Type-as-you-go suggestion popup above the input that filters known
command names by substring (so `LEADER` surfaces both LEADER and
MLEADER). Up / Down navigates the list, Enter dispatches the
highlighted entry, clicking a row dispatches directly. Esc on a
non-empty input clears the buffer instead of escalating to
command-cancel.
The name corpus comes from `inventory::submit!` blocks placed next
to each `CadCommand` impl rather than a hand-maintained static list.
Adding a new interactive command means dropping one `inventory::submit!`
line in its own module file — the autocomplete picks it up at the
next compile. 98 submissions seeded across 60 module files;
`command::all_registered_command_names()` flattens them at runtime.
One-shot dispatch arms in `app/commands.rs` (NEW, OPEN, SAVE, UNDO,
…) don't have a `CadCommand` impl and are therefore absent from
autocomplete by design.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
Add a ▸/▾ toggle button on the right edge of the input row. Tapping
it pops up the full backlog of everything the command line has pushed
since startup — useful once the on-overlay rows have faded out.
The dropdown is a fixed-height scrollable panel rendered above the
input pill. `anchor_bottom` keeps the newest entry in view on open
and `opaque` blocks wheel events from bubbling through to the
viewport shader behind it (without it, scrolling the history zoomed
the drawing). Submitting a command closes the dropdown so the new
prompt is immediately visible.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
The overlay used to keep the last four history lines on screen
indefinitely, leaving stale command output cluttering the viewport.
Stamp each `HistoryEntry` with its `Instant` of insertion and filter
the rows shown on the overlay to entries younger than 3 s.
`CommandLine::has_visible_history()` gates a per-frame tick
subscription so the view re-renders and the entry disappears
on time. The subscription auto-stops once no entry is fresh enough,
keeping the idle CPU draw at zero (#18).
Full history stays in the backing `Vec` — the upcoming dropdown
button will surface it on demand.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
The command-line panel used to consume a full-width row at the bottom
of the main column, taking vertical space away from the drawing area
even when no command was active. Move it onto the viewport stack as a
fixed-width 720 px pill aligned to the bottom-centre, with rounded
corners so it reads as a floating element. The status bar still owns
the very bottom of the window.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
Two follow-ups to the always-on keyboard work:
1. **Repeat last command on empty Submit.** Pressing Space (or Enter)
after a command ended did nothing — the trailing-space submit path
reached `CommandSubmit`, found no active command and no buffered
text, and bailed. `CommandFinalize` already had the repeat-last
shortcut via `tab.last_cmd`; mirror it inside `CommandSubmit` so
the space-as-submit path picks it up too.
2. **Drop the per-event focus_cmd_input calls** added in the previous
commit for `ViewportLeftPress`, `ViewportClick`, `TabNew`,
`TabSwitch`, `RibbonSelectTab` and `RibbonToolClick`. The global
keyboard subscription routes unhandled character / backspace
events into the buffer regardless of focus, so the only places
that still need an explicit refocus are `cmd_result.rs` (command
lifecycle), `boot()` (initial caret), and the per-char append
handlers (caret feedback while typing).
Co-Authored-By: Claude Opus 4.7 <[email protected]>
Two behaviour changes that bring the command line in line with
established CAD workflow:
1. **Always-on keyboard capture.** The input was blurred whenever no
command was running, so typed letters went nowhere until the user
clicked into the field. Replace the blur with a re-focus on every
command lifecycle transition (cmd_result.rs) and on viewport / tab
/ ribbon clicks (update.rs), focus the input at startup, and add
a global key subscription that catches any unhandled character key
or backspace and routes it into the buffer via two new messages
(CommandAppendChar / CommandBackspace) — so the command line keeps
accumulating keystrokes even if focus parks somewhere not
focusable. Removes the unused blur_cmd_input helper.
2. **Space submits.** Typing a space at the end of the input now
submits the buffer the same way Enter does. Only exception: an
active command that's currently collecting free-form prose
(TEXT / MTEXT / DDEDIT body, ATTDEF prompt / default value,
INSERT attribute value, ATTEDIT cell). Those advertise the new
`CadCommand::wants_text_with_spaces()` flag so the Space character
reaches the buffer as a literal.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
`HatchModel.boundary` stores multi-region hatches as a single Vec
with NaN-NaN sentinels separating each disconnected sub-path
(documented on the field). The paper-canvas hatch fill ignored the
sentinels and forwarded every vertex straight to the iced canvas
path builder. When a layout contained a multi-region hatch, lyon
received a polygon containing NaN coordinates and panicked with
`Tessellate path.: Internal(IncorrectActiveEdgeOrder(4))` from
inside `iced_wgpu::geometry::Frame::fill`, taking the whole app down
the first time Layout 1 painted.
Walk the boundary instead, opening a fresh sub-path on each sentinel
and closing whichever was active. Each region is now filled as its
own closed polygon. Reproduces with #22.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
The polyline arm of the hatch wire outline guarded its NaN separator
with `if count > 0 && pts.len() != start_idx`. `start_idx` was assigned
to `pts.len()` two lines earlier with nothing in between to grow `pts`,
so the inequality was structurally always false — the separator was
never emitted.
When a hatch carried more than one boundary path (multi-region hatch),
the renderer drew a straight segment from the last vertex of path A to
the first vertex of path B because the two polylines lived in `pts`
back-to-back with no NaN break. The result was a visible stray line
between hatch regions.
Switch the polyline arm to the same `if !pts.is_empty()` guard the
Line / CircularArc / EllipticArc arms use, and capture `start_idx`
*after* the separator so the closed-loop close-back logic still points
at this polyline's first real vertex.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
YAML issue forms with required fields so new reports arrive with the
context needed to act on them (version, OS, reproduction steps for
bugs; problem-before-solution for features). Auto-applies the matching
label.
blank_issues_enabled: false forces use of one of the three templates.
config.yml also surfaces the Patreon link as a contact option.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
The Hatch boundary-mirror transform called
`crate::scene::transform::reflect_xy_point(&mut x.x, &mut x.y, p1, p2)`
nine times with rustfmt expanding each long-path call into a six-line
block. Total: ~55 lines for nine reflects.
Add a local `use crate::scene::transform::reflect_xy_point as reflect;`
at the top of the closure so each reflect collapses to a single line.
-19 lines, no behavior change.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
The same three-arm `match entity { Solid3D | Region | Body }` block
appeared at five callsites — twice in tessellate.rs (point_of_reference,
fallback wires) and three times in scene/mod.rs (mesh tess dispatch,
apply_grip pre/post position).
Add three accessors in entities/solid3d.rs:
point_of_reference(e) → Option<&Vector3>
fallback_wires(e) → Option<&[Wire]>
tessellate_volume(e, color, facet_res) → Option<MeshLodSet>
Callsites shrink from 5–8 lines to one call. The apply_grip path
collapses two parallel pre/post matches plus the entity-list `matches!`
guard.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
Two near-identical match-on-EntityType tables lived far apart:
scene/properties.rs::entity_type_name returned mixed-case UI labels;
app/commands.rs::entity_type_name returned uppercase DXF record names.
Both enumerated ~30 variants; adding a new entity meant remembering to
touch two files in two crates' worth of source layout.
Move both into a single entities/names.rs as `ui_name()` and
`dxf_name()`. Adjacent so a new variant gets both labels added at once;
function names make the intended audience explicit at the callsite.
scene/properties.rs and app/commands.rs each lose their local match
plus one re-export, gaining a `crate::entities::names::*` call.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
Every entity file ended with the same three-trait wrapper block:
impl Grippable for T {
fn grips(&self) -> Vec<GripDef> { grips(self) }
fn apply_grip(&mut self, ...) { apply_grip(self, ...) }
}
impl PropertyEditable for T { ... }
impl Transformable for T { ... }
— 23 lines of pure delegation around five same-named free functions.
Replace with `crate::impl_entity_basics!(T);` (a macro in
entities/traits.rs that expands to the three trait impls).
Applied where the file already follows the free-fn delegation
convention (Arc, Circle, Ellipse, Insert, Leader, Line, LwPolyline,
MultiLeader, Ole2Frame, Point, Spline, Viewport). Files that inlined
the trait method bodies directly (Ray, XLine, Shape, Solid, Tolerance,
MLine, Mesh, Solid3D/Region/Body, Table, RasterImage, Wipeout,
Underlay, Hatch, Polyline, Polyline2D, Polyline3D, Dimension) left
alone — they already skip the indirection. Text/MText/Attribute use
the with-text-style flavour and will move once that variant is needed
beyond the existing two manual impls.
A second macro `impl_entity_basics_with_text_styles!` is included for
the text-like flavour but no entity uses it yet.
Net: -204 lines.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
The same DXF/DWG bulge → (center, radius, sweep) formula was inlined
in six places (hatch outline, polyline tess + truck path, lwpolyline
×4, scene hatch fill, explode bulge_to_arc) — each with subtly
different sweep-direction handling for the bulge sign.
Add `entities::common::BulgeArc { center, radius, start_angle,
end_angle, sweep }` with one `from_bulge(p0, p1, bulge)` constructor
and a `sample(t)` helper. Callsites collapse to a few lines.
Net: -121 lines, one canonical sweep-direction convention shared by
all consumers.
array.rs (the path-along-curve f32 helper) keeps its inline math —
its sign convention is inverted relative to the others and changing
it without exercising the array-path command risks regressing the
output.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
Two free helpers in app/commands.rs lived as big match-on-EntityType
blocks: massprop_entity (Circle/Arc/Line/LwPolyline/Ellipse area +
perimeter + centroid, ~140 lines) and entity_text_content /
replace_entity_text (Text/MText/AttributeDefinition/AttributeEntity).
Move each to its own trait in entities/traits.rs and implement
per-entity:
MassPropsCalc — Circle, Arc, Line, LwPolyline, Ellipse
TextContent — Text, MText, AttributeDefinition, AttributeEntity
EntityTypeOps gets dispatch shims that route to the trait via the
existing macro, so callsites just call entity.mass_props() /
entity.text_content() / entity.replace_text(search, rep).
The remaining free helpers in commands.rs (entity_list_details,
entity_extra_info, entity_type_name, build_data_extraction_csv) are
single-command formatting helpers tightly coupled to LIST / EXTRACT /
text-format output. Distributing them as per-entity methods would
inflate every entity file with command-side formatting concerns
without making the command-side any cleaner — left in place.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
Five chunks moved out of the 5965-line scene/mod.rs to live with the
entities they belong to:
- wide_lwpolyline_fills → entities/lwpolyline.rs::wide_fills
- wide_polyline2d_fills → entities/polyline.rs::wide_fills
- polyline_segment_fill → entities/common.rs (shared by both above)
- text_obb_corners_native + text_baseline_points + text_greek_obb_tris
→ entities/text_support.rs (handles Text/MText/Attribute alike)
- append_insert_attribute_wires → entities/insert.rs
Net: scene/mod.rs 5965 → 5404 lines (~560 lines extracted). Three
helpers in scene/render.rs / scene/block_cache.rs widened from
pub(super) to pub(crate) so the relocated functions can still reach
them. scene::entity_aabb also widened for the same reason.
scene::tessellate_entity_dim_text stays put — it's a thin wrapper
around the private tessellate_entity dispatcher and moving it would
require exposing tessellate_entity's full signature across the crate.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
Each of the five dispatch functions in `entities/traits.rs` was a
hand-written `match self { EntityType::Line(e) => ..., EntityType::Arc(e)
=> ..., ... }` block listing ~35 variants. Adding a new entity meant
adding five match arms across the file.
Replace with a `dispatch!` macro that takes `(self, |e| body, [variant
list], _ => default)`. Each dispatch function keeps its own variant
list (the supported sets differ per trait), but the per-arm
boilerplate is gone.
File shrinks from 386 to 145 lines. Adding a new entity now means
adding one identifier to one (or more) of the variant lists.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
The trait isn't deprecating anything — it carries the path entities
take when they're not covered by the truck topology pipeline
(Viewport, Insert, Hatch outline, Ole2Frame). "Fallback" names the
purpose; "Legacy" implied retirement that isn't actually on the
roadmap.
Same rename for LegacyGeometry → FallbackGeometry,
legacy_geometry() → fallback_geometry().
Co-Authored-By: Claude Opus 4.7 <[email protected]>
Each entity's tessellation now lives alongside its definition in
entities/*.rs, attached to the entity type via a local trait:
LegacyTess — Viewport / Insert / Hatch / Ole2Frame
(fallback path for entities not on the truck pipeline)
LeaderTess — Leader
MultiLeaderTess — MultiLeader
DimensionTess — Dimension (carries its dim-only helpers too:
~1700 lines of dim_geometry, dim_text_*, format_*,
tolerance, suppression, etc.)
scene/tessellate.rs keeps only the cross-entity dispatcher plus the
shared dim machinery reused by leader/multileader/dimension
(ArrowKind, DimGeom, append_arrow, arrow_from_block, add_segment,
add_polyline, normalized_or, push_tri, color_or_inherit, entity_z,
offset_snap_pts). Cross-entity arc helpers + aci_to_rgba moved to
scene/tess_util.rs.
scene/tessellate.rs shrinks from ~3300 to 809 lines. Compile clean,
all 9 tests pass.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
Bumps bytemuck 1.14->1.25, glam 0.27->0.33, rfd 0.15->0.17,
windows-sys 0.59->0.61, ureq 2->3. ureq 3 changed the agent API
and feature flags, so update_check.rs switches to Agent::config_builder
and body_mut().read_to_string(), and the feature toggles from tls
to rustls.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
Joining the OpenAEC Foundation (open-aec.com). Crate/struct become
OpenCADStudio, user-facing strings use the spaced "Open CAD Studio"
form, repo URLs point to HakanSeven12/open-cad-studio. Logo (window
icon + assets/logo.svg) redrawn with OCS initials. Drop examples/
binaries — unused.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
mapped_at_creation, rayon parallel emit, instanced rendering, and
Unorm8x4 color packing all landed. Per-segment upload dropped from
576 B to 76 B (~7.6x). No follow-ups left.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
Step 4 (final) of the wire pipeline cold-open optimization plan. Color
shrinks from `[f32; 4]` (16 B) to `[u8; 4]` (4 B) and the vertex attribute
format flips to `Unorm8x4`; the GPU still hands the shader a `vec4<f32>`
in [0, 1], so wire.wgsl needs no change. 8-bit per channel is below
display precision.
Net: 76 B per instance, ~14 % smaller than the Step-3 instance layout and
~7.6× smaller than the pre-Step-3 per-vertex format. Wire upload memory
and PCIe bandwidth on cold open scale down accordingly.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
Step 3 of the wire pipeline cold-open optimization plan — the biggest
single lever.
Each segment used to upload as 6 vertices × 96 B = 576 B with the segment
payload (pos_a/pos_b/color/half_width/distance/pattern_length/pat0/pat1)
duplicated across all six corners of the quad. Wire pipeline now uploads
one INSTANCE per segment (88 B) and the vertex shader derives `which_end`
and `side` from `@builtin(vertex_index)` against a 6-entry corner table.
Net effect on cold open: ~6.5× less wire vertex memory and PCIe upload,
plus a smaller Vec to allocate and copy on the CPU side.
Wire instance attributes also collapse distance_a / distance_b into a
single mix() inside the vertex shader, so per-segment distance still
interpolates correctly for the GPU-side linetype dash test.
Touches the wire, wire_xray, and face3d-edges draw passes.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
Step 2 of the wire pipeline cold-open optimization plan. Extracts the
per-wire vertex expansion into a free function and drives it from
`from_batch` via `par_iter().map(...).reduce(...)`. Per-wire emission is
already independent (no shared state, no order coupling — `block_cache`
groups wires by style upstream), so the parallel-then-concatenate shape
is correctness-preserving.
Cold-open wire upload now scales with available cores; a drawing with
many short wires shows the biggest win. Also collapses two near-identical
emission loops (`from_batch`, `build`) into one shared helper.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
Step 1 of the wire pipeline cold-open optimization plan. Replaces
`create_buffer_init` (intermediate staging-buffer memcpy) with a
`mapped_at_creation` buffer the build path writes into directly.
Cuts one alloc + one memcpy per wire buffer; on cold open with hundreds
of MB of wire data this shows up immediately as lower peak memory and
faster upload.
Guarded against size-0 buffers by allocating a single-vertex placeholder
when the wire is empty (the renderer already skips draws on vertex_count
below the per-segment minimum).
Co-Authored-By: Claude Opus 4.7 <[email protected]>
Documents the 4-step plan to reduce first-file-open latency by attacking
wire vertex upload volume: mapped_at_creation, Rayon parallel emit,
instanced rendering (the big lever, ~6.5x memory cut), and per-instance
packing. Each step lands as its own commit.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
LWDISPLAY (Show Lineweight) flipping the toggle no longer triggers a full
retessellate. `render_style_for` always bakes the entity's resolved (layer-
inherited) lineweight, and the wire shader reads `Uniforms.lwdisplay_enable`
to either honour the per-vertex half_width or collapse the quad to a 1 px line.
Side effect: PDF export now uses the real lineweight regardless of LWDISPLAY,
matching AutoCAD's plot semantics (LWDISPLAY only affects screen render).
Co-Authored-By: Claude Opus 4.7 <[email protected]>
Wide LWPolyline/Polyline2D fill bands now sit on the centerline:
- polyline_segment_fill arc center matched explode.rs / acadrust convention
(signed d_perp = r·(1-b²)/(1+b²), handles major arcs correctly)
- per-vertex hw0/hw1 taper applied along arc; tess density 24/circle
- synced_hatch_models subtracts hatch_offset from band boundary so it lives
in the same offset-rel space as the wire tessellation
LWDISPLAY:
- new command `LWDISPLAY [ON|OFF]` toggles header.lineweight_display
- LWT status-bar pill (between ORTHO and POLAR) drives the same flag
- bump_geometry() on change so the viewport re-tess immediately
Co-Authored-By: Claude Opus 4.7 <[email protected]>
acadrust's DWG reader read the wipeout/raster-image clip boundary
points only to advance the bit cursor and then discarded them, so every
wipeout fell back to `Wipeout::new()`'s default placeholder rectangle
(`-0.5..0.5`). For files that store the real polygon mask in clip
vertices (anteen.dwg's `DIMBLOCK` style: u_vector ≈ 643k, size = (1, 1),
clip polygon ~0.0016×0.0033 in pixel space), the wipeout rendered at the
full image bbox — hundreds of thousands of units wide — instead of the
intended ~1000×2000 mask.
Pinned to the new `fix/wipeout-clip-vertices` acadrust branch (carries
the upstream parser fix). On the H7CAD side, the polygon mapping now
follows the DXF convention:
x_off = (clip.x + size.x/2) × u_vec
y_off = (size.y/2 − clip.y) × v_vec (image-Y points down)
…in `wipeout_boundary_2d` (fill), `Wipeout::to_truck` (border line),
and `Wipeout::grips` (vertex handles) so all three stay in sync. The
fill path also explicitly closes the polygon loop — without it the GPU
`in_polygon` ray-cast skipped the v(n-1)→v(0) edge and the solid mask
bled far past the boundary.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
The update notice dialog now mirrors what GitHub publishes: side-by-side
Installed → Latest version cards (the new one accent-tinted to draw the
eye) and a scrollable release-notes panel under them. Light markdown
rendering covers `##` headings, bullet lists, and inline bold/code
markers — enough for the GitHub release body format we ship.
- update_check::check_for_update() returns UpdateInfo { version, body };
body is parsed from the releases API JSON with a tiny custom
string-field extractor (\n / \" / \u handling) so we keep the
serde_json-free dependency footprint.
- Window grows from 480x240 to 560x460 and becomes resizable to make
room for the notes scroller.
- Layout tightens: header bottom padding drops 14 → 6, notes heading
10, and the notes block is wrapped in a Fill-height container so it
claims every leftover pixel — no more dead space between heading and
footer.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
Both the on-screen paper canvas and the PDF / print pipeline only knew how
to draw the boundary outline of pattern hatches (ANSI31, ANSI37, dotted
brick, etc.) — the actual hatch lines were a GPU-only feature, so any
paper-space view or PDF export showed a hollow polygon where there should
be ruled fill.
Add a CPU rasteriser as `HatchModel::pattern_segments()`:
- builds an edge list from the boundary, splitting at NaN sentinels and
closing each sub-path so even-odd inside tests handle islands / holes;
- for every PAT family computes the world-space step vector
(`(dx,dy)` rotated into the family's frame, scaled by `model.scale`)
and the inclusive k-range bounded by the boundary AABB plus a 4096-
line safety cap;
- intersects each line with every boundary edge, sorts the hit
parameters, walks them in pairs (even-odd) and emits one segment per
in/out interval. Dash patterns are walked along each clipped span with
phase preserved via `t mod period`, matching the GPU shader's
absolute-t dash evaluation;
- pattern origin (`family.x0 / y0`) is rotated by `angle_offset` and
scaled the same way as `hatch_batched.wgsl`, so PAT patterns with
non-zero origins (brick / offset herringbone) line up with the GPU
render.
`paper_canvas::draw_hatch` and `pdf_export::emit_hatch` now route
pattern hatches through this rasteriser; solid / gradient hatches keep
their existing fast paths. The previous outline-only behaviour was a
visible regression on every layout containing a ruled fill.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
The PDF / print pipeline only walked `WireModel.points` (lines), so every
hatched region disappeared the moment a sheet left H7CAD. Route the same
`paper_canvas_hatches()` / `paper_canvas_wipeouts()` lists that the
on-screen canvas consumes into `pdf_export::export_pdf` and
`print_to_printer::print_wires`, and emit each one as an
`Op::DrawPolygon` with the EvenOdd winding rule so multi-ring boundaries
(islands / holes) render correctly:
- Solid → fill the boundary polygon in the hatch colour.
- Pattern → stroke the outline (PAT line families stay procedural;
follow-up work will rasterise them into segments).
- Gradient → fill with the averaged colour, mirroring paper_canvas.
Wipeouts emit first, then hatches, then wires — same z-order as the
on-screen canvas, so wires draw on top of the fills they belong to.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
Lift the MText parser + renderer into a reusable `layout_mtext` helper in
`text_support` and re-point the remaining text-bearing entities at it, so
every inline format code (`\f`, `\C`/`\c`, `\H`, `\W`, `\Q`, `\T`, `\A`,
`\p…`, decorations, stacked fractions, paragraph indents, tab stops, …)
reaches the stroke output uniformly.
Per-entity changes:
- TABLE cell ([entities/table.rs]) — anchor + alignment math now feeds
`layout_mtext`; `\C7;Red`-style codes no longer render as literal
glyph noise.
- MLEADER MText content (both entity path in [entities/multileader.rs]
and the LOD-aware path in [scene/tessellate.rs]) — drops the
`strip_mtext_codes` + per-line `tessellate_text_ex` chain; the LOD 0
baseline / LOD 1 greek rect / LOD 2 stroke ladder and the text-frame /
bg-fill rectangle now read line widths from the layout.
- ATTRIB / ATTDEF — when `mtext_flag` is MultiLine / ConstantMultiLine
the value goes through the MText pipeline (SingleLine stays on the
Text-style anchor maths because the DXF spec disallows codes there).
- DIMENSION text_override — `dimension_text_entity` emits a synthetic
MText (instead of Text) whenever the value contains MText escapes;
the LOD ladder via `tessellate_entity_dim_text` handles both kinds.
Infrastructure:
- `MTextVAnchor` gains `MiddleOfTopLine`, `MiddleOfBottomLine`,
`BottomOfTopLine` so MLEADER's 9 attachment variants all map cleanly.
- `MTextLayout` returns `line_widths` / `line_height` / `v_offset` so
callers that draw chrome (frames, fills, baselines, greek rects)
share the same numbers as the rendered strokes.
- `mtext_line_count` re-routed through the new parser so LOD row counts
and rendered row counts can never disagree.
- Legacy helpers (`strip_mtext_codes`, `split_mtext_lines`,
`measure_mtext_chars`, `word_wrap`, `cxf::measure_text`, the
`v_offset_for_attachment` MLEADER offset table, `MTextLine::plain_text`)
deleted — all callers migrated.
Audit (TEXT_FORMAT_CODE_INTEGRATION.md) removed; its plan is implemented
here.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
A pinned "Start" tab now lives at index 0. It cannot be closed, opens
on launch, and replaces the model-space viewport with a welcome page —
similar to Thunderbird's first-run screen.
- Boot opens with only the Start tab; Drawing1 is no longer auto-spawned.
- The Properties left panel becomes a Recent Documents rail on the
Start tab. Recents persist to disk
(%APPDATA%/H7CAD/recent.txt on Windows, $XDG_CONFIG_HOME/H7CAD/recent.txt
on Linux, ~/Library/Application Support/H7CAD/recent.txt on macOS), up
to 20 entries, saved on every open and rehydrated at boot.
Each row opens or removes the file in place.
- The Support ribbon group (Donate / Report / About / Changelog) is
retired; the same actions surface on the welcome page instead. The
Donate button is the primary call-to-action — solid #B03020 brand
fill with a soft brand-tinted ambient glow halo behind the action row
(Thunderbird-style coloured glow against the neutral page).
- Drawing-only overlays (mouse-area, viewcube, navbar, dynamic input,
selection markers) are suppressed on the Start tab so its clicks
don't trigger CAD commands. OpenFile's "current_is_empty" check
refuses to replace the Start tab, and TabClose ignores requests on
it.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
Moves the Dimension and Table baked-block paths up into
`tessellate_entity` so sub-Text/MText recurse through the same LOD
ladder (baseline / greek / full) the top-level entities already use,
and routes the synth-dim text path through `tessellate_entity` as well
so files that omit `block_name` (e.g. typical DXFs) also get LOD.
Extends the text-height LOD branch from Text/MText to
AttributeDefinition, AttributeEntity, and Tolerance — the OBB helpers
gained the same coverage so the baseline/greek rect lines up.
When a sub-entity inside a baked dim/table block has ByBlock colour,
the recursive call's resolved colour is replaced with the parent's so
dimension text no longer renders in the raw ByBlock fallback hue.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
AutoCAD bakes a per-instance block with the final, displayed geometry
into Dimension and Table entities (typically *D### / *T###, but custom
names also occur — sample-file-H uses DIMBLOCK*-4NP). The block's text
already encodes DIMTXT * DIMSCALE (or the table's scale factor), so
synthesising the entity from the dim/table style on top of that
double-applies the scale.
When the referenced block exists with content, render it directly:
- Dimension: dim.base.block_name -> block_record
- Table: block_record_handle -> block_record
The synthesis fallback stays in place for files where the block is
missing or empty.
Fixes sample-file-H.dwg dimensions rendering at 60000 instead of 200
(DIMTXT=199.196 * DIMSCALE=304.8).
Co-Authored-By: Claude Opus 4.7 <[email protected]>
- MultiLeader: stop double-scaling when ctx.text_height (WCS-baked) is set;
use it as-is and only multiply by scale_factor + anno on the fallback path.
ctx.text_width gets the same treatment.
- Text / Attribute: width_factor and oblique_angle are entity-authoritative.
The previous entity*style multiplication / addition mis-rendered any text
whose TXSTYLE had a non-default width_factor or oblique_angle (RR 0.6,
STYLE2 15°, Romans/ARIEL/A32 0.65-0.8 in the sample file).
- MText: honour TXSTYLE is_backward (negative width_factor mirror) and
is_upside_down (180° rotation). measure_mtext_chars uses positive
magnitude so anchor math stays correct under backward text.
- Tolerance: use tol.dimension_gap when set instead of the hard-coded
0.35 * height heuristic.
- Table: cell text height / font / alignment resolve through
CellContent -> CellStyle -> TableStyle.{title|header|data}_row_style
-> compiled-in defaults. Row classification respects the suppressed
flags. Cell + content rotation applied.
Co-Authored-By: Claude Opus 4.7 <[email protected]>
