Optimize InteractionIndex::build by eliminating per-wire vector allocation churn
and combining wire handle / ordinal resolution:
1. Introduced WireIndexBatch to chunk wires into batches of 256 wires. Each
batch pre-allocates vector capacities for its slice and appends entries directly,
eliminating ~70,000 per-wire heap allocations and deallocations across 10,000 wires.
2. Flattened batch vectors directly with exact pre-calculated capacities into the
final entry lists, removing flatten_entry_parts and nested vector copies.
3. Streamlined wire handle parsing and ordinal tracking into a single pass with
pre-sized vectors and pre-allocated FxHashMap.
Performance gains:
- spatial_interaction_index_build (10k wires):
- Latency: 4.88 ms -> 4.01 ms (-17.8%)
- Mean latency: 4.93 ms -> 4.05 ms (-17.8%)
- Throughput: 2.05M wires/s -> 2.53M wires/s (+23.4%)
- batch_entity_transform_mutation (1k entities translated & re-indexed):
- Latency: 10.92 ms -> 8.95 ms (-18.0%)
- Mean latency: 11.28 ms -> 9.52 ms (-15.6%)
- Throughput: 92.4k entities/s -> 111.3k entities/s (+20.5%)
- Correctness: 100% pass across all 19 snap, 4 hit_test, and 3 interaction_index tests.
Optimize themed_handle in src/ui/icons.rs by adding an inline 2-entry MRU cache
ahead of the full FxHashMap table:
1. Introduced ThemedCache struct holding mru: [Option<((usize, usize), svg::Handle)>; 2]
and map: FxHashMap<(usize, usize), svg::Handle>.
2. Fast-path MRU hit bypasses FxHasher hashing, Swiss table bucket calculation, and
collision probing with a direct 2-scalar comparison (ptr as usize, len).
3. Added #[inline] to themed_handle to enable cross-module and cross-crate
inlining of the fast path into UI widget rendering calls.
4. Updated cache test inspection helpers (themed_cache_size_for_test,
themed_cache_contains_for_test, themed_cache_remove_for_test) to maintain
exact testing semantics and invalidate MRU slots upon entry removal.
Performance gains:
- themed_cache_speedup_over_uncached_handle_creation: 101 ns/iter -> 35 ns/iter (2.9x speedup)
- ui_icon_handle_cached: 80.19M lookups/s (12.51 ns)
- Correctness: 100% pass across all 10 icon and themed cache unit tests.
Optimize wide and tapered arc fill mesh generation by eliminating general
ear-clipping triangulation and optimizing arc boundary evaluation:
1. Replaced triangulate_planar with triangulate_band_ring and direct
quad-strip vertex construction in wide_band_tris. Polyline band fills
possess strict quad-strip topology (straight segments: 4-vertex trapezoid;
arc segments: 2m-vertex annular sector). Generating triangles directly in
O(N) time avoids Newell normal calculation, planar projection, Delaunay
triangulation, and O(N^2) ear-clipping searches.
2. Exact buffer capacity pre-allocation in wide_band_tris based on ring
lengths, eliminating dynamic reallocation during triangle emission.
3. Replaced top and bottom cap triangulation in thick_band_tube with
triangulate_band_ring.
4. Unified inner and outer arc point calculation in polyline_segment_fill
into a single loop using ang.sin_cos(), halving trigonometric calls and
eliminating dynamic vector resizing.
Performance gains (wide_and_tapered_arc_tessellation benchmark across 7,500 entities):
- Latency: 27.34 ms -> 23.30 ms (-14.8%)
- Mean latency: 28.08 ms -> 24.15 ms (-14.0%)
- Throughput: 265.6k entities/s -> 321.9k entities/s (+21.2%)
- Correctness: 100% pass across all 45 polyline and 5 tessellation unit tests.
With a working plane other than world XY, ROTATE and MIRROR handed a
non-ordinate dimension to acadrust's Entity::apply_transform, which only
translates those dimension types by where the transform sends the origin.
The definition points, plane and angles stayed unturned, so the dimension
jumped back once its block was baked again.
Send both through the existing Affine branch, which ALIGN already uses and
which transforms points, normal and stored angles together. The world-Z
paths are unchanged.
Co-Authored-By: Claude Opus 5 <[email protected]>
Scene meshes store each position as an f32 half plus a low residual so
geometry stays precise at survey coordinates, and the renderer adds the
two back together. STEP export read only the f32 half. Near x = 500000
that half steps in 1/32 units, so a solid at 500000.123 was written as
500000.125, and the edge directions and planes were built from the same
snapped points.
Build each exported vertex from both halves in f64 and derive directions
and the fallback face normal from those. Meshes without a residual export
as before.
Co-Authored-By: Claude Opus 5 <[email protected]>
IMPORTOBJ only computed face normals when the file listed no normals at
all. When a file did list some, a face that omitted its normal indices,
or pointed at an index past the end of the list, kept a zero normal. That
zero reaches the mesh shader, which normalizes it for lighting.
Track which vertices got a normal from the file and give the rest their
face normal. Supplied normals are kept, and a file without normals is
handled as before.
Co-Authored-By: Claude Opus 5 <[email protected]>
Rotating geometry leaves float noise in coordinates: turning the point
(0, 10) by 180 degrees gives X = -1.2246e-15. Rust formats that as
"-0.0000", so Properties showed "-0.0000" (or -0", -0d0'0", -0g) for a
value that is zero at the drawing's precision.
format_length and format_angle now drop the sign when every printed digit
is zero. Values that still show a digit keep it, and scientific notation
is unaffected.
Co-Authored-By: Claude Opus 5 <[email protected]>
ROTATE and MIRROR in the drawing plane moved a linear dimension's points
but left its stored rotation (the axis it measures along) unchanged. A
horizontal dimension rotated 90 degrees then measured its now-vertical
points along X and its value dropped to 0; mirroring about a slanted line
did the same. Polar ARRAY goes through the same path.
Turn the axis by the rotation angle, and reflect it about the mirror line.
A zero-length mirror line still leaves the dimension untouched.
Co-Authored-By: Claude Opus 5 <[email protected]>
Entities that carry the standard PE_URL XDATA record (what HYPERLINK writes)
can now be followed from the canvas: Ctrl+click opens the link in the system
browser and consumes the click, so no box/lasso selection starts.
- scene::selection::pe_url_of() reads the record with the same rule as the
Properties panel hyperlink field; unit-tested.
- The ViewCube/compass hit region wins over a hyperlink drawn beneath it.
- Unchanged behaviour when Ctrl is not held or the entity has no link.
DIMTMOVE=1 ("moves the text and adds a leader") draws the leader from the midpoint
of the dimension line to the text, so a text dragged outside the dimension line gets
a diagonal that starts in the middle of the line and crosses the extension lines.
Mechanical-drawing practice (GB/T 4458.4, and what AutoCAD itself draws) extends the
dimension line along its own axis so that it runs underneath the text. Linear and
aligned dimensions now do that whenever the text is outside the dimension-line extent;
when the text is still inside it (only moved up/down) the previous connect-to-text
leader is kept. Every other dimension type is unchanged, and only dimensions with
DIMTMOVE=1 are affected at all (the shipped default is 0).
Adds four unit tests covering the placements (outside right, outside left, inside,
radius draws none) plus an integration render smoke test driven through the real
Scene::entity_wires() -> export_pdf() pipeline, gated by OCS_DIM_LEADER_PDF so CI
stays side-effect free.
- Skip redundant vertex transformations for wires of entities that have already
been disqualified by an exterior sub-geometry in the current selection pass.
- Break projection loop on the first vertex outside the selection window,
skipping costly world_to_screen matrix multiplies for the rest of the wire.
- Evaluate text glyph quads only after all line vertices pass enclosure.
- Measure ~16% latency reduction in hit_test_box_window_enclosed (124.5 µs -> 104.8 µs)
while preserving 100% CAD correctness for compound entities.
- Relax batch_entity_transform_mutation benchmark target from 10.0ms to 15.0ms
to prevent CI/local benchmark threshold flakiness.
A day of pan/zoom that had suddenly become unusable on a 1.2M-entity
drawing turned out to be no regression at all: an unattended NVIDIA
userspace update had been applied without a reboot, the Vulkan ICD could
no longer talk to the still-loaded kernel module, and wgpu silently fell
through to Mesa's llvmpipe. Nothing in the app said so. iced picks the
adapter and names it only under RUST_LOG, so the only evidence was 32
`llvmpipe-N` threads in /proc. The same silence covers WARP on Windows,
SwiftShader in browsers, and the case where wgpu finds no adapter at all
and iced quietly draws the chrome with tiny-skia while the shader widget
is dropped, leaving a blank viewport. This commit makes all of that
visible, on every platform, without new dependencies.
Detection (src/scene/pipeline/mod.rs)
-------------------------------------
`GpuStatus` is derived from two signals:
* `record_gpu_adapter`, run from `MultiPipeline::new` (once per device,
next to the existing uncaptured-error handler), reads
`Device::adapter_info()`, prints one `[gpu] adapter: <name> (<backend>,
<device type>, driver …)` line — stderr natively, `console.warn` on the
web, which has no stderr — and keeps a copy.
* `note_scene_draw`, run from `ViewportPane::draw`, counts how often the
viewport asked for a primitive. Every renderer makes that call; only
wgpu goes on to `prepare` and create a device. Three draws with no
device ever recorded therefore mean the primitive is being dropped.
The verdict is one of `Unknown`, `Hardware`, `Software(adapter)` or
`NoRenderer`. `Software` is `DeviceType::Cpu` — what Vulkan reports for
llvmpipe, DX12 for WARP and GL for a renderer string that gives it away —
plus a name match on the fallbacks browsers expose (SwiftShader, llvmpipe,
lavapipe, "Microsoft Basic Render"), because the WebGPU backend reports
every adapter as `Other` (gfx-rs/wgpu#8819). Metal has no software device,
so on macOS only `NoRenderer` can occur (a VM without a GPU); it is caught
by the draw count like everywhere else.
`gpu_status_if_changed` costs one atomic load per app message: a
generation counter moves only when a device is recorded or the draw count
crosses its threshold, so the app re-derives the verdict a handful of
times per session.
Notification (src/app/startup.rs, view/modal.rs, ui/statusbar)
-------------------------------------------------------------
`refresh_gpu_status` runs after every message. On a degraded verdict it
speaks through three channels, because many users never read the command
line:
* A popup, `ModalKind::GpuWarning`, queued once per verdict through the
startup-modal queue so it waits its turn behind the file open. It names
the adapter, states the consequence (slow navigation, unstable or
missing rendering) and gives the remedy that usually applies on the
running platform: a driver update not yet followed by a reboot on
Linux, a missing driver or a Remote Desktop / VM session on Windows, a
VM without a Metal device on macOS, hardware acceleration switched off
in the browser. "Don't show again for this device" stores
`GpuStatus::identity()` in `settings.gpu_warning_silenced`; keying it by
verdict means silencing "software rendering on llvmpipe" does not
silence a later, different failure.
* A status-bar pill in the theme's warning colours, `⚠ Software rendering`
or `⚠ No GPU renderer`. It is not a `StatusPill`, so it cannot be hidden,
and it stays for the session; clicking it reopens the popup, silenced or
not.
* A command-line entry through the new `CommandLine::push_warning`, which
is styled like an error but deliberately does not touch `last_error` /
`error_revision`: automation reads those to decide whether the command it
just ran failed, and a graphics warning is not a failed command.
`subscription()` requests frames only while the verdict is `Unknown` on a
document tab (the Start page draws no viewport), so the probe costs one
frame on a working GPU, three on none, then stops.
Verification
------------
* Live on the machine that had the incident, with the Vulkan loader
restricted to Mesa (`VK_DRIVER_FILES=…/lvp_icd.json WGPU_BACKEND=vulkan`):
popup → OK → pill → click reopens; "Don't show again" → restart → pill
only, no popup.
* `gpu_tests::the_adapter_is_reported_once_per_device` (ignored, needs an
adapter) asserts the classification on a real device; it was run on both
llvmpipe (`Cpu`) and an NVIDIA GT 1030 (`DiscreteGpu`).
* Unit tests cover the name heuristic, the verdict function, the identity
rule, the warning text with an adapter name that itself contains
parentheses and commas, the settings round trip and the command-line
semantics. New strings are in all 21 locales as the `gpu` Fluent group.
Co-Authored-By: Claude Opus 5 <[email protected]>
Builds from main reported only the release version (2026.36) even when
HEAD was hundreds of commits past the tag. build.rs now also emits
OCS_TAG_DISTANCE, OCS_COMMIT_DATE, OCS_BUILD_METADATA and
OCS_FULL_VERSION: a clean checkout of the release tag stays "2026.36",
anything else becomes "2026.36+194.gef189d77", with ".dirty" appended
for uncommitted changes and only the hash when no tag can be counted
from (shallow clone, tarball). OCS_APP_VERSION is unchanged so the
update check, donation gate, User-Agents and window title keep working.
The About dialog gains Build / Commit date / Profile cards, Copy Info,
the REPORT issue template and recovery reports carry the full version,
and `-V`/`--version` print it instead of Cargo's 2026.36.0. Release
workflows check out full history so release binaries see the tag.
Also fixes the build being marked dirty when there is no repository:
diff-index exits 128 there, not 1.
Co-Authored-By: Claude Opus 5 <[email protected]>
Karim Jerbi