* Render results: keep up to eight completed renders and select or discard them
* Render results: Live and render compare with a divider, and render all shots
* fix: use comparison theme tint and reveal reset test targets
* fix: restore Unicode expectations in render workspace tests
* claim: #6528
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Lost GPU: the per-frame probe opens a command list, as the upload would (#6528)
On a reset D3D12 device the empty submission before a frame's uploads
went through (it only signals a fence), and the first queue write then
panicked inside wgpu-core (PendingWrites::activate unwrapping Lost).
The probe now submits a command buffer with a debug marker: finishing it
makes the same begin_encoding call, through a path where wgpu loses the
device and runs its lost callback instead of panicking. The picture
capturer's probe does the same.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #7159
* Measure: a fillet's torus or a sphere reports its radius (#7159)
The face of a fillet round a circular edge is a torus and a ball's face a sphere, which
Measure listed by area only. Their face item now carries radius, diameter and surface
(and the ring radius of a torus); they stay out of Shape, so distances are unchanged.
Fixes#7159.
Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 5.5 <noreply@anthropic.com>
* claim: #9111
* Preview: new surface is tinted whatever its size, and the tints can be turned off (#9111)
A preview decided which faces of a changed body were new surface by asking
whether their triangle centroids lay on the old body, within a tolerance that
follows the part's size. A round or chamfer small beside the part never leaves
that tolerance, so it was drawn as if it had always been there, and sampling at
a fixed step told alike faces apart by nothing but how many triangles they had.
The kernel already records which faces an operation generated, so `build` takes
that record alongside the geometric test. The test stays because a body carries
the history of whatever last made it, which is not the previewed operation when
a parameter change re-evaluates the design: the record only ever adds faces to
the answer. The vote itself now asks about every centroid of a small face,
spreads a large face's samples without a resonant step, and calls a face new
above an eighth of its samples, which a blend tangent on both of its sides
reaches with the one row of facets in five that is clearly off the old surface.
Graphics gains two settings, both on as they come, so nothing changes out of
the box: Tint new surface in a preview draws a changed body as itself when it
is off, working out no face at all, and Show removed material in red leaves the
see-through cut and the ghost of the shape a feature replaces fully
transparent, drawing only their outlines.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
* claim: #9084
* Edit Extrude: close on the first Escape (#9084)
An edit dialog whose value names a parameter (`T` in Edit Extrude) opened with
that field focused and its completion list showing the longer names (`tan(`),
unarmed. The list took the first Escape, so the dialog needed two.
An unarmed list now leaves Escape to the window until the box is typed in, as
it already leaves Enter and Tab; once typed in, Escape dismisses the list first
as before. Its hint drops "Esc dismiss" in that state.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #9827
* Sketch Offset: an Instances count stacks offsets one behind another (#9827)
Offset made one offset at a time, so a set of evenly spaced outlines meant
running it again on the curves it had just made, once per ring: a separate undo
step and a separate driving dimension each time.
`sketch.offset` now takes `instances`, a whole number or a unitless expression
from 1 to 100 (1, today's behaviour, by default). The per-group work the command
did inline becomes `offset_rung`, and `offset_ladder` runs it once per instance,
each rung offsetting the one before it the same distance again, so instance n
stands n distances from the curves picked. Every rung of a side is driven by
that side's own dimension, so editing it moves the whole set and keeps the
spacing even, and each side of a Two Sides or Symmetric offset grows its own
ladder. A rung the shape has no room for refuses the whole command and names
which instance it was.
A later rung reads its side from the geometry, through the rung's own
normal-distance annotation, rather than from the distance's sign: an output's
stored ends need not run the way the chain through them is walked, and a
reversed walk reads the same sign as the other side. A circle is the exception
and keeps the sign, because it offsets concentrically with no chain to walk, and
a step past a ring smaller than the distance lands beyond the centre, where the
circle's "outside the current radius" side test reads as the outward side: an
inward ladder turned round and grew again (1, 4, 1, 4 mm) instead of running out
of room.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
* Sketch Offset: the Instances row in the offset window (#9827)
The offset window gets an Instances row under Distance: a whole number with step
arrows, 1 by default, so the count can be walked up a ring at a time and the
preview follows each step. The count is not kept in the design, as Mirror's and
Scale's values are not, so the window sends the number it evaluates to rather
than the text.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
* UI kit: a count's step arrows are not text (#9827)
The arrows stand inside the editor, which asks for the text cursor across its
whole rect, so hovering one showed an I-beam and the arrows read as part of the
number. They take the ordinary cursor back after it, and the pointer is read
from the arrow's rect rather than from the hover the editor underneath competes
for, which keeps the hover highlight in step with it.
The pattern Quantity rows share this field and had the same thing.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
* Sketch Offset: Instances label in the complete catalogs and the other shipped languages
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Sketch Offset Instances: translations for the complete catalogs
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
Co-authored-by: bherbruck <brennen.herbruck@gmail.com>
* claim: #9873
* Add the Preferences design reference: pages, figures, measurements, tokens
* Fix the selection look in the design reference to match the app
File › Export STEP…, STL…, 3MF…, OBJ… and URDF… were UI actions around
the shared file.export, so Keyboard Shortcuts had no row for them.
file.export_step, _stl, _3mf, _obj and _urdf now do what those items do
when started from a key, command search or the menu (a save picker; Save
As Mesh for STL and 3MF), and take a path when called with JSON. The File
menu shows their keys. Each has a contract and a reply type
(schema/file.rs).
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #10600
* Sketch: a value box must not swallow a click that lands on a sketch point (#10600)
The construction line tools' Length box waits beside the pointer. When the
pointer reaches the reference line's own endpoint it can land under that box
(the box is anchored to the tool hover, which freezes while the pointer rests
on the box), and the box's text edit takes the click, so the tool never sees
it. A release over the tool's own value box or construction/normal switch that
lands on a snapped sketch point now goes to the tool; a click on the box away
from any sketch point still edits the box.
---------
Co-authored-by: Dahak Builds <dahakbuilds@users.noreply.github.com>
* claim: Assemblies: use lookup maps when building the world state
* Assemblies: use lookup maps when building the world state
world_state_reusing, made again after every geometry command in a design
with components, filtered all bodies for every placement, searched the
timeline for every sketch and construction of a placed part, and searched
all bodies twice per tap and placement. It now builds the feature to
component map, the bodies of each component (in body order) and each
tap's component and body offset once per call. The result and its order
are unchanged.
On a 100-part assembly placed 5 times (800 sketches, 200 taps) one world
state drops from 124 ms to 9.5 ms (release, median).
A test-only counter of searches through all features or all bodies
checks that the work no longer grows with the sketches and taps.
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>
A join, cut or intersect without explicit targets reads where every body
of its component is placed. Its placement digest searched the model for
each of those bodies by name, then the timeline for its component and the
placements for that component's frame: O(B^2 + B*F) per feature, on every
evaluation, also when every feature comes from the cache.
The bodies, their components (Document::model_body_components) and each
component's frame are now looked up once per digest. The bodies are still
hashed in name order with the same values, so every digest stays the same
bit for bit and saved caches stay valid.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #10393
* Per-component History hiding: flag, commands, Browser menu, bar toggle (#10393)
* Carry the new History-hiding strings in the complete catalogs (#10393)
---------
Co-authored-by: Dahak Builds <dahakbuilds@users.noreply.github.com>
* Sketch: a construction line with a length can be a normal line (#10258)
`construction: false` on a construction line command that has a
`length` makes the segment a normal sketch line, ready to close a
profile, instead of construction geometry. The default stays
construction, and an infinite line is always construction.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Sketch: a length box in the construction line tools (#10258)
The horizontal, vertical, parallel and perpendicular construction line
tools show a Length box beside the pointer from the start. It stays
empty (shown as infinite) and leaves the keyboard to the tool's
shortcuts until a digit is typed; then the box takes the keyboard.
Without a length the tools work as before: one click, an infinite line.
With a length the first click sets the start and a second click the side
the segment runs to, sent as the command's `length` and `near`, so the
segment gets its length dimension. The angle tool has a Length box
beside its Angle box. A typed length keeps the start's snap, and the
preview draws the segment the click will create.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Sketch: a construction/normal switch in the construction line tools (#10258)
A Construction | Normal switch sits under the Length box of the
construction line tools. Normal sends `construction: false`, so a
segment with a length becomes a normal sketch line, ready to close a
profile, without converting it afterwards. The switch starts at
Construction with each start of a tool. An infinite line is always
construction, so without a length the switch is greyed out and its
tooltip says why. It belongs to the tool alone: the Sketch Palette's
drawing line type (X) is left as it is.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: Sketch drag: drop the second document copy per frame
* Sketch drag: drop the second document copy per frame
edit_solved copied the document to edit it, then stored it with
`*s.doc_mut() = doc`. The command's undo snapshot still holds the old
document, so doc_mut copied the whole design a second time only to
overwrite it. Store the edited document with Session::set_doc instead,
which forgets the same caches and puts it in a new Arc. The design lands
at a new address as before and the command's end records it for the
crash journal (#7502).
A live drag frame on a design of 200 sketches (20k curves) drops from
19.2 ms to 9.4 ms; 50 sketches: 2.3 ms to 1.7 ms.
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>
* claim: Move: say a part's joint conflicts instead of 'no movable joint axis'
* Move: say a part's joint conflicts instead of 'no movable joint axis'
When a joint of the selected part conflicts with the others (the solver
can't close its loop), Move found no axis and said "This part has no
movable joint axis.", pointing at joint types and limits. It now names
the conflicting joint and suggests editing, suppressing or deleting it.
Parts with no free joint value and no conflict keep the old message.
Refs #10331
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>
* claim: #10484
* Sketch: Intersect can pick sketch lines and model edges (#10484)
The Intersect dialog only took faces or whole bodies, so a click on a
line of another sketch (the sides of a rectangle crossed by a plane
through its middle) picked nothing, although sketch.intersect already
cuts edges and sketch curves with the sketch plane. The pick list now
takes faces, edges and other sketches' curves, and the selection filter
reads Geometry | Bodies like Project's.
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>
* claim: #10471
* Show given names in bidi display order; find RTL prefs labels typed in reading order (#10471)
crate::bidi ran over catalog text only, so a multi-word Hebrew or Arabic
body name showed its words left to right in stored order, and Preferences
search compared display-order labels against reading-order queries.
given_name now returns right-to-left names in display order (browser rows,
tabs, status bar all draw names through it), and text_matches also scores
the query after the same bidi transform.
---------
Co-authored-by: Dahak Builds <dahakbuilds@users.noreply.github.com>
* Configuration rules: Then says it lists only the table's columns, and how to add one (#10532)
A rule drives or restricts a column of the table only, so Then leaves out a
feature without a Suppress column or a parameter without a column, where
When lists them all. A note under the list now says so and points at Add
Column.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Configuration rules: the two notes in the complete catalogs (Arabic, Hebrew, Japanese, Vietnamese, Chinese)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: View: work out consumed sketches and hidden bodies once per revision
* View: work out the sketches features use once per design
Each sketch_shown call walked every feature of the design to tell whether
the sketch is used, and the viewport asks about every sketch several
times a frame. The used sketches are now gathered into a set once per
design (keyed on the design itself, so any edit reads them again).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* View: look hidden bodies up in a set
body_visible walked the hidden-bodies list for every body asked about, and
the viewport asks about every body several times a frame. The hidden
bodies now join the hidden joints and folders in the view's visibility
index, built when the list changes.
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>
* claim: View: work out consumed sketches and hidden bodies once per revision
* View: work out the sketches features use once per design
Each sketch_shown call walked every feature of the design to tell whether
the sketch is used, and the viewport asks about every sketch several
times a frame. The used sketches are now gathered into a set once per
design (keyed on the design itself, so any edit reads them again).
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>
* claim: Canvas: store reference image data as shared text so edits stop copying it
* Canvas: store reference image data as shared text so edits stop copying it
Every edit clones the document, because the undo history still holds the
old one, and the canvas image was a plain String: each of the up to 200
undo states owned its own copy of the base64 text. Canvas::data is now
Arc<str>, like the shared STEP text, meshes and point clouds; serde's rc
feature writes it as the same JSON string, so design files stay byte for
byte the same.
A 4.2 MB image with 250 edits: peak memory 1.09 GiB -> 42 MB, run time
1.16 s -> 0.12 s.
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>
* Browser: a folder that holds only a new group shows it (#10526)
New Group from the root's context menu made an empty group in the Bodies
folder, but the Browser draws no folder without items, so in a design
with no bodies yet nothing appeared. A folder that holds groups is now
drawn with them, even before anything is in it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Browser: test that a new group in an empty folder shows and takes a dragged body (#10526)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: Extrude dialog: work out the automatic operation once, not every frame
* Extrude dialog: work out the automatic operation once, not every frame
An open Extrude window with the automatic operation on asked the engine
for `auto_operation` every frame. That probes the bodies with
`Mesh::contains`, rays through every triangle, so a large part cost tens
of milliseconds per frame with nothing changing.
The window now keeps the answer with a key over what it depends on: the
extrude's parameters (participants included), the design's revision and
id, and the active component and occurrence path. The latch still runs
every frame as before.
Open Extrude over a 1M-triangle mesh body: 29.8 -> 0.87 ms per frame
(headless harness, median of 3).
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>
* claim: Extrude dialog: keep the taper limits while the profile is unchanged
* Extrude dialog: keep the taper limits while the profile is unchanged
The taper handle is drawn every frame while the extrude window is open,
and every frame it worked out the inradius of each picked profile and of
each of its holes: a grid search over every side, about 200 ms a frame
for a plate with 64 round holes. Only the height changes between frames.
The radii are now kept in the window's state with the profiles they were
worked out for, and worked out again only when the profiles differ. The
limits come from the same arithmetic, so they 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>
* claim: #10016
* Sketch: a Line chain survives Undo (#10016)
Undo could take away the segment the chain was drawing on from, and every
further click was then refused with "unknown sketch point". The chain now
remembers where each segment started and goes back to the last one still there.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: Raytracer: map faces to looks once
* Raytracer: map faces to looks once per body
Preparing a body for the CPU raytracer walked the body's face look list
one entry per work step for every triangle, so a body cost T x (F + 1)
steps (each through the job host) instead of T. The face looks are now
sorted once per body, keeping the first entry per face as before, and
each triangle finds its look with one binary search. Pixels are
unchanged; with 10 000 faces and 50 to 500 face looks, preparation
drops by 40 to 88 %.
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>
* claim: Opening: validate the document once when a model cache is present
* Opening: validate the document once when a model cache is present
Opening a design that keeps its model parsed and validated document.json
twice: reading the model cache checked it before decoding the cached
bodies, and read_document, which follows straight after, runs the same
migrations and validation and fails the open when they fail. The open
path now reads the model without that check; geometry(),
read_design_geometry and the preview keep it. A broken document fails the
open with the same error as before.
3,000-feature design with its model cache, bench-open parse step:
426 -> 293 ms (median of 8, heavy machine load). Saving the opened
design gives a byte-identical file.
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>
* claim: #10483
* ViewCube: an option to invert the curved roll arrows (#10483)
A new Navigation preference, "Invert the curved roll arrows"
(invert_curved_roll_arrows, off by default): with it on, the two curved
arrows at the cube's corner roll the view the other way within the
screen plane - the clockwise arrow rolls counterclockwise and the
counterclockwise arrow rolls clockwise. The sibling of the cardinal-arrow
inversion from #10286, which left these arrows alone.
---------
Co-authored-by: Dahak Builds <dahakbuilds@users.noreply.github.com>
* claim: Mesh imports: compare by pointer before comparing contents
* Mesh imports: compare by pointer before comparing contents
MeshData derived PartialEq, so the dirty check and the autosave check
walked every coordinate and index of a scanned mesh after each edit,
although the document and its saved copy share the same arrays. Compare
shared arrays by pointer first and value by value otherwise, as Points
and Colors already do. About 1.9 ms per comparison for a mesh of one
million triangles goes to nothing.
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>
* claim: Sketch: use a lookup map for curves in pattern upkeep
* Sketch: use a lookup map for curves in pattern upkeep
Pattern upkeep runs before every solve, each drag frame included, and
looked up every copy curve and its arc points with a linear search over
all ids. Dragging another line in a sketch with a 500-copy slot pattern
spent about two thirds of each frame there. The ids now resolve through
maps built once per call; the result is the same to the bit.
The solve bench gains a circular slot pattern section.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Sketch: hash sets for the stale-pattern check before each solve
The check whether a pattern lost a seed, copy or relation built three
ordered sets of all ids before every solve. Only membership is asked, so
hash sets give the same answer; a drag frame in a sketch with a 500-copy
pattern takes another third less.
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>