* claim: #7225
* fix: headless pictures no longer keep the previous picture's highlight
Capturer::rgba_highlighted left the highlight slot empty when a picture had
no highlight. The renderer keeps its last highlight scene while the slot is
empty, so a selected edge's band drawn for one picture showed in the next
one on the same capturer. The depth tests share their capturers across test
threads, so they failed at random: an edge "192 of 612 px drawn", a washer
showing through its plate, a cut showing what lies behind it. A picture now
always gets a highlight scene of its own, empty without one.
Clearing the leak showed that a_selected_edge_on_a_grazing_face_draws_whole
passed at 88 degrees in a design of radius 30 only by counting the leaked
band: its reference line lies 143 mm behind the sheet's middle, past that
design's far plane, and drew nothing. The small design now holds the edge.
New test: a_picture_without_a_highlight_shows_none_of_the_one_before.
Full `--test depth` runs on Apple silicon: 10 of 10 failed before, 0 of 20
after.
* claim: #7274
* Renaming an imported body with numbered siblings evaluates nothing again
Since #2467, Model::rename_body evaluates the design whenever another body's
name is worked out from the renamed one's (`X (2)`). Every STEP import whose
solids share a label has such names (`Part`, `Part (2)`, ...), so renaming
one of its bodies re-read the import and meshed every body again.
body.rename keeps the names of a body's siblings and of later bodies named
after it, so for a body no later feature changes (the same mesh from the
feature that made it to the end), renaming it in the model gives the names
evaluating the renamed design gives. That case is followed again, as before
#2467; bodies later features change still take #2467's checks.
* claim: #7192
* Web app: Report a bug saves its screenshot as a download (#7192)
Save screenshot… wrote the PNG with image's save, that is std::fs, which
the browser doesn't have. The screenshot is now encoded in memory and
written through vfs, as every other save is, so the web app hands it out
as a download. Natively the same file is written and ui.screenshot
replies as before. A failed save says why in plain text instead of a
Rust debug value.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Web app: a screenshot without a path no longer panics (#7192)
`ui.screenshot` without a `path` saved to `std::env::temp_dir()`, which panics on
wasm32-unknown-unknown (std has no file system there), and the web app's control
requests come through `window.solidcraft.request`. `vfs::temp_dir()` now returns
`vfs::BROWSER_TEMP`, a folder of the in-memory files, in the browser, and
`save_screenshot` and `vfs::design_folder_refusal` (reached when a request names
`allow_paths` while a design is open) use it. Natively nothing changes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Web app: keymap and theme exports download (#7192)
Both exports wrote with `std::fs::write`, which always fails in the browser. They
now write through `vfs::write` like the other saves, so the web app hands the file
out as a download. Natively the file is the same; an agent's folder scope now
applies to them as to every other write.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Tests: say what the folder-scope checks of the saves show (#7192)
The screenshot test wraps the write in a folder scope by hand: that proves the write goes through vfs, not that ui.screenshot's write runs inside the request's scope. ui.screenshot checks the agent's folders when the request arrives; the PNG is written in a later frame, outside that scope, as on main.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
The seat of a subscribed connection on the DesignWatch lived in the
connection's serve thread. It left only when that thread saw the end of
the stream, so one unbounded blocking read stood between a closed client
and the count. Under load (GPU backend probing and crash-dump children
on Windows) that read can sit for seconds; the test's 10 s deadline then
expired and `!WATCH.subscribed()` failed (#7226).
Move the seat onto the events thread. The thread ends on `stop`, or on
the first write a closed client refuses, so the next publish retires the
seat. The test now publishes while it waits, and its wait no longer
depends on the serve thread.
Neither theme set the File view's rail colours, so SolidCraft derived the rail from the accent and
the start page's whole sidebar was the primary orange, against the design system's one-or-two
ember budget (#4868 derives an unset rail from the accent). The rail is now n0 (void, beside the
app bar), its hovered and open row n3 (raised), and its text n11.
Session::refresh evaluated the model and then solved the joints, which
writes the solved placements into the design. A feature that reads where
a part is placed kept the result made for the placement before the
solve: a mirror of a jointed part's body, made from the root, ended up
4 mm below the part after a move the slider took back.
When the solve moves parts, the model is now evaluated again for the
placements the design holds, and the joints solved for that model, at
most three rounds. Features that don't read placements come from the
cache.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
A tangency at a point the two curves share (or a line's end held on the round) is solved as a
block of its own, and the orientations a solve keeps left it out. When a point placed by
parameters moved far, the line could land on the other tangent from its far end (its mirror
image) with every constraint met and no warning. It now keeps the round's side of the line, as a
tangency away from the ends already does: kept where the move can be walked there, reported as
a flip where it can't.
Fixes#7428.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #7206
* Browser: a body's name hides the body, not a sketch of the same name (#7206)
`browser.visibility` looked a key up as a feature first, so when a sketch
had exactly the name of a body (a sketch named after the body it makes),
the body's eye in the browser hid and showed the sketch instead, and the
body could not be hidden at all. The browser names a sketch by its id and
a body only by its name, so a name that is exactly both now names the
body. The sketch still hides and shows by its id; matches that differ
only in case are left as they are (#6967, #7004).
Fixes#7569 too.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FN5eEMrqWWXqi65FcK8GXq
---------
Co-authored-by: sascha2101 <221609842+sascha2101@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #7131
* feat: Complete the French UI catalog (#7131)
- Add the 1,153 messages missing from fr.json as a pending fr block
- Keep the terminology already used in fr.json and French typography
---------
Co-authored-by: P'tite Tête <asso@ptitetete.com>
* claim: #7171
* Report a bug: take the screenshot once the preview is done instead of giving up (#7171)
A report opened while a preview was being computed took no screenshot
and asked to be reopened, which loses what was typed. It now waits for
the preview with the report open and takes the screenshot as soon as it
is done; after 10 s it shows the preview unfinished and says so in the
report (a hanging preview is often what is being reported).
The preview the report waited for belongs to its context unless the
design, the dialog or the selection changed meanwhile, so the retake
check compares those alone. What is typed while the capture hides the
report waits for it, and its focused field gets the focus back.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Report a bug: keep the new helpers apart from the other report changes
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Report a bug: keep what is typed as the waited-for screenshot is taken
The screenshot a report waits for starts in the frame the preview
finishes, before the form is drawn, so what was typed in that frame went
nowhere. It is now held with what comes while the form is hidden. The
report's context also leaves the dialog's own state out: a preview error
the dialog shows meanwhile (a sketch fillet's) no longer counts as a
change that refuses the screenshot, since it is often what is reported.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #7188
* macOS menu: disable design commands behind an open Report a bug draft (#7188)
The native menu bar kept running its actions while a Report a bug draft was
open, although the app's own toolbar and shortcuts wait behind it: Edit > Undo
or Cmd+Z removed the last feature, so the report described a design that no
longer existed. FileAction::enabled now refuses every action but Quit while a
draft is open. Settings, Keyboard Shortcuts and Agent Chat are refused too:
their windows open over the report but take no input until it closes. The
menu's Undo and Redo still go to a focused report field
(FileAction::menu_enabled), which no longer leans on Settings being enabled.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Menus: Quit and Help's windows wait behind an open Report a bug draft
Quit switched to the first design with unsaved changes under the draft and
asked about it in a prompt that can't be answered while the report blocks
the app; with nothing unsaved it quit and lost the typed report. Like any
modal window, the draft now turns Quit off. Help's items that open a
window (About, Keyboard Shortcuts, Messages, Crash Reports, Connect AI
assistants) wait too, in the native menu and in the app's Help menu;
Documentation and Report an Issue… stay.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #7190
* Report a bug: focus the empty description once, not every frame (#7190)
After Continue on GitHub found What happened? empty, the form moved the
focus back to it on every frame, so typing in Summary went into the
description. Continue now re-arms the one-time focus (and scroll into
view) instead; the other fields stay free to type in.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #1156
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Startup tests: start the backends the app would, no OpenGL driver on Windows (#1156)
The software-adapter test enumerated every wgpu backend and the probe test
asked OpenGL on its own even where the app stops at D3D12, so both loaded the
GPU vendor's OpenGL driver (atio6axx.dll on AMD) into the test process, where
it once aborted the process at exit with c0000409. The adapter test now starts
what the software renderer starts (D3D12 alone on Windows) and the probe test
probes in the app's order, stopping at the first backend with a GPU.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Startup tests: read the software renderer's backends from wgpu_options
Review: the adapter test now takes the backends from the app's own
wgpu_options(true, None) instead of a copy of its choice, and the probe test
no longer re-checks the order choose_backends' own test covers.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #7281
* cargo xtask skills: check the files as Git stores them, on a CRLF checkout too (#7281)
A Windows checkout with core.autocrlf gives AGENTS.md and the SKILL.md files CRLF line endings,
so `front_matter` found no `---\n` and every skill was reported, and each line's `\r` counted
towards the size caps. The check now reads the files with CRLF taken as LF, as #4960 did for
icon_audit. The new test checks a CRLF skill filled to the cap.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #7299
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Sketch: a circle centred on an existing point shows its Coincident glyph (#7299)
The centre circle tool sent a centre snapped to an existing point as that
point's id, so the circle shared it: attached, but with no constraint and so
no glyph to see, pick or delete. A plain point snap now goes as its place,
where `connect` ties the new centre to the point with a Coincident the sketch
draws. Midpoint, on-curve, crossing and body-vertex snaps still go as refs,
so the centre stays held there (#3099 keeps passing).
Fixes#7299
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Phanoob <phanoob@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Claim: centre arc on an existing point shows its Coincident glyph (Refs #7299)
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Sketch: an arc centred on an existing point shows its Coincident glyph (Refs #7299)
The Centre arc tool sent a centre snapped to an existing point as that point's id, so
the arc shared it with no constraint and the sketch drew no glyph. A plain point snap
now goes as its place, where draw's connect ties the new centre with a Coincident;
midpoint, on-curve, crossing and body-vertex snaps still go as refs.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Phanoob <phanoob@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #7156
* Highlight the selected sketch in the viewport (#7156)
Picking a sketch in the Browser or the timeline selected it, but the
viewport lit nothing: the highlight only drew the bodies and faces a
feature made. It now draws all of a picked sketch's curves in the
selection colour, over the model and also while the sketch is hidden,
so a sketch is easy to find without hiding and showing it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #7348
* Report a bug: include the Messages log and other failing features (#7348)
About half of the reports say "No command error was recorded" and often
can't be reproduced. The error was often in the Messages log, or on a
feature that was neither selected nor the latest. The technical details
now list every failing feature (id, name, kind, error; at most 10) and the
last 8 messages with level and age. Each entry is one quoted line of at
most 300 characters, with the home folder shown as ~.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #6977
* Open from computer: let iOS and iPadOS pick .solidcraft files (#6977)
The iOS file picker matches an input's accept list by system file type,
not by extension. No system type has the .solidcraft extension, so the
designs were greyed out. On an iPhone or iPad the file inputs now offer
every file; other devices keep the extension filter.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #7007
* A surface left without a configuration is configured again, not a crash (#7007)
wgpu reports device-loss errors to no error scope. On D3D12 a driver reset
found by Surface::configure drops the old configuration and reports nothing,
so the surface was taken as configured and the next acquire panicked with
"Surface is not configured for presentation". A configure that meets a lost
device or an unsupported surface is now a refused configure, the acquire runs
under an error scope, and its Validation status reconfigures the surface.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
model_render_follows_the_grid_and_origin_switches_unless_asked compares nine
renders byte for byte and has failed once in a full run without coming back
alone. A mismatch now names the step, counts the differing pixels, lists the
renderer that drew each picture and keeps both PNGs.
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The value box beside a command's geometry and the placement fields kept a number's width, so a
formula or parameter names scrolled out of sight while typing. They keep that width for a number
and grow with the text, up to three times it.
* claim: #7166
* Toolbar: a panel with no buttons keeps one that opens its menu (#7166)
Removing New Sketch from the Model tab (Remove from Toolbar, or a drag off
the toolbar) left the Sketch panel empty above its label, so Sketch seemed
to be missing from the toolbar. A panel whose buttons are all gone is now
drawn like a collapsed panel: its first command's icon over a caret, which
opens the panel's drop-down, in the same room.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #7643
* Web app: Import… reads shortcuts and theme files through vfs (#7643)
Import… on the Keyboard Shortcuts page and the theme strip's Import… tile
read the picked file with std::fs::read_to_string. In the browser a picked
file is kept in vfs's in-memory files under /upload/, and std::fs on
wasm32 is unsupported, so every import failed. Both now read through
vfs::read_to_string, capped (a theme file at theme_files::MAX_BYTES, a
shortcuts file at the new toolbar_custom::MAX_IMPORT_BYTES), like the
other imports. Natively it is the same file on the disk.
The new tests import a file outside a vfs folder scope (refused, nothing
imported) and inside it (imported). With std::fs both fail: the file
outside the scope is imported.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
timeline.redefine evaluated every later feature in its trial session, then again in the session:
the evaluation cache keeps only the last few states of a design with large bodies, so the trial's
results were gone by then. The trial keeps them in the session's kept results (or its own), and
the session's evaluation takes them from there. EvalCache::with_budget gives tests a small cache.
* claim: #7152
* Preferences: the Colour vision preview shows the chosen mode again (#7152)
Since the preferences redesign the Colour vision section previewed a mode
with four status chips (Selected, Warning, Error, Done). No colour vision
mode changes those colours, so the chips looked the same in every mode.
The section shows the sample of what the modes recolour again (sketch
lines, axes, bends, Draft Analysis, a failed feature), drawn in the mode
being chosen (Tokens::get_in). The status chips stay as they are.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01FN5eEMrqWWXqi65FcK8GXq
---------
Co-authored-by: sascha2101 <221609842+sascha2101@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #7091
* Report a bug: scale a large window's screenshot down instead of refusing it (#7091)
A frame over 16 million pixels (a 6K display, a window across two 4K
screens) was refused with "Screenshot dimensions are unsupported", so the
report went out without its picture. Such a frame is now averaged down by
the smallest whole factor that fits.
The frame is also fitted to the longest side a texture can have: a frame
under the pixel limit but wider than the GPU's limit (three ultrawide
screens, 10320 x 1440, on a device limited to 8192) made the preview's
upload fail wgpu validation, and the window's device panics on that.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Inspect: center of mass measures the world state, as Measure does (#5246)
inspect.center_of_mass read the model state, so in an assembly a moved
occurrence was measured where its bodies were made, a second placement
was left out, and a placed copy's world name was "no such bodies". It
now reads the world state, as inspect.measure does, and agrees with
component.properties' mass centre at uniform density.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Center of mass test: unground the placed part before moving it
Upstream now grounds a design's first part when it is made.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* claim: #6281
* assets: add Render appearance, library and paint icon drafts for owner review
* assets: add Render lighting, environment and light-type icon drafts for owner review
* assets: add Render camera, shot, frame and strip icon drafts for owner review
* assets: add Render output, result and compare icon drafts for owner review
* assets: list the Render icon drafts in the inventory and ATTRIBUTION
* assets: remake the Render light icons as one family with a shared dotted stem
* assets: remake the Render look icons around the lit look sphere
* assets: redraw the Render look icons so each action reads at a glance
* assets: record the owner's acceptance of the Render light icons
* assets: simplify the Render look icons and make Remove Override an alias of From Material
* assets: record the owner's acceptance of Render look icons
* assets: mark the removed look on From Material with a cross instead of an arrow
* assets: record the owner's acceptance of further Render icons
* assets: redraw the Render library, presets, environment, ground and backdrop icons
* assets: restore the first Render library and backdrop drawings and equalise the cube in the studio icons
* assets: record the owner's acceptance of the Render studio icons
* assets: redraw the Render camera and shot icons as one family
* assets: redraw the Render All Shots and Lens icons so they read at a glance
* assets: put the play badge back on the Render All Shots icon
* assets: redraw the Render output and result icons around one picture base
* assets: give Render Image its own play badge and enlarge the Compare handle
* assets: record the owner's acceptance of the Render output icons
* assets: describe the final Render icon drafts in RENDER_DRAFTS.md and ATTRIBUTION.md
* claim: #6281
* ui_kit: lit sphere thumbnail, cached and bounded
* ui_kit: look-chain widget (four steps, one active, empty steps dashed)
* ui_kit: add both to the gallery and the UI kit notes
* claim: #6281
* docs: record the Render workspace interaction design and implementation roadmap
* docs: plan the Render panels so each has one purpose
* docs: replace the tabs-and-ribbon plan with the stage-bar design
* docs: drop the note about the superseded draft
* docs: plan every Render panel and where each old control goes
* docs: defer the shared lighting rig, UI steps go first
* Patterns: a body of a moved component is copied where it is shown (#4763)
A pattern of bodies made outside the bodies' component (a screw's body patterned from the root)
moved each body's stored geometry, which is in its component's frame, by transforms in the
pattern's frame. The copies landed where the body sits before its component is placed: near the
origin, for every pattern kind. As Mirror does, each copy is now moved as the pattern moves the
body in the world and kept beside its source, in its component.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Pattern test: unground the placed component before moving it
Upstream now grounds a design's first part when it is made.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Hovering many bodies re-emits every edge in the highlight's colours. Each edge
interned its three colours into the palette again, each segment was written in
eight small appends, and segments without their faces' normals encoded the
same zero normals again. Lines are now one record, the edge colours are
interned once per highlight, the buffers are reserved per edge, and the shared
width word is worked out once.
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
The no-OIT glass cache compared the blocks' vertex and run buffers by address
without holding them, so a freed buffer's address could come back for other
contents. It now holds them and compares with Arc::ptr_eq.
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Instanced placements were culled per view, but a block drew every body in it
whatever the camera saw. Blocks now keep each body's vertex span and box; when
the view changes, each class's runs are drawn without the bodies whose boxes
are out of the view. A block whose bodies are all in view draws its runs as
they are.
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Arabic and Hebrew are complete catalogs since #3768, so
every_complete_catalog_carries_every_english_message fails a pull request
that adds an English message without them. The localization guide's list of
languages a new label needs now says so.
Co-authored-by: rootarchitect <7781866+rootarchitect@users.noreply.github.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>