* Sketch: Break Link in the context menu of projected geometry (#10417)
Projected curves and points (model edges and vertices, another sketch's
curves) could only be unlinked through command search, although Trim and
other refusals say "use Break Link on it". The sketch context menu now
offers Break Link when the selection holds projected geometry; text keeps
Explode Text.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Sketch: test Break Link in a projected edge's context menu (#10417)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Browser: Features folders scrolled out of view skip their mark
A folded Features folder shows the worst error or warning mark of the
features it holds. Working that out scans the design's whole history, and
from 16 features builds a table of all of them: per placement, on every
frame, whether the row is on screen or not. In a long tree of placements of
a part with a long history this was most of the window's frame while
orbiting.
A row scrolled out of view paints nothing (`tree_row` only takes its space),
so its mark is now worked out only when the row is on screen, as the
placement rows already do for their own state. The rows on screen show the
same marks.
Refs #6083
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Scale tests: time each frame of an orbit around a large design
`scale_tests::orbit_frame_cost` (ignored, manual) turns the view with the
right button over a design (a STEP file inserted as the app opens one, a
design file, or the generated assembly) with the model scene built, as the
app builds it once a renderer is there. Per frame it times the window
thread's frame and its named steps (`frame_watch`), egui's tessellation,
and the renderer's prepare, draw recording and GPU work on an offscreen
device (`OffscreenFrames::frame_parts`); without an adapter it times the
window thread alone. `painted` fingerprints the tessellated picture once
the view rests, to compare two builds.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
On XZ a sketch's y runs along world -Z, so a value written as a height lands
below the plane; only the modelling skill's reference said so. The command's
parameter text now gives the axes of all three origin planes (XY: +X, +Y;
XZ: +X, -Z; YZ: +Y, +Z), and a test pins them.
Assisted-by: LLM
`component.insert_design` links a design saved as a .sc container or as plain
JSON (`.json`), but `component.relink` refused every path that was not a
container ("choose a saved .sc design"), so a component linked to a JSON
design could not be pointed at another one. Relink now takes any design path
Insert Design takes (`is_design_path`: .sc, a legacy extension or .json); the
file is read by the same `design_bodies`, which already reads both. The Relink
Source file dialog opens on a filter that lists .json among the designs, too.
Assisted-by: LLM
* claim: #10258
* Sketch: construction lines with a length (#10258)
The horizontal, vertical, angle, parallel and perpendicular construction
line commands take an optional `length`: the line becomes a construction
segment from `at`, running towards `near`, held by a length dimension
that follows its expression. Without `length` the line stays infinite,
as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Markus Kaltenbrunner <kaltenbrunnermarkus@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
- In entry_row (crates/ui-egui/src/browser.rs), use live rather than cached for the property
- Ensures eye toggle dynamically fades or unfades rows for sketches and bodies without requiring a cache rebuild
- Add regression test in browser_card_tests
Fixes#9953
OpenSteps splits doc.open of a design into steps for a caller that must
keep working meanwhile (the web app, which has one thread): start reads
the file, each step evaluates the next feature by moving a copy's history
marker one feature on (so every result before it is reused), and finish
puts the design in the session's place with doc.open's reply. The next
feature's imported STEP text is offered first, so a caller can read it
in parts. Session::opened and opened_at are split into small helpers so
both paths build the session the same way; doc.open is unchanged.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Saving a design with its model cache asked, for every state, for each sheet-metal body it holds,
and built that sheet's JSON record each time only to find it written already. A design of 906
states with 300 sheets spent most of its save there. The sheet's text written straight from it
is the key now; its record is built only the first time. The records and the file are the same.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Find a saved model's sheets by their text, not a record built per state
Saving a design with its model cache asked, for every state, for each sheet-metal body it holds,
and built that sheet's JSON record each time only to find it written already. A design of 906
states with 300 sheets spent most of its save there. The sheet's text written straight from it
is the key now; its record is built only the first time. The records and the file are the same.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Write the model cache manifest without copying the model's records
The manifest was built as a JSON tree, which copied the whole model's records (13 to 33 MB of
records in large designs) before writing them, and a cache too big for the file wrote it again for
each try at fewer states. The manifest is now written from the records where they are, its fields
in the sorted order the tree wrote them, so the file is the same byte for byte.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #10063
* docs: add a design approach and workflows chapter to the user guide (#10063)
Adds guide/design-approach.md after the window tour: the parametric idea, principles for designs that change well, the timeline and rollback marker, bodies vs components and occurrences, common workflows with links, a term map for users of other CAD software, and where to find help.
* docs: match the chapter to the app's labels and sketch colours (#10063)
Checked in the web app (0.29.0): the timeline strip is headed History, the toolbar shows New Sketch and Parameters for Create Sketch and Change Parameters, and fully constrained geometry turns white in the default dark theme rather than dark.
* docs: tighten the design approach chapter's wording (#10063)
The glossary covers most, not all, of the chapter's terms; Redo moves forward; the timeline is shown as the History strip; Report a bug drafts a report for the user to file and sends nothing itself.
A flat node's roundoff allowance took a hypot per control point. Twice
|x| + |y| bounds each hypot from above with room to spare for rounding,
and rounding keeps sums monotonic, so whenever the flatness test passes
with that larger allowance it passes with the exact one too. Only nodes
whose chord distance comes near the 0.5 px bound (or with a non-finite
term) now take the exact hypots, and they decide as before.
Same result for every node: the reference comparison test still matches
bit for bit.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
A caller that only needs the bodies' names (assembly occurrences included,
e.g. "Body2 (Part:2)") had to measure every body, or read document.inspect,
which lists definition bodies only. `inspect.measure {names_only: true}`
returns each world body's name and its definition's name, honouring the
`bodies` filter; an unknown name is refused before anything is measured.
The MCP `measure` tool passes it through, directly and as a batch step;
`summary` leaves such a list as it is, since there is nothing to narrow.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #8876
* Add optional viewport-only Kawase blur and native UI composition
* Fix vendored panel hook imports and deprecated float constant import
* Drain texture updates in the full GPU blur fixture
* Hold the blur evidence fixture camera fixed across radii
* Address blur compositor lint checks
* Share animated Render preview background with blur source
* Use resolved Design background for blur source
* Exercise blur rendering independently of software adapter policy in CI
* Use stock egui hosts instead of vendoring the full crate
* Preserve feature-window open handling with stock egui hosts
* Target the native Window in close-handling regression fixture
Show F2 beside Rename on context menus where F2 already renames the
target (bodies, features and sketches, construction planes, components and
occurrences, browser groups, and canvases). Saved views and joints remain
without a shortcut label because F2 does not rename them without a menu.
When a context menu is open, pressing F2 routes through press_shortcut to
rename the menu's target (such as a clicked browser group or canvas) and
close the menu, rather than renaming the viewport selection.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Gemini 3.8 Flash <noreply@google.com>
Each node of the display curve sampler recomputed a control point's
seven clip distances for every clip plane it tested (up to 7 times for
the reject test, 3 more for the depth test, once more for the cut), and
projected each point to the screen twice. They are now worked out once
per node into arrays and read from there.
Most nodes are split rather than drawn. A node is flat only when every
control point lies within the pixel bound of its chord after adding a
roundoff allowance that is never negative, so a point already farther
than the bound decides "split" on its own; the allowance (four hypot
and divisions per point) is now left out for those nodes.
Same values in the same order: the reference comparison test from the
previous commit still matches bit for bit.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
The doc comment of the ground shadow had slid onto build_sketch_scene when
that function was added above ground_shadow. Give it back to ground_shadow
and describe what build_sketch_scene builds.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Sketch Coincident: a selected line midpoint is a point it takes (#10311)
A selected midpoint goes to the command as {midpoint: line}, as
Horizontal/Vertical's alignment takes it, but Coincident wanted a point
id and refused it with "b must be a point or curve". The command now
puts a point at the line's middle, tied there by a Midpoint constraint
(or uses the one already tied there), and makes that point coincident
or puts it on the other curve, in one undo step. The toolbar button and
the context menu send a selected midpoint that way.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Sketch Coincident: test a line midpoint in the command, the toolbar and the context menu (#10311)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Engine: open a design a feature at a time (#10198)
OpenSteps splits doc.open of a design into steps for a caller that must
keep working meanwhile (the web app, which has one thread): start reads
the file, each step evaluates the next feature by moving a copy's history
marker one feature on (so every result before it is reused), and finish
puts the design in the session's place with doc.open's reply. The next
feature's imported STEP text is offered first, so a caller can read it
in parts. Session::opened and opened_at are split into small helpers so
both paths build the session the same way; doc.open is unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Web app: open a design a few features a frame, with progress and Cancel (#10198)
On the web, a design to open is now a staged file load like a STEP file
(#1376): its features are evaluated for about 30 ms a frame through
OpenSteps, a feature's imported STEP file is read a solid per frame first
and held so the feature finds it read, the progress card counts the
design's features, Cancel works between steps, and the meshes are made
a few per frame afterwards. The desktop app keeps its worker thread.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Web app: design links and the files window open designs with progress and Cancel (#10198)
Design links (?open=URL, as embedded viewers use), the files window and
designs opened from the computer went through the one-shot open_path,
which froze the page on a large design. They now use the staged open; a
design link becomes an unsaved copy under the link's name once the open's
reply arrives (SolveApp::open_path_answered).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Engine: open a design a feature at a time (#10198)
OpenSteps splits doc.open of a design into steps for a caller that must
keep working meanwhile (the web app, which has one thread): start reads
the file, each step evaluates the next feature by moving a copy's history
marker one feature on (so every result before it is reused), and finish
puts the design in the session's place with doc.open's reply. The next
feature's imported STEP text is offered first, so a caller can read it
in parts. Session::opened and opened_at are split into small helpers so
both paths build the session the same way; doc.open is unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Web app: open a design a few features a frame, with progress and Cancel (#10198)
On the web, a design to open is now a staged file load like a STEP file
(#1376): its features are evaluated for about 30 ms a frame through
OpenSteps, a feature's imported STEP file is read a solid per frame first
and held so the feature finds it read, the progress card counts the
design's features, Cancel works between steps, and the meshes are made
a few per frame afterwards. The desktop app keeps its worker thread.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
A sketch holding a text whose words come from a text parameter is left
unsolved after the words change and its links then move far, with the
warning "the sketch constraints conflict (k1, ..., k10)". The same two
changes the other way round solve.
Repro: a sketch with three projected points, a construction rectangle on
them and a centred text along its far side, the text's words given by a
text parameter. Change the parameter's words, then move the source
geometry about 280 mm. Without the word change it solves; with it, the
sketch warns of conflicting constraints.
Other words are other outlines: the text is built anew with new points
and curves. The solve that walks a sketch from its stored state to a far
link move (and the check for sides the jump turned over) refused the two
sketches as not the same entities, and the single jump left alone failed.
`Sketch::stored_with_rebuilt_texts` gives the stored sketch as this one
has its entities when they differ only by rebuilt texts: the texts'
entities as they are now, every other point, radius, arc turn and
constraint value as stored. `Document::solve_sketch` walks from that, so
the points pair up again. Sketches that differ in anything else are
refused as before.
Test: sketch_project::side_tests::a_label_follows_its_source_far_after_a_text_parameter_gave_it_other_words
(fails without the fix with the conflict warning above).
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
The pixel-bounded curve sampler that draws sketch curves and profile
outlines ran again on every camera move, and each subdivision step
allocated: every span's control points were a Vec, splitting one in
half built a Vec per de Casteljau level, each node's projection was
collected into a Vec, and every drawn part grew from two vertices.
The control points now live inline in a fixed array (`Hull`, at most a
degree-9 Bezier), the halves are worked out in place with the same
`mix` level by level, the projection goes into a reused array, and a
new part starts with room for eight vertices. The arithmetic and its
order are unchanged, so the points are the same bit for bit.
Tests: `samples_the_same_points_bit_for_bit_as_the_reference` keeps the
previous sampler as a reference and compares paths, quality and budget
use over every segment kind, slanted and far-off planes, orthographic
and perspective views from far out to far in, clipped and unclipped,
with ample and spent budgets. Ignored timings:
`sampler_cost_against_the_reference` (render) and
`sketch_scene_cost_while_the_camera_moves` (ui-egui).
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
The web build cannot start the open-meshing thread. Its fallback counted
every body done but listed none ready, so whatever the file load opened
on the web drew no bodies and kept "Loading parts n/n" up for good. The
fallback now lists every body ready and the viewport meshes as it draws,
as the fallback meant to. Found while opening a design through the file
load in the browser (#10198).
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #10218
* Assembly: move sketches to the part they're drawn on (#10218)
`component.move_sketches.to_parts` moves each given sketch, or every sketch in the design, whose
plane is an exactly attached face of a part's body into that part through
`component.move_sketches`, as one undo step. Sketches with no single part to go to stay where they
are, and the reply lists what moved and what stayed, with the reason. The Sketches folder's menu
offers it for the folder's sketches (or the selected ones), and a sketch's menu for that sketch or
the selected sketches; the status bar says how many moved and why the others stayed.
`component.move_sketches` now keeps a face-attached sketch where it is on its face: the sketch is
authored in the new component's placement (the face's own when the component owns the face), and
its picked point and repair plane are carried into it. Before, a face-attached sketch moved out of
the design was left without an author placement and failed to evaluate.
* Tests: the moved part is free to move; face anchor path
* Tests: move the to_parts test to the component command tests
* Shorter label: Move Sketches to Their Parts
* claim: #4581
* View angles card: per-axis zero buttons and locks (#4581)
The readout card under the orientation gizmo already showed and edited the
view's yaw, pitch and roll in degrees. Each axis row now also has a zero
button that turns that angle to 0 degrees at once, and a lock: a locked
angle stays put while orbiting, turning with the arrow keys or using a 3D
mouse, and its field and zero button go inert, so a symmetrical view can
hold its yaw and roll at 0 while only the pitch is adjusted.
* clippy: drop needless borrow in keep_locked call
---------
Co-authored-by: Dahak Builds <dahakbuilds@users.noreply.github.com>
* Sketch Ellipse: Major radius and Minor radius boxes while drawing (#10279)
The Ellipse tool showed no value boxes. After the centre it now shows
Major radius, after the end of the major axis Minor radius; a typed value
places the point at that size and becomes the ellipse's dimension, as for
a circle's Diameter. sketch.draw.dimensioned takes major_radius and
minor_radius for sketch.ellipse, so Dimension new shapes covers it too.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Sketch Ellipse: test the typed radii in the tool and in sketch.draw.dimensioned, and say so in the guide (#10279)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Sketch Ellipse: rustfmt the new engine test (#10279)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: faster highlight with many bodies selected
* Viewport: find selected bodies by name once per highlight build
With every body of a large assembly selected, each hover change built the
highlight again and looked every selected body up along the world's bodies
(`ModelState::body` scans them), which is quadratic: about 2.2 million name
comparisons on 2,116 bodies, some 3 ms of every hover change.
The selected bodies, faces and edges are now found in a map by name, made at
the first lookup. A name held by more than one body finds the first, as the
scan did, so the same bodies are lit in the same order.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Viewport: measure hover changes with many bodies selected
An ignored scale test: every body of a design selected (a file named by
SOLIDCRAFT_SCALE_DESIGN, else a generated grid of 1,440 boxes), the pointer
moving from face to face, timing each hover change's highlight build headless
and as the app builds it on a GPU (bodies drawn from their plans, and from the
model scene's blocks). Run with `--ignored --nocapture`.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Markus Kaltenbrunner <kaltenbrunnermarkus@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Fold the pending design format changes (align-axis-mask, equation-variable-unit, link-file-digests) into format 46.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Keep full names in the tooltip only when the name cell clips them, and keep
the double-click rename hint without stacking a second popup. Cover short
and long names, favorites, and double-click input with a regression test.
Co-authored-by: Codex <noreply@openai.com>
* claim: #10254
* ViewCube: an option to invert the directional rotation arrows (#10254)
A new Navigation preference, "Invert the view cube rotation arrows"
(invert_navigator_arrows, off by default): with it on, an arrow around the
cube orbits the camera the way the arrow points, as in Fusion and Inventor,
instead of rotating the cube toward the side the arrow sits on. Left and
right, and up and down, trade places, and each tooltip names the action
actually performed. The curved roll arrows are not inverted: they name the
rotation they perform, not a direction.
---------
Co-authored-by: Dahak Builds <dahakbuilds@users.noreply.github.com>
* claim: consolidate Gruvbox contrast variants
* Consolidate Gruvbox variants into light and dark contrast choices
* Check both original Gruvbox viewport gradient endpoints
* Reset Gruvbox scenario state before independent theme checks
* claim: #10215
* Sketch: an option to make the part active when sketching on its face (#10215)
Preferences › Sketch › "Sketching on a part's face makes that part active", off by default.
When it is on, a sketch started on a face of another component's body activates that component
at the picked placement first and is made there, as after an explicit activate. The part stays
active, so what is built from the sketch lands in the same part.
For scripts the same is `sketch.create` `activate_part: true`; the reply's `activated` names the
component and placement. On an error the active component goes back to what it was.
* Tests: a free part, no edge projection, the sketch on the face plane
* claim: #10151
Signed-off-by: Maurizio Monticelli <maurizio@whatwedo.ch>
* Sketch: clicking the running tool's button ends it, so clicks select again (#10151)
The lit ribbon button of the running sketch tool started the tool over
when clicked again, so it gave no way out: every click in the view kept
drawing or measuring, and Ctrl+click never reached selection. Esc and
right-click already end a tool, but the pressed button looks like the
natural switch.
A second click on the running tool's button (large ribbon buttons and
the constraint grid) now ends the tool. Persistent analysis toggles keep
their own on/off. The old behaviour stays behind the Sketch preference
"Clicking the running tool's button ends it" (ribbon_click_ends_tool).
Fixes#10151
Signed-off-by: Maurizio Monticelli <maurizio@whatwedo.ch>
---------
Signed-off-by: Maurizio Monticelli <maurizio@whatwedo.ch>
* claim: #10083
Signed-off-by: Maurizio Monticelli <maurizio@whatwedo.ch>
* 3D mouse: turns a rolled or upside-down view the way the cap is pushed (#10083)
In the default upright (turntable) orbit the cap's Spin rolls the view,
but its Tilt and Turn went straight into yaw and pitch, which ignore the
roll. Rolled upside down, every push moved the model the opposite way on
screen; rolled on its side, tilt moved it sideways.
The device now takes Tilt and Turn back through the view's roll before
the orbit, for the cap and for a device app over the navigation bridge.
An unrolled view behaves exactly as before; Free orbit was already
view-relative and is unchanged.
Fixes#10083
Signed-off-by: Maurizio Monticelli <maurizio@whatwedo.ch>
---------
Signed-off-by: Maurizio Monticelli <maurizio@whatwedo.ch>
* claim: #6606
* Sketch: held solves that cannot succeed give up sooner (#6606)
A status-only solve (the held solves of a drag) only asks whether it
succeeds. When LM sits in a least-squares valley of conflicting
constraints it crawled on to its 500-iteration limit, and a held solve
runs LM up to three times. Such a solve now leaves a component once the
iteration has stalled far from a solution: over the last 20 iterations
the sum of squared residuals fell by at most 1e-5 of itself while the
largest residual is still above 1e-4 mm.
Successful solves are unaffected: on about 62,000 LM runs that succeeded
(the sketch and engine tests and 120 generated sketch scripts with 2,900
drags) the slowest still fell by 1.4e-3 over any 20 iterations while
that far off. Solves with an analysis keep iterating (their report names
the violated constraints), and live frames are unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Markus Kaltenbrunner <kaltenbrunnermarkus@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
- sketch.rectangle.center and two_point: a corner given with expressions is a fixed point that follows its parameters, as with sketch.line
- sketch.circle.center: a radius or diameter naming a parameter is a driving dimension; a string naming none is just the number; bare numbers read in the design unit
- A radius that is not a number is refused with its own message and no longer passed over for diameter
- sketch.circle.two_tangent: a string radius reads bare numbers in the design unit
Co-authored-by: P'tite Tête <asso@ptitetete.com>