The picker shows the Brush Settings panel's angle/roundness widget left
of its Size and Hardness block, the same widget made reachable from the
picker. Its drags go through tools.setBrush as one journaled change
each, and Escape reverts an angle or roundness edit as it does a Size
or Hardness edit. A pointer position that isn't finite leaves the tip
as it was.
The sampled colour is read from stored pixels, but the foreground was compared unrounded, so in
8- and 16-bit documents a foreground between two levels missed its own stored colour: Find Edges
erased it, and a foreground stored as the sample no longer protected anything. The foreground is
now rounded to the document's sample type first (32-bit float unchanged).
protect_off_is_unchanged now pins the whole alpha row, not only the first and last block.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`PhotocraftApp::run` puts what a command answered into the status line, but a refusal raised
before the command runs - the automation gate a control or MCP-driven click passes through -
returns from `run_command_at_depth` ahead of that report, and the button handler only reacted
to the hidden-layer alert. A denied synthetic click therefore did nothing at all, which reads
as a dead button rather than a refused one. It now reports any error it gets, the way
`eraser_ui` and the toolbar flyout do, and keeps opening the reference's alert for the hidden
layer alone.
`Session::edit` keeps the selected layers valid after an edit but not the active id, so an
edit that drops the active layer leaves one behind that the document does not hold. The new
enabled check read that id and answered "live", and with an explicit `crop` the command then
rotated and cropped the whole canvas for a layer that is not there - the one outcome of this
state that changes pixels. Refusing it is what `select_layer` and `restore_layer_selection`
already do with a gone id, and it costs the resolve check the target lookup was doing anyway.
No command shipped today drops the active layer without re-targeting, so no real document
changes behaviour: this is the guard the review asked for, and the test reaches it through
`Session::edit` directly.
The filmstrip that main added while #9991 was open holds raws opened together.
Only the shown raw read its sidecar crop, so a raw that was never shown, or was
shown and left, opened uncropped. The crop is now read when the raw is queued.
Co-Authored-By: ljkx <177671075+ljkx@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Conflicts with main's --no-maximize-compatibility (#8064) in the CLI
export args, output mapping and dev-dependencies, and in the
file-format parity status line and revision history: both sides kept.
Co-Authored-By: ljkx <ljkx@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
#11795 didn't measure where the popup sits after a rotation, so that case keeps today's placement
(the top-left of the mapped rectangle) and only the size rule applies. Image Size and Canvas Size
still keep the popup's offset from the icon.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
PSD export writes its default popup for an all-zero one; moving it with the icon turned it into
a 0 x 0 rectangle elsewhere, which export then wrote as is.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
CodeRabbit on #9396: with both ends of a short open path in reach (as when
zoomed out), the canvas showed the continue-path pointer, but the press
resumed the path and then hit its first anchor, so the release closed it.
The press that resumes a path now never closes it, and the nearer end is
the one carried on. The path a Pen path carries on is now borrowed instead
of cloned, so the hover check no longer copies the whole path each frame.
Co-Authored-By: Leel Wasanatha <260480244+kavishkashehan726-sudo@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Importing an .abr with select picked its first brush without the 1 %
opacity floor, so a preset saved at 0 % showed 1 % but painted nothing
(#11812).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A brush remembered at 0 % opacity (saved before the 1 % floor) read back
at 0: the options bar showed 1 % but strokes painted nothing. Restoring
now floors opacity at 1 %, like tools.setBrush (#11812).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
#9970 reopened Image > Adjustments > Photo Filter... at the filter, colour,
density and Preserve Luminosity of its last OK. Measured in the reference
application: after OK at 60, 70 and 74 %, with a preset or with Color and a
custom colour, it reopens every time at Warming Filter (85), the orange
swatch and 25 %; only Preserve Luminosity carries over (off after an OK with
it off, on after an OK with it on). Opening now starts from the neutral
settings and takes just that one value from the last OK; Alt still reopens
at all the last settings (#8900). Tests and the ui-parity row follow the
measured behaviour.
Co-Authored-By: ljkx <177671075+ljkx@users.noreply.github.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The typed-text draft remembers the setting it was stored with and is only
used while that setting is unchanged, so a value set meanwhile (a tool
preset while the options bar was hidden) shows from the first frame.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
pointer takes the held modifiers. A new drag with Shift draws a square
of the shorter drag side, with Alt a box centred on the press at twice
the drag, with both a centred square; the live box follows the
modifiers. Shift moves a dragged corner along the axis it mostly
follows; Alt keeps a rectangle one, the opposite corner fixed. Ctrl
and no modifier keep today's behaviour.
After two clicks the preview is the two points with the pointer as the
third and fourth corner, after three the pointer is the fourth; it is
drawn with its grid in the finished quad's style, and the clicked
corners are square handles. The canvas sends button-up moves to the
tool while corners are pending, as it does for the Magnetic Lasso.
While the field has the keyboard it gets back the value typed so far
instead of the clamped one, so a leading 0 on the way to 05 no longer
turns into 1 (and 15). The stored value is still clamped at once.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>