* Render appearance: click, drag and paint to assign, Pick Up, copy and paste, and Remove
* Exclude the workspace Shots strip from dialog width checks
* Exclude the collapsed Shots control from dialog width checks
* Retain assembly cache listings and measure actual dialog bounds
* fix: handle toast button rectangles in assignment status
* fix: restrict gallery note measurement helper to tests
* test: measure assignment help at the catalogue footer
* fix: share the command revision snapshot after restacking
* claim: #7445
* Color picker: add shared Hex input mode (#7445)
Accept strict RGB/RGBA Hex input while preserving exact opacity,
opaque/PBR restrictions, invalid drafts and existing U8/F behavior.
Keep explicit English fallback entries for the three new Hex messages
in Arabic and Hebrew, which final integration main now requires as
complete catalogs. No Arabic or Hebrew translations are claimed.
The user authorized this feature-only commit after full cargo xtask ci
stopped at the pre-existing geometry nonminimal_bool lint. The same
failure was reproduced without this patch at the published claim and
main. No unrelated lint fix or suppression is included.
* i18n: cover complete catalogs for Hex picker (#7445)
Add the three Hex message keys to Vietnamese in the feature's pending
locale file. Keep the unchanged English source values as explicit
fallbacks; no Vietnamese translations are claimed.
Audit the complete-catalog set (ar, he, ja, vi, zh) rather than only
the first catalog reported by CI. Preserve the existing completeness
test and all picker/Appearance code.
The completeness regression changes from red to green. Full local CI
and wasm checks remain blocked by failures independently reproduced
in clean upstream main; no unrelated fixes or suppressions are added.
* Appearance: use the shared opacity-aware color picker in Render (#7445)
Route the merged Render Appearance editor through the UI-kit picker,
passing exact opacity only for non-PBR looks. Keep a non-editable mixed
label and treat explicitly accepting the first color as a mixed edit.
Remove the external Hex field and duplicate parser, replace its gallery
entry with the shared mixed-aware picker, and update the control contract.
Preserve existing regression assertions and add actual Render UI tests
for alpha edits, precision, invalid drafts, PBR rejection and mixed edits.
Full local CI and wasm were run before this commit and remain blocked
by the separately reproduced unchanged upstream geometry lints and
prefs.bytes wasm error. No unrelated fixes or suppressions are included.
---------
Co-authored-by: mynoc <298906611+mynoc-96@users.noreply.github.com>
* claim: #8086
* Configuration table: a filter in the Add Column menu (#8086)
The + menu listed every parameter, feature suppression and body
material/appearance with no way to narrow it. A filter field now sits at
the top, takes focus when the menu opens, narrows the entries and the Face
Appearance submenu as typed, picks a lone match on Enter and clears when
the menu closes. A click into the field keeps the menu open (plus_header
closes on outside clicks; entries close it themselves), and the list keeps
its full height after the filter has emptied it once.
New kit element ui_kit::menu_filter (glyph, frameless editor, rule) so the
field does not read as an entry; gallery, i18n guard and docs/ui-kit.md
updated. Regression test in manage_dialogs_tests.
* claim: #6756
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Windows: never load a dxcompiler.dll found on PATH; FXC unless DXC ships beside the app (#6756)
wgpu's Dx12Compiler::Auto loads dxcompiler.dll by name, so Windows takes the
first one on PATH. SolidCraft ships none, and DXC 1.8.2505 from other programs'
folders (Obsidian, NVIDIA SDK Manager) crashed every start with
EXCEPTION_ILLEGAL_INSTRUCTION during the viewport shader compile.
solidcraft_gpu::dx12_compiler() now chooses: WGPU_DX12_COMPILER as wgpu reads
it when it names a compiler; otherwise DXC from the dxcompiler.dll beside the
program, by its full path, else FXC. The window, the headless/CLI instance and
the test helpers start with it, and dx12_compiles_with_fxc follows it.
WGPU_DX12_COMPILER=dxc keeps the old lookup.
Fixes#6756
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Sketch: points tied only through a removed curve's end stay tied
Removing a curve removed its orphaned end points with their constraints, so
two points made coincident with such an end lost their tie although they
still met: two rectangles' corners on one arc end came apart when the arc
was trimmed. They are now tied to each other before the point goes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Sketch: keep the tie test with the other trim tests
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Naming a face swept from a sketch flattened every curve of the sketch to
a polyline and measured its distance to each, for every face. On a sketch
with thousands of curves (a sheet with many cut-outs) that grew with
curves times faces. The curves are now flattened once per feature, each
with its box, and a curve whose box is farther than the best distance so
far is not measured.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Spanish (es-ES): catch up with en.json
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Navigation preferences: name the Shift key through the catalog
Spanish now translates the key name, so the page painted it in English.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: bherbruck <brennen.herbruck@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #7356
* Orbit about the cursor: on an edge or sketch line, turn about the point under the cursor (#7356)
With Orbit around: point under the cursor, a drag that starts on a body
edge turned about the edge's middle, which on a long edge is far from
the cursor, and one that starts on a sketch curve turned about the
centre of the bodies in view. Edge and sketch curve hits now carry the
point under the cursor, which the pick already measured, and
pivot_under uses it. An edge hit's `mid` is unchanged, so selections
still record the edge's middle.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #8139
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019FAWPuFoh9qxn2uS2j53QZ
* 3MF export: a component's bodies as one object with parts (#8139)
file.export and file.save_mesh wrote every body as its own 3MF object and
build item, so a slicer loaded a multi-body part as separate objects and
could arrange them apart on the bed. With objects: "component", each placed
component that holds two or more bodies is written as one object whose
<components> are its bodies' meshes, built once, so PrusaSlicer, OrcaSlicer,
Bambu Studio and Cura load it as one object with a part per body, each of
which can take its own filament. The design's own bodies, and a component
with a single body, stay objects of their own. The default (objects: "body")
writes the same file as before.
- io: model_xml_parts and write_3mf_parts write the parts; export_with_parts
maps body names to them. model_xml_with and export_with_quality give the
same output as before.
- engine: mesh_parts reads `objects` (component is refused for formats
other than 3MF); component_parts groups the placed bodies by placement,
each part named after its occurrence (Bracket:1, Bracket:2).
- file.save_mesh passes the parts on when it also sends the file to a slicer;
file.send_to_slicer's own folder export is unchanged.
Fixes#8139.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019FAWPuFoh9qxn2uS2j53QZ
---------
Co-authored-by: Claude <noreply@anthropic.com>
* Render wear: a Wear row and editor for the aging layers
* test: read Wear header text from accessibility label values
* test: restore body picks for Wear header recovery assertions
* Render environment editor: rotation, intensity, backdrop colour and ground through render.environment.set
* Render environment editor: rotation ring in the view with one undo step per drag
* Render environment: Source, orientation, backdrop, ground and atmosphere sections
* Keep segmented controls inside their allocated rows
* fix: compile legacy environment controls only for tests
Where several curves cut at one place, trim pinned the new end onto the
first of them. At the touch point of two circles that was a rectangle side
passing through, and once a later trim removed the side the arc ends were
loose: a trimmed dog bone kept two untied ends.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Sketch Coincident: refuse a line's two ends
Making a line's two ends coincident shrinks it to a point; a Parallel or
Perpendicular on it afterwards has no direction to hold and a later solve
could throw points far away while reporting solved, as with a line's own end
as its midpoint (#8562).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Sketch Coincident: keep the own-ends test with the other Coincident tests
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #3693
* Fix timeline playback to show each rebuilt feature
* Move History playback into the Inspect ribbon
* Refuse Play History up front when playback couldn't run
Starting Play and the per-frame check that stops it were looking at
different app state, so Play could be accepted during an apply, a delete
check or queued history work and then stop silently on the next frame.
Both now use one playback_blocked check.
The UI also reads the engine's PlaybackAction as the typed enum instead of
matching strings, so a bad reply is an error rather than a quiet toggle.
* Put Play History in the History controls
History now shows five controls in transport order: start, previous,
play, next, end. Play runs timeline.play, so it turns into Stop while
playback runs, same as the ribbon button and the shortcut.
It's on by default (#3688, #3740, #4216). Anyone who wants the old four
controls back can right-click them and pick Hide Play Button, or clear
history_play_button under Preferences > Interface. In a named vertical
History too narrow for five buttons beside its grip, Play is left to the
ribbon rather than overlapping the grip.
* Draw the History Play button as a plain play/pause glyph
The Play button in the History controls now uses the same drawn glyphs as
its neighbours: a filled triangle, and two bars while History plays. The
triangle is bigger than the step arrows so it doesn't read as Forward.
The ribbon command keeps its timeline_play and timeline_stop icons.
* Keep the playback changes clear of PRs queued ahead
Several PRs in the merge queue touch the same lines, so this moves ours
aside without changing behaviour:
- history_play_button and its reset-test row sit with the other History
preferences, not beside view_confirm_corner (#7248 adds status_bar there).
- History keeps main's history_transport_width() for four controls (#6430
and #7459 call it); the Play button's count uses history_controls_width.
- A new or opened document stops playback after dim_view::cancel(), clear
of the line #7819 adds after reset_chat.
- The Play/Stop label swap moves from interactive_command_info, which #7906
rewrites, into the ribbon's command_info.
* Stop playback after the command completes, clear of #6433#6433 rewrites the session.execute line in execute_command, and our playback check sat right under it, so the queue couldn't merge the two. Checking the revision after command_completed does the same job a couple of lines later.
---------
Co-authored-by: cfaulkingham <colin.faulkingham@gmail.com>
* claim: #7834
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Viewport: vertex markers where edges are drawn (#7834)
Each body's vertices (the ends of its pickable edges, once each) show as
small screen-space squares in the edge colour in the styles that draw
edges: shaded with edges, wireframe, and shaded with hidden edges, where
hidden vertices show faintly like hidden edges. Plain shaded draws none.
A marker is a lifted line from the vertex to itself in the body's block,
so it is instanced with the edges, cached with them, and depth-tested
like a highlight band. The existing vertex hover ring stays on top.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Viewport tests: count edge segments apart from vertex markers (#7834)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Vertex markers: an option, off by default (#7834)
Adds the Vertex markers preference (Preferences > Graphics, off by default) and draws the dots only when it is on.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: bherbruck <brennen.herbruck@gmail.com>
Open refused a saved model whose text records (a feature's warning, a body or
sketch name) held more than 65,536 bytes, although Save writes them, so such a
design was evaluated again on every open.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Viewport: cut dashed lines to the view, so a huge distance can't fill the memory
epaint makes one shape per dash and steps along a line in f32. The
viewport's dashed lines (the Extrude taper handle, Measure's legs, the
fillet radius guide, the inference and snap guides, a sketch text's
frame) were dashed over their whole projected length. With an Extrude
distance of 1e9 mm the taper line was billions of points long: the
frame never finished and filled the memory within a second.
`viewport::dashed_line` dashes a line that lies in view exactly as
before, and cuts one that reaches far off screen to the view (grown by
two dash periods) first, keeping its dashes in phase with its start.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Viewport: compute a cut dashed line's ends in f64
A line a trillion points long lost its cut ends in f32: they collapsed to one point, and the part in view drew nothing. The new test caught it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Two new commands, view.next_tab and view.previous_tab, with Ctrl+Tab and
Ctrl+Shift+Tab as their default keys: they appear on the Keyboard Shortcuts
page and can be rebound or cleared like any other. The window steps along the
tab strip and goes round at either end; one design alone stays as it is, and
a "save changes?" prompt keeps its tab shown. The engine only echoes the
step (no design edit, no undo step); engine.batch refuses them like the
other steps that go to another design.
Co-authored-by: Phanoob <phanoob@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #5066
* Pattern counts: name a length-unit parameter, refuse fractional counts (#5066)
A count parameter created with a length or angle unit now gets an error
that names it and says to set its Unit to none. A count that evaluates to
a fraction is refused instead of being rounded silently.
A mesh body's row (an imported STL, 3MF, OBJ, ...) now shows a new
mesh_body icon, a triangulated cube in the mesh accent, instead of the
solid body's grey cube. Hovering a body row says Mesh body, Surface body
or Solid body. Body::is_mesh() decides; nothing new is saved.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #7744
* Vendor winit 0.30.13 with Wayland touchpad pinch and rotation gestures
The 0.30 Wayland backend never binds zwp_pointer_gestures_v1, and
WindowEvent::PinchGesture / WindowEvent::RotationGesture only come from the
macOS and iOS backends, so a trackpad pinch produces no input at all on
Linux and Wayland (two-finger scroll still works, since X11/XWayland
reports it as a wheel). egui-winit already maps the events to
egui::Event::Zoom and navigation::handle already zooms the camera from
that, so the Wayland backend was the only missing piece.
Upstream the gesture support is part of winit 0.31 (rust-windowing/winit#4338),
which eframe 0.36 cannot use while it still requires winit 0.30, and the
0.30 backport rust-windowing/winit#4778 was closed unmerged. So ship the
same fix locally, upstream-style-applied and stripped of example and test
targets, with its unit tests for the gesture state machine kept under
cargo test -p winit --lib -- pointer_gesture:
- src/platform_impl/linux/wayland/seat/pointer/pointer_gesture.rs: bind
the global (versions 1..=3), one pinch gesture object per pointer,
destroyed before the pointer is released; begin/update/end become the
two events for the window the gesture began over. Two fingers only;
the protocol's scale is relative to the gesture's beginning, so the
event carries the change since the previous update; rotation is
negated to winit's counterclockwise degrees; a non-finite scale is
dropped, and so are updates of a gesture whose begin we never saw.
- seat state wiring: bind the global, track and drop the gesture object.
- wayland-protocols' unstable feature, platform notes on both events,
changelog, patch-section entry in Cargo.toml, vendor/README, and
Linux/Wayland notes in the troubleshooting guide and the UI tour.
Retire this vendoring once an eframe release consumes winit 0.31 or
later; the cleaned branch is kept as
https://github.com/purp0w/winit/tree/wayland-pinch-gestures.
* ci: retrigger after flaky occurrence-cache test
---------
Co-authored-by: Francisco das Chagas Danta Filho <dev@dantas-filho.me>
Parallel, collinear, equal or concentric of a curve with itself was added
as a constraint that always holds; perpendicular or tangent to itself was
refused as geometry that can't move. These six now refuse one curve
picked twice, as Coincident already does for a point, and leave the
sketch as it was. Constraints already in a design are untouched.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #7919
* Pick by the body's own size, not the camera's distance (#7919)
The occlusion test that decides whether an edge or vertex is hidden by the
face in front allowed a slack of `scene_radius(app) * 0.02`, and
`scene_radius` ends with `.max(app.cam.half_height() * 2.0)`: in an
orthographic view that is `0.016 x camera distance`. Zooming out grew the
slack without bound until a whole body fitted inside it and every edge behind
the front face counted as visible, and edges and vertices are offered before
faces, so they won the pick. Cutting the slack from the scene also gave a
small part in a large assembly a tolerance many times its own size.
`pick_slack` now takes a size and returns the mesh's own chord tolerance with
room to spare, `DisplayQuality::Coarse.chord(size) * 4.0`: a facet sits at
most a chord tolerance inside its surface, and that tolerance is already a
fraction of the body's size, so the slack no longer moves when the camera
does. Each body is tested with its own bounding box, which the loop already
holds. `view.pick` does the same, so an agent's pick still matches a click on
the same pixel. Sketch profiles and origin planes belong to no body and keep
a scene-level value, `scene_pick_slack`.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
A line's own end held at its middle shrinks the line to a point. The
solve accepted that, and a later angle, tangent or parallel on the line
had no direction to hold: its points went millions of millimetres away
while the sketch still reported solved. MidPoint now refuses the pick,
saying why, and leaves the sketch as it was. Constraints already in a
design are untouched (only the command refuses).
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #4750
Co-Authored-By: Claude <noreply@anthropic.com>
* Extend: pick two curves to extend both to where they meet (#4750)
`sketch.extend` gains a second form: `curves: [a, b]` names two lines or
arcs, with an optional `at: [[x,y],[x,y]]` naming the end of each to grow
(default: the end nearest the meeting point). Both grow along their own
line or circle to where they cross and the two ends are joined with one
Coincident, in one undo step. `moved` is the joined end's id.
- Where two crossings are possible (line and arc, two arcs) the one with
the least total growth is used; a tie is refused.
- A curve that already reaches the point does not grow. An end held by
other geometry or a constraint stays and gets a new line or arc.
- Parallel or collinear lines, curves that never meet, a point behind the
chosen end, the same curve twice, splines, ellipses, infinite lines,
offset geometry and malformed `curves`/`at` are refused with no change.
- The single-curve form is unchanged; its code is shared through
`grow_line` / `grow_arc`.
Canvas: in the Extend tool, Shift-click two curves. Shift goes through
the tool modifier registry as the new `extend_pair` action, and Extend
turns off Shift's midpoint snap. Escape drops a pending first pick. The
sketching guide describes the gesture.
Co-Authored-By: Claude <noreply@anthropic.com>
---------
Co-authored-by: Claude <noreply@anthropic.com>
* claim: #7893
* Toasts: one stack in the view's bottom-right corner, one width, buttons as in dialogs (#7893)
* Control protocol: toast_place in the sorted ui.prefs list (#7893)
* Toasts: a sketch test clears the status notice before clicking the view's corner (#7893)
The test's line end lands in the bottom-right corner, where notices now stack
and take clicks as they did at the bottom centre; the test is about the circle.
* Toasts: the pattern scenario's circle sits clear of the view's corner (#7893)
After the refit, a circle at y 0 landed in the bottom-right corner, where the
tool's prompt now stacks and takes clicks. The circle moves 40 mm up; the
pattern it checks is the same.
* claim: #8358
* Components: a component named like the design is reachable by name (#8358)
Document::find_component resolved the design's own name to the root component (id 0) before
searching the component list, so a component that shares the design's name was unreachable by
name: appearance.assign reported `no component `CurtainRailPlug`` (the same lookup serves every
command that takes a component by name).
An exact component name (or component id) match now wins over the design-name fallback; "0" and
"root" still always mean the design.
Regression test: a_component_named_like_the_design_takes_its_appearance_by_name (fails on main
with the reported error, passes with the fix).
Nathan Dane