection changes.
Resolve the parity-doc conflict: keep main's status line and entries, re-add this PR's entries.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Resolve the parity-doc conflict: keep main's status line and entries, re-add this PR's entries.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Resolve the parity-doc conflict: keep main's status line and entries, re-add this PR's entries.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Resolve the parity-doc conflict: keep main's status line and entries, re-add this PR's entries.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>