PLOT/EXPORTPDF started from the menu, the toolbar, Ctrl+P or the command line had no path,
so the PDF was built in memory and dropped. They now open the save dialog (PDF filter first)
and write the file there. JSON calls from scripts, the control channel and MCP still never
open a dialog. Ctrl+P was shown in the menu but not bound; it is now.
eframe's default wgpu instance enables PRIMARY | GL on Windows. Creating
the OpenGL backend can crash inside AMD's atio6axx.dll before the window
appears, so the app exits silently at launch. When WGPU_BACKEND is not
set, the Windows instance now enables only DirectX 12; an explicit
WGPU_BACKEND is still honoured. macOS, Linux and web are unchanged.
Same fix as storytold/designcraft#168 and storytold/filmcraft#179.
Claude-Session: https://claude.ai/code/session_01LyVbT6Vv4BwWZ4hY8973qD
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Linux: present without a vsync queue so the crosshair follows the mouse (#25)
The canvas hides the OS pointer and draws its own crosshair, so every frame
the swapchain holds back shows as cursor lag. Under vsync (FIFO) wgpu asks for
one queued frame, but Vulkan clamps it to the driver's minimum swapchain size:
three images on X11 (Linux Mint's Cinnamon session), i.e. two frames queued
behind the one on screen. Windows DX12 honours one. The canvas also requested
a repaint on every frame while hovered, which kept that queue full.
- gpu::surface_config: AutoNoVsync (Immediate, then Mailbox, then FIFO; valid
on every surface) on Linux/BSD, frame latency 1. CADCRAFT_VSYNC=1 restores
vsync, CADCRAFT_VSYNC=0 selects low latency elsewhere.
- gpu::FrameCap (240 fps) when presenting without vsync: nothing else paces
frames on X11, and a 1000 Hz input stream rendered ~900 frames per second.
- No continuous repaint while hovering the canvas; input repaints by itself.
The status-bar message schedules its own repaint to clear on time.
- Hover/selection highlight looks up primitives in a per-list sorted index
instead of scanning the whole display list every frame (~2.5 ms per frame
on a 270k-entity drawing).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Move frame-cap and status-repaint hunks away from other PRs' to avoid merge conflicts
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* Light and System interface themes (System follows the OS appearance live)
CADCraft had one hard-coded dark interface: Tokens::get() always returned
Tokens::DARK and Visuals::dark() was installed once at startup.
- theme.rs: Tokens::LIGHT (light grey chrome, dark text) beside the unchanged
Tokens::DARK; Tokens::get() returns the palette of the theme in use (a
UI-thread thread-local set once per frame). Both egui styles are built from
the tokens and egui is switched to the shown theme, so standard widgets,
menus and popups follow too. The drawing area (canvas, grid, axes,
selection, grips, ViewCube, UCS icon) keeps its model-space colours in both
themes: drawing colours are not part of the interface theme.
- ThemePref { System, Light, Dark } in UiState (`theme`), resolved every
frame against ctx.system_theme(), so System follows the OS (winit, or the
browser's prefers-color-scheme) live without touching the saved choice.
No reported appearance: System shows Dark (SYSTEM_FALLBACK). Default stays
Dark (today's look).
- View > Interface Theme > Use System Setting / Light / Dark (ui.theme.system,
ui.theme.light, ui.theme.dark) and ui.theme {theme} for scripts and the
control channel; ui.inspect reports the theme shown.
- Hard-coded colours on the command line (history, border, keywords,
AutoComplete list), the status-bar message and the layer status icon go
through tokens; overlays drawn on the canvas use a new canvas_ink token.
- Desktop app: the choice persists via eframe storage (`persistence`
feature, key cadcraft.prefs); window geometry and egui memory are not
persisted, as before.
Fixes#59.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Document the theme field away from other PRs' edits to avoid merge conflicts
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
`convert` and `run --export` always fitted the model-space extents, so
`zoom`, `pan` or `view.set` in a run had no effect on the image and there
was no way to export a region.
- io: `write_framed(d, name, window)` frames PNG, SVG and PDF on a
model-space rectangle (fitted without margin, centred, no distortion);
`write` is unchanged and still fits the extents. `check_window` rejects
non-finite, inverted, zero-size and out-of-range (±1e12) windows.
- CLI: `run --export` shows the current view when the commands changed
it, otherwise the extents as before. `--window x0,y0,x1,y1` and
`--view extents|current` on `convert` and `run` choose the framing
explicitly. The usage text and README say how images are framed.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* cadcraft-cli mcp: --connect without HOST:PORT is an error, not a headless session (#103)
`mcp --connect` with no address (or an empty or flag-like one) fell through
to the headless backend, so a client meant for the app got successful edits
in a separate drawing. The mcp arguments now go through parse_mcp_args:
--connect must be followed by a non-empty value that doesn't start with `-`,
unknown arguments and a repeated --connect are rejected, and every error
exits non-zero before any MCP request is served. Plain `mcp` is still
headless; a real address still connects, and an unreachable one still fails.
- Unit tests for the parser; integration tests (tests/mcp_connect.rs) run the
binary: missing/empty/flag-like address, headless control, loopback
endpoint routing, closed port.
- docs/mcp.md says what happens without an address.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
* Move tests away from other PRs' hunks to avoid merge conflicts
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
TAN and PER snaps need the other end of the line, so the first point of a
LINE (no base point yet) could never snap tangent: a belt line tangent to
two circles was impossible to draw.
Now, as in AutoCAD, a point prompt can accept *deferred* snaps:
- snap: with no base and the new `deferred` flag, osnap offers "Deferred
Tangent" (circles, arcs) and "Deferred Perpendicular" (circles, arcs,
lines, xlines, rays) where no other snap wins. The hit carries the curve
(`SnapHit::deferred`, `Deferred`, `SnapCurve`).
- `resolve_deferred` solves the line once the other end is known: tangent
or perpendicular from a point, tangent-to-tangent (common tangents of two
circles, new `Circle::common_tangents` in geom) and the perpendicular
mixes. Among several solutions the one nearest the picks wins. No
solution (point inside the circle, concentric or nested circles,
non-finite input) is `None`, never a panic.
- prompt: `Prompt::deferred()` opts a prompt in; `Input::Deferred` reaches
only such prompts, everywhere else the session turns it into the point it
was picked at, so other commands snap as before.
- LINE opts in for its first point (and the next one while the first is
still deferred). No solution reports and re-prompts; Undo drops the
deferred point. The rubber band shows the resolved line, also the
tangent-to-tangent one while hovering the second curve
(`Session::cursor_deferred`, set by the canvas).
- canvas: passes the prompt's flag to osnap and sends a deferred pick as
`Input::Deferred`.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
A pick after one edit at 200k entities drops from ~110 ms (full rebuild) to ~4 ms;
steady-state picks stay at ~1 us. Adds a parity test against fresh builds and a
perf row for the incremental path.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
MarekSpodBiedry