Every weight under 0.26 mm was clamped to the screen's 1 px floor before
the PDF and SVG exporters saw it, so 0.00, 0.09, 0.13 and 0.25 mm all
plotted alike. The floor now applies on the GPU only.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Every further reference to an already expanded block (a translated copy
of the same definition, at the top level or nested) was materialised as
a full copy of every wire, although the GPU already draws those copies
from one shared source with a per-copy offset. Such a copy is now a
single stub wire holding the shared prototype and its offset (stubs nest
inside prototypes, so nested repeats collapse too).
- The block renderer expands stubs into instance references while it
groups by source, so drawing is unchanged.
- Interaction queries expand only the stubs near the query, so snapping,
picking, window and crossing selection see each copy's geometry.
- Plot, boundary detection, viewport projection and entity_wires() get
fully expanded wires; count and reference-edit recolouring expand the
set while they are active.
- OCS_NO_BLOCK_STUBS turns it off.
A drawing with 480 references to a block of 100 nested 50-line blocks
(2.4M lines) now opens with 11k wires instead of 2.4M: peak memory
5.4 GB -> 0.37 GB, wire assembly 2.4 s -> 36 ms, interaction index
3.7 s -> 23 ms. The viewport, a plotted PDF, snaps and selections are
identical with and without stubs, and on five real drawings the view
renders the same.
Refs #1698, #1542
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Hatches are drawn before the wires and write depth, so a transparent
hatch above lines in draw order still hid them through the depth test.
Transparent hatches now go to their own renderer drawn after the wires:
the lines behind show through the blend and the lines in front still
win the depth test.
Fixes#1720
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- The background image texture is not an sRGB view: the image is drawn with
its own colours, as the solid and gradient colours are.
- An entered paper-space viewport shows its named visual style (Realistic,
...) in the render-mode picker, and a mode picked there is recorded as the
viewport's visual style.
- The view_background automation action takes offset=x,y and scale=x,y.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A fresh web-vs-desktop audit found the remaining gaps; this closes the
top ten.
Raster images: IMAGE attach read the picked file via image::open on a
name-only path (dead on wasm), and raster references inside opened
drawings decoded the same way. Picked bytes are now registered in a
name-keyed store (the PDF-underlay pattern), decoding runs from memory,
and dropped .png/.jpg/.bmp/.tif files register by name so unresolved
references in opened drawings re-resolve and render (and export to PDF).
Command gaps fixed with the bytes-at-pick / download_bytes pattern:
- WBLOCK now downloads the extracted block instead of failing on fs.
- SAVEALL saves every dirty tab instead of printing a no-op message.
- SCRIPT picks and reads the .scr file through the browser picker.
- CENTERLTYPEFILE picks the .lin/.dnl file and populates from text.
- Custom .ctb tables picked on web now parse, register for the
session, appear in the plot dialog list, and apply (load_named
previously returned builtin-only on wasm).
- LAYTRANS 'Load standards file' parses the picked drawing on the
existing web worker.
- INSERT 'from file' in the block palette parses picked bytes and
feeds the shared import path.
Text fidelity: TTF-named styles rendered as thin stroke glyphs on web
because the font database was empty. Fetched Noto faces are now
registered into it with a family alias table (Arial, Times New Roman,
Cambria, ... -> Noto Sans/Serif), so those styles render filled, shape,
export to PDF as vector fills, and appear in the style picker.
ViewCube labels request semibold again on web.
docs/native-vs-web.md: the hatch and font rows were stale (both work on
web now), the GPU row claimed WebGPU is used where available (the
backend is pinned to OpenGL), and the parallelism row missed the parse
worker — all corrected.
(cherry picked from commit c643a10e)
Ported onto main (#1717 review): conflicts resolved toward main — the
font-file lookup, PDF searchable text, purge service, block palette and
plot-style copy stay as they are on main; the web pieces (font registry
and aliases, session plot styles, SCRIPT/SAVEALL/CENTERLTYPEFILE picks,
block-from-file on the web) are layered on top.
A drawing whose plan content is built from blocks reaches thousands of
definitions, and each one took a buffer of its own for its segments and
another for its constants, plus one per placement run. The payloads are a
few hundred bytes, but every allocation costs a page: 14k definitions ask a
device for 1.8 GB to hold 12 MB of geometry. D3D12 runs out, the failed
allocations stay invalid, every later frame writes to them, and the session
dies in the staging belt.
The definitions of a build now share one arena: constants in a single
uniform buffer addressed by dynamic offset, segments in a single geometry
buffer addressed by first_vertex, placements in a single instance buffer
addressed by first_instance, with another arena started whenever one of the
three would outgrow the per-buffer budget. The cache holds the records
rather than the buffers, so a build the device rejected poisons nothing.
Co-Authored-By: Claude Opus 5 <[email protected]>
- Text styles keep their TrueType typeface through save and reopen (codec
reads and writes the STYLE record's ACAD extended data) (#1633)
- Unnamed shape-file entries no longer show in text style lists (#1634)
- Save As offers DWT 2018; NEW asks for the template to start from, while
Ctrl+N and the quick-access New run QNEW with the default template (#1635)
- An edge next to an earlier fillet of the same radius rounds together with
it (kernel); the solid stays drawn and pickable while edges are chosen
(#1625, #1107)
- A full cache rebuild drops the B-rep cache too, so an undone fillet does
not come back with the next edge operation (#1107)
- Selected solids in shaded styles highlight only their visible edges (#1031)
- VIEW ISO / SWISO / SEISO / NEISO / NWISO snap to the isometric corners
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Turning a 141k-entity layer off/on took 1.5-5 s; it now takes ~0.6-0.9 s.
- Off/freeze toggles go through invalidate_layer_visibility: entities on
the layer keep their tessellation memo (visibility gates them before
tessellation); only block users re-expand. No mesh recolor.
- The resident patch reuses a surviving memo entry instead of
re-tessellating, guarded by the same display-context hash as the full
build.
- The patch looked up `layout.order` linearly per re-shown entity, which
made turning a big layer back on quadratic.
- recolor_meshes_for_handles resolved materials for handles with no mesh.
- Wire arena: a hidden slab keeps its instances and sets a `hidden` flag
in its consts (collapsed in the vertex shader); an unchanged re-show only
clears the flag. Const writes are batched into merged spans.
- The `layer_visible` automation action indexed the document's layer
table instead of the sortable panel rows the Layer* messages use.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Each band segment was a rectangle ending square at its vertex, so the
outer corner of every joint showed a notch. Every segment end now carries
the tangent of half its turn, and the shader slides that end's corners
onto the joint's bisector so neighbours meet in a miter. Turns sharper
than the limit keep butt ends. Constant-width runs inside a polyline split
around its arcs no longer count as tapered, which had kept them unmitred.
Fixes#1565
Co-Authored-By: Claude Opus 5.5 <[email protected]>
- Lighting follows the view: the default source is a key light from the
upper left and a fill light from the lower right over an ambient term,
the other source one light from above; POINTCLOUDLIGHTING 1 lights the
front faces, 2 both.
- A selected or hovered point cloud shows its extents box.
- A move-gizmo arrow can be grabbed along its shaft; the grab offset is
taken on the press, so click-and-place and typed distances start from
the insertion point.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Points carry their world normal: POINTCLOUDLIGHTING lights them in the
shaded visual styles (1 lit from both sides, 2 with back faces dark),
from the view (POINTCLOUDLIGHTSOURCE 0) or the default lights (1).
The 2D wireframe style draws no cloud points: a found cloud shows its
extents and "Point clouds are not displayed in 2D visual style.",
or nothing with POINTCLOUD2DVSDISPLAY 1. While the view moves only
POINTCLOUDRTDENSITY per cent of the points are drawn; the points are
stored in an order where every leading part samples the whole cloud.
The colour map dialog lists its schemes by name.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Scans (.rcs) and projects (.rcp) are read and their points drawn as
screen-sized dots, coloured by the cloud's stylization. The cloud's
placement carries its scale and rotation; its extents and crops stay in
the scan's coordinates. A missing scan shows the extents with the saved
path. -POINTCLOUDATTACH / POINTCLOUDATTACH, POINTCLOUDCROP and the
POINTCLOUD* settings follow the reference's prompts and messages, the
Properties list is the reference's, and a locked cloud is left alone by
edits.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
DWFATTACH / -DWFATTACH and DGNATTACH / -DGNATTACH attach a DWF / DWFx sheet
or a DGN (V7 or V8) model with the reference's prompts; ATTACH routes the
file types to them. The sheet or model is read into vector paths, drawn as a
transparent underlay raster (contrast, fade, monochrome and background
adjustment apply) and offers its geometry to object snaps. Definitions are
kept in ACAD_DWFDEFINITIONS / ACAD_DGNDEFINITIONS, the External References
palette lists and reloads them, and image textures get line-keeping mip
levels so thin underlay lines stay visible when zoomed out.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
An image's transparency flag decides whether transparent pixels show what
is behind the image or draw in their stored colour: the image pass and PDF
plots now follow it (it was ignored, so every PNG alpha showed through).
TRANSPARENCY selects images and asks for the mode, defaulting to the first
image's current one.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Raster image textures use nearest magnification; PDF pages and OLE
pictures keep bilinear filtering.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The codec and kernel repositories and crates are renamed opencadcodec and
opencadkernel. Inside OCS they are depended on under short keys, so code
reads `codec::`, `kernel::` and the new `graph::` (opencadgraph, the node
graph engine). File-format and protocol spellings stay as they were: the
ACIS product id `acadrust`, the ACADRUST_* switches and the plugin
version keys `acadrust_version` / `acadrust_source`.
A button left of MCP on the command line opens a translucent node graph
over the viewport, with a palette of object, input, math, logic, list,
text and geometry nodes added by click or drag. An object node owns
drawing entities and its rows are the Properties panel's own rows for
them: the single-object page is extracted from refresh_properties as
entity_property_sections, values use the control protocol's
property_json, and writes go through set_property_value with a property
target override, so the panel, MCP and the graph share one read/write
path. A list reaching an object row makes one entity per element. Each
graph action is one undo step; the canvas pans and zooms.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Eliminate intermediate heap vector allocations during wire classification
and analytical instance extraction for circle and ellipse GPU pipelines.
1. Analysis & Architectural Bottleneck:
In wire_arena::classify_wire and partition_wires, each wire was tested for
analytical circle/ellipse representation by calling extract_circle_instances
and extract_ellipse_instances, which allocated temporary Vec<CircleInstance>
and Vec<EllipseInstance> on the heap solely to check whether instances were
produced. Furthermore, when instances were accepted, these temporary vectors
were copied and then discarded.
2. Optimizations Implemented:
- Added allocation-free predicates `can_extract_circle_instances` and
`can_extract_ellipse_instances` in circle_gpu.rs and ellipse_gpu.rs.
- Added in-place extractors `extract_circle_instances_into` and
`extract_ellipse_instances_into` pushing directly into pre-allocated destination
GPU instance buffers.
- In wire_arena.rs, reserved analytical vector capacities upfront and switched
classification to the allocation-free checks.
- In pipeline/mod.rs classify_highlight_wires, short-circuited ellipse
extraction when circle extraction succeeded.
3. Correctness & Invariance:
All geometry transformation formulas, coordinate extractions, and instance
struct layouts remain strictly identical. Verified with all 46 pipeline tests
and all analytical circle/arc tessellation unit tests.
4. Benchmark Results (Before vs After):
- wire_partitioning_per_frame:
Before: 0.18 ms (mean 0.19 ms, p95 0.38 ms)
After: 0.06 ms (mean 0.07 ms, p95 0.28 ms)
Improvement: -63.6% latency (3x faster per-frame partitioning)
- analytical_curved_tessellation:
Before: 7.79 ms (359.5k entities/s)
After: 6.87 ms (407.7k entities/s)
Improvement: -11.8% latency (+13.4% throughput)
- wide_and_tapered_arc_tessellation:
Before: 6.36 ms (235.8k entities/s)
After: 5.48 ms (273.8k entities/s)
Improvement: -13.8% latency (+16.1% throughput)
- analytical_curved_zoom_100x:
Before: 12.75M FPS (0.01 ms)
After: 13.97M FPS (0.01 ms)
Improvement: +9.6% FPS
Top Front Right/Left kept Y (north) as up when starting from Top,
tilting the base because orbit is turntable (never banks). Make
snap_to_direction deterministic: world +Z horizon except near +-Z
where north +Y is used, sign-preserved only for intentionally
inverted views. Expand opposite() from faces-only to all 26 regions
so already-there flip works for edges/corners.
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.
Keep fractional raster placement and matching integer texture coverage
through compositing. Apply parent clipping without changing the scale.
Fixes#1267
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]>
Brings in 54 commits from origin/main: a selection-performance overhaul,
surface thickening, GPU-native wide/tapered stroke rendering, continued
ocs_doc_api evolution, and assorted fixes/polish (block double-click
editing, DWG hatch/database repair, clipboard, collapsible Properties
sections).
Six real conflicts, all where this branch had independently re-applied
a fix upstream later revised or already carried natively:
- src/app/automation.rs, src/app/command_driver.rs: combined both
sides — kept this branch's sketch-constraint/named-parameter loading
on "new"/"open", adopted main's Scene::clear()/material_base_dir and
its newer rebuild_derived_caches() over the older bump_geometry().
Scene::clear() (src/scene/entity.rs) now also resets
sketch_constraints/named_parameters, so every caller gets this for
free instead of each call site repeating it.
- src/io/mod.rs, src/scene/render_graph.rs: adopted main's design
outright (commit 973af87d, authored by Hakan directly) — a large
MINSERT is no longer purged as "corrupt" at load, only its rendering
is bounded, preserving the source data. This branch's own earlier
design (reject at load) predates that revision.
- src/scene/pipeline/mesh_gpu.rs: adopted main's valid_rgba_size()
guard (a real robustness improvement over this branch's version).
- src/snap.rs: replaced wholesale with main's version — this branch's
copy was an explicit verbatim snapshot from an earlier upstream PR
(commit 4c264b8c) with zero constraint-system-specific content, and
main has since evolved it considerably further.
Also removed two now-dead duplicates the merge surfaced without a
conflict marker (both branches had independently fixed the same
things): the OPTIONS/OP command dispatch/autocomplete entries, and
automation.rs's open_finalizes_and_purges_like_the_ui_open_path test
(kept main's version, which covers strictly more: solid_models/
material_base_dir reset on "new").
Full workspace suite passes (812 lib tests + this branch's own 24
constraint/persistence integration tests) except the one pre-existing,
unrelated font-test failure already present on main. Zero clippy
findings in the hand-written merge resolution itself; the ~1,171
warnings clippy reports are pre-existing in the 54 absorbed commits,
not introduced here.
Co-Authored-By: Claude Sonnet 5 <[email protected]>
- Repurpose CircleInstance center_high.w and params.w for start_width and end_width (preserving exact 128-byte stride)
- Route wide polylines, tapered polylines with bulges, and donuts through split_mixed_polyline into GPU analytical arc instances
- Compute physical annular bands, width tapers, and round end caps in circle.wgsl using world-space distances to prevent 3D perspective foreshortening distortion
- Expand bounding quad margins dynamically with grazing-angle compensation to eliminate clipping
- Relax pick_tris check in wire classification and partition routines so wide entities retain CPU picking while drawing analytically
- Add unit and hardware GPU integration tests, including tilted 3D perspective rendering verification
Resolve curve intersections through cadkernel, retain empty exact results, gather analytic candidates independently of tessellation, and bound material uploads. Preserve large MINSERT data while limiting rendering work; remove obsolete review narratives.
Ports the fixes from fix/osnap-circle-arc-ellipse-intersection (and
this session's audit follow-up) onto this branch. All are independent
correctness/reliability fixes, unrelated to the parametric-constraint-
system work also in progress here.
1. OSNAP Intersection snap used a tessellated approximation instead of
exact geometry (GitHub discussion #1052): a circle/arc always
tessellates at a fixed 48 segments in the resident wire set
regardless of radius or zoom, so every circle/arc intersection snap
was off by construction. Added exact_circle_intersections, a
closed-form, frame-independent circle/arc-vs-circle/arc solver,
wired into both Intersection-snap code paths ahead of the
tessellated sweep. This branch's TangentGeom predates the
PlanarEllipse variant added upstream, so the ellipse/line
generalization from the upstream fix isn't ported here — this
covers the literal reported bug (a circle-circle case) using only
variants this branch already has.
2. Unbounded MINSERT allocation: array_offsets() reserved one Vec
element per array instance with row/column counts parsed as
unchecked u16s — a corrupt 65535x65535 pair requests ~96 GiB.
Fixed with a documented instance budget (20,000), enforced at both
the load-time corrupt-entity filter and a render-graph failsafe.
3. MINSERT spacing was scaled by the block's own scale:
insert_instance_transform composed the offset translation before
the instance transform, so x/y block scale multiplied the row/
column spacing too. Fixed by orienting the offset with rotation +
OCS only, then translating the already-scaled instance transform.
4. Automation "open" skipped the finalization (block-origin
normalization, raster-path resolution, source_path) and corrupt-
entity purge a UI open always runs. Added
io::load_bytes_finalized(path, bytes), used by automation's "open"
handler, which now also reports the purge count.
5. Material maps were uploaded without checking
device.limits().max_texture_dimension_2d, unlike raster-image
rendering (which already tiles to that limit). A mesh material is
sampled by UV, so it can't use the same tiling trick — instead,
upload_rgba_texture now downscales an oversized image before
upload, falling back to the 1x1 material color if malformed.
6. Sketch-constraint persistence/undo, audit-found while reviewing
this branch's own constraint system:
- Clearing every constraint in a scope after a prior non-empty save
left the stale XRecord in place, resurrecting the deleted
constraints on the next load. materialize_sketch_constraints_for_
save now actively removes the XRecord when a scope goes back to
empty.
- ERASE removing an entity took its constraints with it (design
doc's deletion policy), but sketch_constraints lives on Scene,
outside document/document.objects, so nothing captured that
mutation for undo — the erased geometry came back on undo, but
its constraints didn't. refresh_sketch_constraints's deletion
pass now records the touched scope's before-image into the same
UndoRecording the entity removal rides in, threaded through a new
sketch_constraints channel on DeltaSnapshot, so one undo restores
both together. HistorySnapshot::Delta is now boxed (clippy
flagged the resulting size skew across undo/redo stack variants).
31 new/adapted tests across all six fixes. Full workspace suite passes
except the one pre-existing, unrelated font-test failure also present
on main.
Co-Authored-By: Claude Sonnet 5 <[email protected]>
Material map images (diffuse/specular/reflection/opacity/bump/
refraction/normal) were decoded at their original file dimensions and
uploaded to wgpu without checking device.limits().max_texture_dimension_2d,
unlike raster-image rendering, which already tiles to that same limit.
An oversized material map produced a wgpu validation error and left
the material unusable; the global error handler prevented a process
abort, but never produced a valid texture either.
A mesh material is sampled by UV across a surface, so it can't use the
same per-quad tiling trick raster images use. Instead,
upload_rgba_texture now checks the decoded image against the device's
texture-size limit and downscales it (aspect-preserving, via
image::imageops::resize) before upload when it's oversized, falling
back to the existing 1x1 material color if the decoded buffer is
malformed (its length doesn't match its reported dimensions).
3 new tests cover the resize math (aspect ratio preserved, both
dimensions respect the limit) and the malformed-input fallback.
Co-Authored-By: Claude Sonnet 5 <[email protected]>
