Follow-up to #1341:
- interactive UCS offers Face / OBject; bare UCS OBJECT starts an entity pick
- one UcsPickCommand serves both FACE and OBJECT
- ucs_from_normal calls the existing arbitrary-axis helper instead of a copy
- prompt catalogue key and en-US text follow the new option list
Co-Authored-By: Claude Fable 5.1 <[email protected]>
Collapse the triplicated send/timeout/deadline/nested-request loop in interactive_event, get_prompt and needs_entity_pick into PluginProcess::call_simple. Timeout floors, error strings and UnexpectedResponse behavior preserved exactly. Net: +64 / -72 in one file, no behavior change. Verified: cargo check clean, cargo test -p ocs_plugin_api --lib --features host (92 passed), clippy shows only pre-existing warnings.
Delete the byte-identical 0.68 muted_style copies across 13 window modules and import crate::ui::style::common::muted_style instead. print_all keeps its local name via muted_style as muted so call sites are untouched. Divergent alphas (0.60/0.62/0.65/0.72) and missing_fonts weak.text left alone to avoid visual changes. Net: +13 / -78, no visual change. Verified: cargo check --all-targets clean, cargo test --lib ui::style (3 passed).
Deduplicate the byte-identical 1px divider helpers previously copy-pasted across 6 modules. form::{hdivider, vseparator} is now canonical, with a form::vsep alias for legacy call sites. style_manager re-exports both so scale_manager and anno_object_scale imports keep working unchanged. Delete locals in style_manager, plotstyle, dimstyle, layout_manager, attribute_editor. Replace thin wrappers in window/plot with direct imports. no visual change.
[SystemFolder] resolves to SysWOW64 on x64, where ie4uinit.exe
does not exist (only System32). This logged MSI 1721 on every
silent install. Icon associations via ProgId/Extension need no
manual refresh, so drop the action instead of switching to
System64Folder.
Older GPUs without usable DirectX 12 Feature Level 12_0 or working
Vulkan support (e.g. legacy Intel iGPUs and other old parts), and old
GLES drivers whose shader compiler rejects iced's own shaders, crash
at startup or on the first drawing viewport: StagingBelt aborts in
Buffer::get_mapped_range (panic=abort, silent exit) on the DX12 path,
and shader compilation fails inside iced's gradient pipeline on the
GLES path. Selecting a working backend via WGPU_BACKEND before iced
boots avoids both classes.
- Probe each candidate backend (Windows: dx12,vulkan,gl; Linux:
vulkan,gl) in an isolated --gpu-probe child: device request,
staging-style mapped upload, render pipeline creation (its fragment
shader reads a private vec4 array, catching old-Mesa compile
failures), offscreen render and readback. First fully-passing
backend wins; explicit --backend / WGPU_BACKEND / --safe-mode
always win, and Windows keeps preferring DX12/Vulkan (#55).
- Software adapters (WARP/llvmpipe/SwiftShader) are the last resort
before a best-effort GL attempt: slow but alive.
- The crash sentinel names the attempted backend so only implicated
backends are skipped next launch.
- Winning probes report adapter limits; GPUs without shader storage
buffers get the packed compatibility renderer automatically, with
a startup command-line notice so the switch is never silent.
On the web build, {"op":"open","path":…} resolves the name against the
browser's recent-files cache, so a file that is not already there cannot be
opened by the control channel. Accept {"op":"open","name":…,
"data_base64":…} instead: the bytes go through the same parse, caching
and recovery path as a file chosen in the browser picker
(io::open_bytes_web, now shared with pick_and_load_web).
Native builds keep path-based open and reject data_base64 with
invalid_request.
Refs #1349
serde_json keeps 10 and 10.0 as different Numbers, so an expected of 10
never matched a stored whole-number float, and a JavaScript client cannot
send 10.0 at all. Compare numbers by value (recursively through arrays and
objects); everything else keeps structural equality.
Fixes#1348
crate::network and the open crate are desktop-only, but the community-font
download (font_repo.rs) and the XREF row Open action call them
unconditionally, so cargo check --target wasm32-unknown-unknown fails on
main. Gate both; on the web download_fonts returns an error, since the web
build has no network client or fonts folder.
- app_with_fixture embedded CARGO_MANIFEST_DIR straight into an
automation JSON string; on Windows the backslashes become invalid
JSON escapes and the open failed. Build the payload with json!.
- driver_options_are_edited_remembered_and_sent_with_the_job exercises
the CUPS options editor, which Windows never opens (PrinterProperties
routes to the system dialog there); gate it to non-Windows.
When a display plot style maps an ACI to a fill pattern, wire fill
triangles are re-emitted as PLOTSTYLE hatches. Those took
depth_override raw — a per-block child label for block wires and 0.0
for top-level ones — while everything else in the hatch/wipeout pass
sorts at the composed depths[insert] + label * half. The fill could
land outside its host's band and get buried under its own block's
masks (same shape as the block-text bug).
Both the display and PDF paths now travel through wire_draw_depth /
the PlotWire's composed depth, and the display cache key rides the
depth generation so a DRAWORDER edit re-ranks the fills without
needing a wire rebuild.
Unselected block text (the BS33-11 Mark/SectionMark labels) was erased by
its own block's wipeout pass and only came back through the selection
xray. Two compounding causes:
- Glyph quads rendered at the bare insert level, an exact depth tie with
every sibling fill, so the later wipeout pass won LessEqual. Wire
batches now carry the composed child rank as depth_override (the same
units the scene graph composes: depths[insert] + label * half), which
the block-text upload resolves per instance. Glyph vertices stay
neutral on purpose — baking the raw per-block label into them
double-counts and, for a negative rank, sinks the text BELOW its own
masks.
- The wipeout pipeline's depth bias (-8 quanta) was larger than the
whole within-block band: once child ranks compress into the insert's
half-band, sibling steps are ~3 quanta and a wipeout-to-text gap is
only 9-16 quanta of the 24-bit buffer. The mask only needs to break
exact ties, so the bias is now WIPEOUT_DEPTH_BIAS_QUANTA = 2.
block_text_draw_depth regression pins the contract: neutral vertices,
override carrying the rank, and a margin over the wipe that exceeds the
bias on a 600-entity scene.
load_file, used by headless export, xrefs and block import, finalizes a
drawing without purging, so the spline degree repair never ran there and
`--export drawing.dxf out.dwg` still hit the knot underflow. Run the repair
when any load is finalized.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Retaining a line's length added a P2PDistance constraint, whose gradient
divides by the current distance. For a zero-length line the solver got a NaN
Jacobian and its SVD fallback never returned, freezing the editor (e.g.
ECONSTRAINT on a line drawn by clicking one point twice). Hold coincident
endpoints together with per-axis differences instead.
Control-point splines read without knots are given a degree their control
points can carry (at most n - 1) and a clamped knot vector when a drawing
loads. Generating knots for fewer points than the degree underflowed during
DWG write: a panic in debug builds, an unbounded allocation in release, and
reachable from autosave. A lone control point with no knots or fit points is
purged as corrupt.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
- ARRAYRECT, ARRAYPOLAR, ARRAYPATH, ARRAY3D and COPY Array reject item
counts above 100,000 (checked as a product of every dimension) and keep
the prompt open. Unbounded counts allocated one transform per item and
aborted on allocation failure. Stored defaults are clamped as well.
- EXPLODE leaves a MINSERT above the same limit intact instead of
materializing every cell.
- Typed lengths and angles reject non-finite values and magnitudes above
1e15; SCALE rejects a non-finite factor.
- Zero-radius FILLET compares candidate intersections with total_cmp and
rejects non-finite results; partial_cmp().unwrap() panicked on NaN.
- SOLID falls back to a minimal kernel tolerance when its corners' extent
overflows, since Tolerance::new panics on infinity.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
Right-to-left line reordering walked each atom's byte range and sliced the
line at every byte, panicking when a neutral multi-byte character such as a
guillemet preceded the first strong character. Walk characters instead.
FIND/FINDALL located the REPLACE keyword in a Unicode-uppercased copy and
sliced the original with that offset; uppercasing can change byte length
(dotless i, ligatures). Uppercase ASCII only, which keeps offsets aligned.
FILLET and CHAMFER inline R/D shorthands now test the original text before
slicing past the first character.
Co-Authored-By: Claude Opus 5 (1M context) <[email protected]>
A drawing saved by a commercial CAD application (tests/fixtures/plot,
provided by Panos: one 10 m line, three layouts) corrects several
assumptions of the previous commit:
- rotation is relative to the medium's own orientation: an unchanged sheet
keeps the driver's exact medium (its PDF driver lists landscape media of
its own, written with rotation 0; a system printer's portrait A3 plotted
landscape carries 90); only a newly picked sheet is written as the
catalogue's portrait medium
- the stored scale factor (DXF 147) is the ratio × 25.4 on millimetre page
setups and the plain ratio on inch ones; the "use standard scale" bit
stays clear for standard and custom scales and is set for fit-to-paper;
a standard code is kept across a paper-unit conversion ("1:1" on an
inch page setup over a millimetre paper space is 1 in = 25.4 units)
- layout plots take the same scale path as every other area, so a foreign
layout's paper-space unit survives
- "plot with plot styles" / "display plot styles" round-trip as the
dialog's checkboxes, independent of a style table being named
- layout limits place the sheet at minus its displayed margins, the way the
source application does (the printable corner is the paper-space origin);
the margin ↔ rotation mapping shared by the printable-area guide now
turns the medium's top margin into the layout's left one at 90°
The fixture round-trips through the dialog with every stored field
unchanged; picking a new sheet on it writes the canonical form.
Co-Authored-By: Claude Opus 5 <[email protected]>
One writer (plot_settings_from_dialog) now serves both the layout's embedded
settings and named page setups, so the two can no longer drift apart:
- media are stored as the portrait sheet with the orientation in the
rotation code (0/90/180/270; upside-down adds a half turn) instead of
swapped width/height with rotation limited to 0/180
- standard scales are written by their code (1:1 … 100:1) with matching
numerator/denominator; other scales as a readable custom ratio in the
layout's own paper unit, converting from the drawing's scale-family unit
(a metre drawing plotting "1:100" stores 10 mm = 1 unit, an inch layout
stores 1 in = 2540 units); reading converts back and keeps a ratio the
scale list has no name for instead of snapping to 1:1
- the layout's paper unit and every flag the dialog does not edit survive a
round trip untouched; layout limits follow the paper-space unit
- new drawings, new layouts, CLEAR and the automation "new" op share one
default page setup (canonical A4 name, rotation 90, none_device, 1:1
layout plot) through Scene::reset_to_new_drawing
Scene gains scale_family_unit_mm, scale_ratio and a scale-name ratio parser
(ratios and architectural names). Round-trip tests cover orientation,
standard/custom/fit scales, unit conversion, named-setup parity, preserved
flags and the default layout.
Co-Authored-By: Claude Opus 5 <[email protected]>
The Display page could only take a typed #RRGGBB, which is precise but
no way to choose a colour by eye. The swatches are now buttons that
open a wheel.
The wheel does not replace the hex field, it sits beside it: typing an
exact value stays the quickest way to reproduce a known colour. On
submit it hands a hex string to the existing ModelSpaceBgChanged and
friends rather than writing the config itself, so the two routes
cannot drift apart on validation, persistence, or the MatchTheme
fallback an empty field triggers.
Also stops the Browser shipping in the default dock layout. It was
added to the right edge, which shifted the block palette to index 1
and broke four dock tests that encode the existing arrangement. Those
tests were right and the change was wrong: the earlier commit claimed
to leave the layout alone. BROWSER now docks the panel on first use,
so nobody's edges rearrange on upgrade for a panel they never asked
for.
nix build now succeeds. The hashes were placeholders before, so it
could not have.
Reading Cargo.toml for the git sources missed two of them: cryoglyph
and dpi come in through iced, which pins its own forks, and they only
appear in Cargo.lock. Six entries, not four. The comment says to
regenerate the list from the lock for that reason.
flake.lock is committed alongside, so a checkout resolves nixpkgs and
rust-overlay the same way next month as today.
Also asserts the snap default instead of leaving it to be eyeballed.
It is only visible in a fresh profile — the app persists settings on
change, so an existing settings.json keeps its own osmode and never
shows what a new install would do.
UCS FACE had no test at all: it was only known to compile. The plane
it builds has two halves and they sit either side of a visibility
boundary, so the coverage splits to match.
tests/ucs_face_plane.rs takes the public route — extrude a disc, then
check the caps come back as distinct planar faces with opposite
outward normals, that an off-centre pick lands on the same face as a
central one, and that the curved wall is never offered as a plane.
The axis construction is pub(crate), so it is unit-tested next to
ucs_from_normal instead: Z is the normal it was handed, the axes stay
orthonormal, a normal along world Z still yields a plane rather than
crossing with itself, the same normal is reproducible, and a zero
vector is refused rather than silently producing a degenerate UCS.
Byte-pair separators in the hex colours, and #[must_use] on the four
constructors, which return a value and do nothing else — dropping one
is always a mistake rather than a choice.
Found by clippy::pedantic. Only this file is touched: the same run
reported ten thousand warnings across the tree, nearly all numeric
cast lints that are load-bearing in geometry code, so fixing them
wholesale is a separate decision rather than a drive-by.
A docked outline of the drawing: the three origin planes, the open
sketch, and every solid labelled with the operation that made it,
recovered through primitive_property_operation. Rows run ordinary
commands, so clicking a plane starts a sketch on it and clicking a
body selects it.
It is a table of contents, not a feature tree. The DWG history records
how each solid was built but nothing re-runs it, so an earlier feature
cannot be edited to rebuild the ones after it. Presenting the list as
editable history would promise exactly that, so the panel states what
it is instead.
Off by default and opened with BROWSER, leaving the existing dock
layout alone for anyone who does not want it.
CREATESKETCH opens a sketch on the current UCS, or on one of the three
origin planes with XY, XZ or YZ. FINISHSKETCH closes it again.
Nothing here is new machinery: the plane is a UCS, the square-on view
is PLAN, the drafting aids are the snapper. What the pair adds is
holding them together and putting them back afterwards, including the
snap master state, so a user who works with snapping off gets that
back rather than silently keeping it.
Squaring the view up is not cosmetic. Drawing on a plane seen nearly
edge-on is the main way geometry lands somewhere the user did not
mean, because a few pixels of cursor travel cover a long way in world
space.
A sketch is a mode rather than a document object, so what it produces
is ordinary model-space geometry and the file stays a DWG other
software can open. Whatever was drawn becomes the Previous selection
set, so EXTRUDE then Previous consumes it without making the user
reselect a profile they just finished.
Draw a profile, extrude it, then sketch on a face of the result: the
operations were all there, but nothing turned a solid's face into the
active drawing plane, so the second half of that loop was unreachable.
UCS FACE takes a surface pick, finds the planar face under it and
builds the plane from the face normal. X comes from the Arbitrary Axis
Algorithm, so the same face always yields the same X axis and a sketch
drawn on it does not spin when the UCS is rebuilt.
UCS OBJECT adopts a planar entity's own stored axes instead of
re-deriving X, so a sketch started from a circle lines up with how
that circle is stored. It reuses the profile extraction EXTRUDE
already runs, which means anything extrudable can also be sketched on.
Bare UCS FACE starts an interactive pick that resolves through the
same inline handler, so the clicked and typed paths cannot diverge.
Fusion Black pairs near-black chrome with a paper-white canvas, so the
drawing is the only bright thing on screen. That breaks the assumption
in theme_canvas_background that the canvas follows the chrome, hence
the explicit lookup: falling through to the palette would paint model
space black and undo the point of the theme.
Both keep a blue accent rather than going fully greyscale. Selection
highlight defaults to the theme primary, and a white primary would be
invisible against a white canvas.
These are Theme::Custom values and so are absent from iced's
Theme::ALL. all_themes() is the list to enumerate from now; the picker
in the options dialog uses it, and parse_theme_name resolves both by
name.
The running snap set was already seeded with endpoint, midpoint,
centre, node, quadrant, intersection and nearest, but the master
suppress bit was set, so none of them did anything until the user
found the status-bar pill. Snapping is what makes a drawing close
rather than nearly close; it should be live from the first click.
Only fresh configs change. A stored settings.json keeps whatever
osmode it already holds, and a legacy file carrying its own $OSMODE
still overrides on open.
The devShell carries the toolchain plus the X11, Wayland, Vulkan,
fontconfig and freetype libraries winit and wgpu dlopen at run time,
and the trunk/wasm-bindgen pair the web build needs. LD_LIBRARY_PATH
is set there too, since a dev binary runs straight out of ./target and
never gets the wrapper the package build applies.
The four git dependencies each need an entry in cargoLock.outputHashes.
They are lib.fakeHash for now; the first `nix build` prints the real
values, and they have to be refreshed whenever a rev in Cargo.toml
moves.
Hakan Seven