Windows shows a window's big icon (ICON_BIG) in the taskbar and Alt+Tab.
eframe sets only the window icon (the small one, in the title bar), so the
taskbar showed a generic window instead of VectorCraft's icon. The app now sets
the big icon from the same 256 px image when the window is created.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
On every quit (Cmd+Q, VectorCraft > Quit, the close button) on a MacBook Pro
with a Touch Bar, AppKit threw from -[_NSTouchBarFinderObservation invalidate]
and macOS reported a crash. AccessKit's macOS adapter subclasses the window's
content view and gives it back its own class when the window is torn down;
AppKit's Touch Bar support observes the view under the subclass, so removing
that observation throws (the same cause Zed fixed in zed-industries/zed#65186).
On macOS with accessibility, the app now ends the process at the end of
on_exit, once the preferences are saved, before eframe tears the window down.
Nothing after the window is needed on a normal quit. While the window is still
starting up it doesn't, so a graphics failure still starts the app again on
another adapter (#651).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Metal reports no PCI ids, so both GPUs of a dual-GPU Mac had the adapter key
Metal:0000:0000. When the integrated GPU failed at start-up, the restart left
out both and the app quit, although the discrete GPU works. An adapter without
ids is now keyed by its name, so the restart leaves out only the one that
failed.
The failure itself was wgpu's indirect-call validation compute shader, which
OCLP-patched Metal on an Iris Pro can't compile. Nothing in VectorCraft or
egui draws indirectly, so that validation is off (WGPU_VALIDATION_INDIRECT_CALL=1
turns it back on).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The DirectWrite font lister (#594) was installed in the desktop app only:
`vectorcraft-cli` (convert, run, MCP) didn't see fonts that font services
such as Adobe Fonts load on Windows, so its exports set such type in a
fallback font.
- The CLI shares the app's `system_fonts.rs` and installs it first thing
on Windows (not the win7 build). The text crate stays free of unsafe
code; the CLI denies unsafe code instead of forbidding it, so the
shared module's own allowance applies.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A browser's Copy Image puts the picture on the clipboard with the image's
address as text (and HTML). Paste takes text before bitmaps, as word
processors offer a picture of copied text, so it pasted the address.
- `clipboard::is_address`: text that is only an address (http, https,
file, ftp, data or blob).
- Desktop paste (Windows, macOS, Linux): such text gives way to the
clipboard's bitmap when it has one. An address alone pastes as text, as
before; nothing is fetched.
- Web: a paste whose text is only an address takes its picture file.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
VectorCraft asks for a path where it needs one and goes on with it at
once, so its file dialogs ran on the UI thread. Windows and macOS dialogs
run the window's events meanwhile; on Linux nothing answered the Wayland
compositor's pings while a dialog was open, and Hyprland offered to kill
the window as not responding.
- ui-egui `picks`: every file dialog (open, save, folder, place) goes
through one place. With the host's `start_pick` (desktop Linux) the
dialog shows on a thread of its own; what asked (the command, the
dialog's OK, a panel's or dialog's button) gives up for now and runs
again with the picked path when the dialog answers. Another dialog
waits for the open one.
- The desktop app passes its window as the dialogs' parent on Linux (the
portal can then associate its window, "Unhandled parent window type"
before) and Windows; macOS shows them as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Font services load their fonts with GDI at run time, from files in their
own folders, without installing them: neither the font folders nor the
registry list them, so the font scan never saw them. DirectWrite's system
font collection holds them since Windows 10 1607.
- vectorcraft-text: `set_platform_font_files` lets the app name font files
the platform knows; each scan and each check for installed fonts asks
again and reads those outside the font folders. A file named by itself
is read whatever its extension (font services keep fonts in files
without one).
- The desktop app lists the local files of DirectWrite's system font
collection on Windows (`system_fonts.rs`, through the windows crate
already in the build), read in place.
- The check for fonts installed while the app was in the background runs
on a thread: DirectWrite's check for updates takes about 50 ms.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
An SVG used as an animation rig lost its hidden alternates and their
rotation pivots: the exporter left hidden objects out (only hidden layers
could be kept), and the importer dropped data-* attributes.
- SVG export: hiddenLayers (on by default for document.save, as before)
now keeps hidden objects too, each written display="none" on its own
element (path, image, text or group). Exports still leave them out
unless asked; the CLI's convert asks, as a conversion keeps the
document.
- An object's own data: ObjectAttributes.data, the data-* attributes in
order (ours, data-name and data-vc-*, aside). The importer keeps them on
the objects and layers they belong to (at most 64 per element), the
exporter writes them back, the native format saves them (missing in
older files), and object.setProps {data: {key: value|null}} sets or
removes them.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Windows lists each GPU under both backends, and Vulkan came first. On an
Intel GPU its driver made the whole window flicker black while the
pointer moved; the same machine was fine with WGPU_BACKEND=dx12 or gl.
adapter_order now ranks a GPU's DX12 adapter before its Vulkan one, after
the power preference, so the GPU chosen doesn't change, only how it is
driven. Vulkan stays the fallback (a failed DX12 start restarts on it)
and WGPU_BACKEND=vulkan still picks it. Elsewhere there is no DX12
adapter, so nothing changes.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
On macOS the menus are in the system menu bar, laid out as a Mac app's (Illustrator on the Mac as
the reference), and the app settings sit in the VectorCraft menu: About, Settings ▸ (one item per
Preferences page, General… ⌘K), Language ▸, Appearance ▸ (UI brightness), Services, Hide
VectorCraft ⌃⌘H (⌘H stays View › Hide Edges), Hide Others ⌥⌘H, Show All and Quit ⌘Q (the app's own
quit, so unsaved documents are asked about); they leave Edit and Help. Window gets Minimize ⌃⌘M,
Zoom and Bring All to Front; Help is the system's Help menu. Windows, Linux and the web keep the
in-window menu bar unchanged; --in-window-menus or VECTORCRAFT_IN_WINDOW_MENUS keeps it on macOS
too, and so does a menu bar AppKit refuses to build.
One menu model: the in-window menus and the native bar draw the same rows (menus::entry), and the
platform-free native_menu module lays them out, decides when to update (a cheap state hash; labels,
enabled and checked in place, a rebuild only when the structure changes) and hands key equivalents
back to egui as the key presses they were, so shortcuts, user overrides and focused text fields
behave as in the window and nothing fires twice. The muda backend (pinned =0.21.1, macOS only, no
unsafe) is apps/vectorcraft/src/mac_menu.rs, mirroring PhotoCraft's. ui.inspect reports
nativeMenuBar. Translations for the new labels in ja, es, zh-hans, zh-hant, cs and pt-br.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- The worker reads the clipboard only while nothing is copied in the app
and a document is open (as the in-line check did before), and idles on
a channel otherwise; it sends a verdict and wakes the UI only when the
answer changes.
- The in-line check (Windows, macOS, web) keeps its short-circuit: it
doesn't open the system clipboard after an in-app copy or without a
document, as before the PR.
- A dead worker hands the check back to the UI thread instead of
freezing the last verdict.
- Dropping the app wakes the worker and waits (250 ms at most) until it
has let go of its clipboard handle: on X11 the last arboard handle
alive hands what the app copied to the clipboard manager, and a
detached worker holding it would lose that handover at exit.
- The factory stays in `Services` until the first frame takes it (no
extra app fields); main.rs installs it with `cfg!(target_os = "linux")`.
- Tests trimmed from five to three that share tests_sysclip's helpers:
the frame never waits on a slow clipboard and only the probe thread
reads it (nothing read without a document), the probe rests after an
in-app copy and lets go of its handle when the app drops, and a dead
probe falls back to the in-line check. The in-line case is already
covered by tests_clipboard.
- ROADMAP (Last updated, Platforms, Shipped so far) and
docs/development.md (Linux: clipboard owners that never answer).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`has()` on an X11 clipboard owner that never answers can block for
seconds; the app called it every 0.25 s on the UI thread to decide
whether Paste is enabled, so a stuck owner froze the frame loop (the
window stopped drawing and could not be closed).
`clipboard_probe` owns a second clipboard handle on a worker thread,
publishes its verdict over a channel and wakes the UI only when it
changes; the frame reads the latest verdict without ever blocking.
Without a factory the check stays in line, throttled, exactly as before
(the web shows SVG text only; hosts that don't install one keep the old
path).
The factory is installed on Linux only for now (the environment the bug
reproduces in); other platforms keep the in-line check, so this cannot
change their behaviour.
Tests: a fake clipboard whose `has()` blocks on a gate, a probe factory
sharing a recorder, and a headless frame whose elapsed time is measured.
They assert the frame does not wait on the clipboard and that the check
runs on the probe thread, that the probe's verdict reaches `system_paste`
and the Paste menu, that dropping the app stops the probe, and that the
in-line check (no factory: the web, older hosts) still works.
Reported: with the Pen there was no way to edit curves in place as Ctrl does in Illustrator on
Windows (Cmd on the Mac), segments couldn't be grabbed, and the curve-editing tools felt slow and
sometimes didn't take a handle.
- Cmd/Ctrl held at a press with any tool but a selection tool now borrows the selection tool in the
engine (Session::pointer), not in the canvas: the selection tool chosen last, or Direct Selection
for the Pen, Curvature and anchor tools until one is chosen. The tool lent to isn't deactivated,
so the Pen goes on drawing its path after the release while that path stays selected; it no
longer draws on another open path selected meanwhile. Agents reach it with pointer_gesture
mods.cmd. With Cmd held the pointer is the borrowed tool's. Choosing a tool or switching
documents mid-gesture gives the tool back first.
- Pen + Alt over a segment of a selected path reshapes it, as the Anchor Point tool does.
- The handles shown are the handles that drag (one helper for Direct Selection, the Anchor Point
tool and Pen + Alt): Direct Selection now drags the handles of a path selected as a whole (as
the Pen leaves it, so Cmd + drag of a handle while drawing works), the Anchor Point tool no
longer grabs handles that Show handles when multiple anchors are selected hides, the nearest
handle wins, and a handle shorter than the tolerance leaves its anchor to be picked. The anchor
tools pick within Selection & Anchor Display > Tolerance instead of a fixed 4 px.
- path.reshapeSegment keeps a smooth anchor at either end smooth (its other handle turns with the
moved one) instead of kinking it.
- A dragged handle gathered its smart guide targets from the whole document on every pointer move
(21 ms per move on 20,000 paths): HandleSnap now gathers them once per drag (1 ms).
vectorcraft-cli perf measures handle and segment drags on a 5,000-anchor path.
- catalog::is_selection_tool and catalog::edits_anchors replace the UI's own lists.
- Tests in tools, engine and MCP (headless pointer_gesture); docs/mcp.md, control-protocol.md,
ROADMAP.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
0.5.0 asked wgpu for the low-power adapter everywhere (#306). On a Linux
desktop whose KDE Wayland compositor runs on an NVIDIA GPU, wgpu sorted the
Ryzen's integrated RADV GPU first: it reports that it can present to the
window, but KWin can't import the buffers it allocates ("importing the
supplied dmabufs failed"), which is a fatal Wayland protocol error. The next
surface configuration then got ERROR_SURFACE_LOST_KHR and egui-wgpu panicked
("The surface isn't supported by this adapter", winit.rs:129).
- Preferences > Performance > Graphics Processor gains Automatic, the new
default: power saving on Windows and macOS as #306 wanted, and on Linux the
system's order, where Mesa's device-select layer puts the GPU the desktop
runs on first. Values are now automatic / lowPower / highPerformance; 0.5.0
saved powerSaving for everyone, which reads as automatic.
- The app picks the window's adapter itself (apps/vectorcraft/src/gpu.rs,
eframe's native_adapter_selector): only adapters that present to the
surface, WGPU_ADAPTER_NAME first, hardware before software, native backends
before OpenGL, then the power preference. Every adapter and the pick are
logged.
- run_native runs under the engine guard. When the graphics fail during the
first frames (a wgpu error, or a panic in wgpu, egui-wgpu or naga), the app
starts again without that adapter (exec on Unix, so an AppImage stays
mounted; a new process on Windows), passing the failed adapters in
VECTORCRAFT_GPU_SKIP, and says so in the status bar. When none is left it
logs why and exits with an error, never a panic.
- docs/development.md, README and ROADMAP: the new default, the restart and
the escape hatches (WGPU_POWER_PREF, WGPU_ADAPTER_NAME, WGPU_BACKEND,
MESA_VK_DEVICE_SELECT, DRI_PRIME).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`application:openURLs:` can pass URLs that aren't files, and on macOS 27 `to_file_path()` returns
the path of an https URL ("/a.svg"), so it was queued as a document. Keep the file URLs
(`isFileURL`) before converting; `opened_files_wait_for_the_ui` passes again on macOS 27.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Discord feedback: "EPS files are always rasterized: they open as bitmaps, in
other programs as vectors". Every EPS saved by Illustrator 2026 fell back to
its TIFF preview (`gcheck` was unknown), and cairo/Inkscape files stopped at
their first image (`status`), dropped masked images (ImageType 3) and failed
on mesh gradients; the warning didn't say where the program failed.
Fallback never silent:
- The warning names the PostScript error, the operator that raised it and
the named procedures it ran in, innermost first ("a PostScript error
(typecheck in `add`), in `inner` in `outer`"); errors caught by `stopped`
aren't reported. It reaches the status bar, document.open, app.open and
file.open (which now return document.open's result, so MCP's open_file over
the control channel gets it), and `vectorcraft-cli info` ("warnings").
- A file whose PostScript can't be read opens as its TIFF preview, else its
Windows metafile preview (as vectors, scaled to the bounding box), else its
EPSI bitmap. es @msg rows for the new messages.
Interpreter:
- Operators programs probe or use: gcheck/scheck, rootfont,
currentcolorrendering, currenthsbcolor, status, token, read,
bytesavailable, fileposition/setfileposition, resetfile, write*, pdfmark
(marks dropped, as Ghostscript does), execform, nulldevice, strokepath,
pathforall, user paths (ufill/ueofill/ustroke/uappend/upath/setbbox, the
encoded form too), setcachedevice2, findcolorrendering, cache, device and
VM queries, revision; (%stdout) (w) file gives a sink.
- `//name` is replaced by its value when read, inside procedures too.
- Tiling patterns become pattern swatches (their PaintProc drawn in pattern
space, one cell; uncoloured ones in the colour under them; makepattern
copies share one swatch); nested patterns stop at 8 levels.
- Type 3 fonts draw their glyphs (BuildGlyph/BuildChar, widths from
setcachedevice/setcharwidth) as outlines grouped per string, spaced by
ashow/widthshow/awidthshow/xshow/yshow/xyshow/kshow; glyphshow and
stringwidth/cshow use them.
- Shadings of types 1 and 4-7 become gradient meshes (data from arrays,
strings or files, packed or not, flags sharing edges), as fills through a
shading pattern too; sampled functions (type 0) are evaluated;
GradientMesh::coons/triangle are shared with PDF import.
- ImageType 3 masked images, interleaved by sample, by row or in separate
sources; ReusableStreamDecode reads its data when made; Flate/LZW PNG and
TIFF predictors.
- Bounded: DeviceN at 32 inks, sampled functions at 2^24 samples and 32
outputs, 4096 patches a shading, strokepath ignores dashes finer than 1e5
pieces, Type 3 glyphs nest 8 deep (a glyph that shows itself overflowed the
stack before the limit), 64-bit index maths for 32-bit targets.
Tests: a synthetic program per generator (cairo/Inkscape, its Type 3 fonts,
matplotlib, CorelDRAW, Affinity, LibreOffice, Ghostscript eps2write,
Illustrator procsets), meshes and function shadings, masked images, fault
messages, probing operators, metafile/EPSI previews, `//name`, user paths,
hostile patterns/glyphs/shadings (run against main, 12 of the 14 generator
and feature tests fail; CorelDRAW- and LibreOffice-style programs already
read there). The import fuzz vocabulary covers the new operators. Checked
against real files made here with Inkscape 1.4 (cairo) and Illustrator 2026
(its own format and the CS-to-CC compatibility formats; all fell back or
stopped early on main and open as vectors now; its Illustrator 8 format
needs the next commit); those files aren't committed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Presentation Mode hides the menus and panels, had no Escape and was saved with the
preferences, so it came back on every restart with no way out. Escape now leaves it (as in
Illustrator), F from it goes back to Normal Screen Mode, and the screen mode is no longer
saved: the app always starts in Normal Screen Mode, and preferences saved in Presentation
Mode by older versions open normally. ui.inspect reports `screenMode` on its own.
Desktop file drags carry no pointer position or modifiers (winit 0.30 on Windows, macOS
and X11), so every drop on a window with a document open was taken for the middle of the
canvas and placed. Where the platform doesn't say, a dropped document (native, SVG, PDF,
.ai, EPS, DXF, EMF/WMF) now opens as a tab of its own and a picture or text is placed in
the middle of the view; pictures dropped together with documents are placed first, so the
last document opened is shown. The web, whose drag events give the position, keeps
Illustrator's choice: placed on the canvas, opened off it (the tab bar).
Illustrator's PDF-compatible part leaves out art outside the artboards (it lives in the
private data alone, which isn't read), so it can't open; importing a file with that
private data now says so in a note. Off-page art a PDF does draw already opens on the
pasteboard; a test keeps it so.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The "no log file" message went through eprintln!, which panics when
standard error is a broken pipe; log::warn! reaches standard error the
same way (the file sink has already given up) and never panics.
ROADMAP: the log file in "Shipped so far" and "Last updated".
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The desktop app installed no logger, so every log::warn!/log::error! in
the workspace was discarded: the graphics adapter in use, a lost graphics
device, the engine guard's report of a recovered panic. A launch from a
desktop menu has no terminal either, so there was nothing to attach to a
bug report.
apps/vectorcraft/src/logging.rs (no new dependency):
- standard error plus logs/vectorcraft.log next to the preferences
(~/.config/vectorcraft/logs, or under $XDG_CONFIG_HOME); each launch
rotates the previous log to vectorcraft.1.log and .2.log, so the log of
a run that crashed survives the next start; the file is capped at 16 MiB
- records logged before the settings directory is known are buffered
(bounded) and written once the file is attached
- default levels: info for vectorcraft*, warn for everything else;
RUST_LOG replaces them with env_logger-style directives
- never panics: an unusable directory or a failed write leaves standard
error as the only sink
- runs with VECTORCRAFT_NO_PREFS (agents' test runs) log to standard
error only, so they don't rotate away the user's own logs
The logger is installed first in main and the file is opened after
argument parsing (--version leaves no file). Log messages of this crate
drop their "vectorcraft: " prefix, which the record's target now shows.
vectorcraft-cli keeps its own MCP logger (vectorcraft_mcp::logging).
Documented in docs/development.md (Logs, Environment variables).
A screenshot on the clipboard didn't paste everywhere:
- macOS: the Paste menu item (which takes Cmd+V there) was enabled for text
only, so with nothing copied in the app Cmd+V did nothing; and arboard
reads the pasteboard's TIFF only. The pasteboard's types now enable Paste,
and its PNG, else TIFF, pastes as it is.
- Windows: bitmaps came through CF_BITMAP and GetDIBits, which drops alpha.
Paste now reads PNG, else CF_DIBV5, else CF_DIB, decoded with their colour
masks and tables (a BMP file header in front, which image's headerless
decoder gets wrong for masks after a BITMAPINFOHEADER), opaque when the
alpha is zero throughout as screenshots leave it, capped at 32768 px a side.
- Web: only SVG text pasted. The page's paste event now hands the first
picture or file that is art to the app (VectorcraftApp::paste_from_host),
pasted with the chord held, like the desktop's paste keys.
- Linux reads image/png already; its Paste menu item still waits for text
(no cheap way to ask arboard for a bitmap), the paste keys take one.
Bitmaps paste as embedded images at the view centre, as before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Over the canvas the window hid the system cursor and painted the tool's
cursor glyph into its own frame. That glyph reached the screen only after
the frame was rendered, queued (two frames by default) and composited, so
on Windows it visibly trailed the real pointer, on any GPU: the "sluggish
mouse" of the report. Frame times were not the cause (0.4 ms empty,
1.5 ms median on a 20 000-path document, and no repaints while idle).
- cursors: the glyphs are collected as shapes, tessellated by egui and
rasterized into straight-alpha RGBA bitmaps (anti-aliased, hotspot on a
pixel corner), cached per cursor and display scale so the same Arc is
handed over every frame and egui-winit keeps its OS cursor.
- canvas: on Windows and Linux the tool cursor goes to the OS through
ctx.set_cursor_image; the system moves it itself. macOS (bitmaps sized
in points) and the web (no bitmap cursors in eframe) keep painting it.
- app: the wgpu surface queues one frame instead of two (LOW_LATENCY):
the GPU only composites the CPU-rendered canvas, so everything drawn in
the window follows the pointer a frame sooner.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
document.inspect takes the same {depth?, childLimit?} slicing as
document.node (a depth-0 read is the layer skeleton), and document.find
{name?|kind?|text?, limit?} locates nodes across the tree with their
ancestor paths and a total. docs/mcp.md documents the drill-down recipe;
the CLI stdio sweep covers both.
Under Wayland the window is native but arboard was built without its
data-control backend, so it read the X11 clipboard through XWayland,
which only holds what X11 apps copied: Paste found nothing and reported
an empty clipboard (egui's text paste into fields uses smithay-clipboard,
which is why that worked). arboard now has `wayland-data-control`, the
same as PhotoCraft; it falls back to X11 where the compositor lacks the
protocol (GNOME's Mutter).
The reporter's clipboard also held text/uri-list: with a file copied in a
file manager, Paste would have taken the path as text before the bitmap.
Copied files now paste first, as the first one that is art (SVG, PDF,
EMF/WMF or a bitmap), told from its first 4 KiB by the engine's format
detection (`clipboard::file_flavour`): text/uri-list through arboard on
Linux and macOS (a trailing CR from CRLF lists trimmed), CF_HDROP on
Windows, read with the clipboard closed again.
Dropping files on the window can't work under Wayland yet: winit 0.30,
which eframe 0.36 runs on, has no Wayland drag and drop (winit#1881,
added in 0.31), so no dropped_files arrive. docs/development.md documents
it and the workarounds (copy and paste the files, File > Place, or run
under XWayland with `WAYLAND_DISPLAY=`).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`depth` / `childLimit` without `summary: true` used to return the whole
full-JSON subtree silently; it is now an invalid-params error. docs/mcp.md
describes the slice parameters and childCount, the engine test covers
fractional values, unsliced reads and limits past the tree, and the CLI
stdio sweep reads a sliced group through run_command.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Select All (Cmd/Ctrl+A) now follows the keyboard, as in Illustrator:
- macOS menu bar: the system menu takes Cmd+A/X/C/V before the window
sees them, so a focused panel, dialog or search field got the document
command. `menus::invoke_from_system_menu` (used by `NativeMenu::poll`)
hands Select All, Cut, Copy and Paste to a focused text field as the
events its keys would send; clicks on those menu items follow the same
rule. Other commands, and these with no field focused, run as before.
- Type tool: `select.all` itself selects all of the text being edited
while the Type tool edits (returns {editing, start, end}), so the menu,
the shortcut, the control channel and headless MCP (`press_key`) agree.
The UI's own Select All interception and the Type tool's hard-wired
Cmd+A are gone; Cmd+A reaches it through the Select All shortcut.
- Windows, Linux and the web already left Cmd+A to a focused field; now
covered by headless app-frame tests (field, Type tool, canvas).
Docs: docs/mcp.md (Select All while editing type), ROADMAP. Not yet:
Select All across threaded frames (one frame's text is selected).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
When the browser dropped the page's WebGPU device or WebGL context, the
editor kept running but nothing it drew reached the screen: Chrome and
WebKit showed a frozen, stale canvas and Firefox a blank page, while
edits and Save went on.
- ui-egui: new `graphics` module. `GraphicsLoss` carries a loss report
from any thread (wgpu's device-lost callback, the canvas's
`webglcontextlost` event) to the UI thread. `VectorcraftApp::
graphics_lost` writes Data Recovery copies of the modified documents at
once. The app now notices when the host hands it a new egui context and
uploads everything again (fonts and theme, the canvas, place cursor
thumbnails); every per-thread texture cache (layer, symbol, swatch,
style, asset and glyph thumbnails, previews, font samples, Navigator,
Flattener Preview) is a `TexCache` that empties for a new context.
Without this, the moved app drew stale or wrong textures and egui
panicked on its missing font families.
- web: the shell watches both loss signals; on a loss it puts a fresh
canvas in place of the dead one and starts the same `WebRunner` again,
handing over the same app (documents, undo, panels), with services made
for the new context and a status-bar note. If no new graphics can be had
(or after three losses) it removes the stale canvas, lets go of the
recovery copies and says to reload.
- engine: `recovery::leave` lets go of a session's area but keeps its
copies, so the reloaded page offers them at once instead of after the
heartbeat goes stale.
- desktop: eframe can't give a window a new device, so a lost device
(driver reset) is logged, the copies are written and the status bar says
to save and restart; closing then skips the Save Changes question it
can't show.
- Tests: loss reporting across threads, `TexCache` emptying, the app
uploading the canvas, a cached preview, theme and fonts again in a new
context (fails without the fix), copies written on loss, and
`recovery::leave` offering copies at once with both lock and heartbeat
stores. docs/development.md (web section) and ROADMAP updated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A heartbeat alone can't tell a crashed tab from one whose timers stopped in
the background, so another tab could discard (or restore) the copies of a
still-live document once its heartbeat went stale.
- Recovery stores without locks can now announce an area for as long as the
app runs (RecoveryStore::announce/announced). The web shell implements it
with the browser's Web Locks (navigator.locks through js_sys::Reflect): a
tab holds a lock named after its area, which the browser releases only when
the tab is gone. Lock snapshots are refreshed every 10 s (once before the
first frame); a snapshot over a minute old counts every area as held.
- An area is gone only when no announcement holds it and its heartbeat is
stale. Where the store can't tell (no Web Locks), the owner repairs: each
heartbeat (also its first frame after a pause) and each recovery save write
again the copies missing from its area, and a heartbeat newer than the one
another tab wrote when taking the area over marks it running again.
- MemoryStore::announcing models Web Locks in tests; engine and UI tests cover
the issue's sequence. Native folder locks are unchanged.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Finder, Open With, a drop on the Dock icon and `open -a` failed with "VectorCraft cannot open
files in the PDF document format":
- The release Info.plist declared no document types, so Launch Services refused the files before
the app saw them. Info.plist.in now declares every format File › Open reads (the engine's
OPEN_EXTS; a test keeps them in step): VectorCraft owns its own documents and templates and is an
alternate for the others, so it takes over no other app's files. `cargo xtask bundle` now fills
in the same template instead of its own, shorter plist.
- macOS delivers such files as open-document events, not arguments. winit 0.30's application
delegate has no `application:openURLs:`, so AppKit fell back to NSDocumentController, which
refuses every file of a non-NSDocument app (the error in both issues). The new
`open_documents` module adds that method to the delegate's class when the app is about to finish
launching (before AppKit delivers launch files), queues the paths and wakes the UI, which opens
them with `io::open_path` like command-line files; errors go to the status bar too.
New direct macOS dependencies `objc2-foundation` 0.3 and `block2` 0.6 were already in Cargo.lock
(through winit, muda and rfd). The module allows `unsafe_code` for the Objective-C runtime call.
Fixes#354
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The web reads dropped files asynchronously, and the place inbox only kept
the drop point, so a file dropped on document A was placed into whatever
document was active when its bytes arrived. A drop on a canvas now carries
place::DropAt (the document's uid, the point, embed), and drop_files makes
that document active again before placing; when it has closed meanwhile,
nothing is placed and the status bar says so. The native drop path is
synchronous and goes through the same code, so it is unchanged in effect.
Test: a drop pending while another document opens lands in the first one
at the drop point; one whose document closes is not placed anywhere.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
On hybrid-graphics laptops (an integrated GPU driving the display plus a
discrete one), egui-wgpu's default power preference is HighPerformance, so
VectorCraft rendered on the discrete GPU and its frames were copied to the
display through the integrated one. On an Asus ZenBook with Intel Iris + Arc
the window flickered whenever the mouse moved over it (#306, #305).
The canvas is rasterized on the CPU (vello_cpu) and the GPU only composites
the canvas texture and draws the UI, so the integrated GPU is plenty. The
desktop app now asks wgpu for the power-saving adapter by default.
Single-GPU machines are unaffected.
- Preferences > Performance > Graphics Processor (`gpuPreference`:
powerSaving, the default, or highPerformance), with a note that it
applies after a restart. The app reads it from the saved preferences
before the window opens. WGPU_POWER_PREF still overrides it.
- Help > About and `ui.inspect` (`graphicsAdapter`) name the adapter in
use, and it is logged at startup, for future GPU bug reports.
- Docs (README, docs/development.md) and ROADMAP updated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The export tool's description, the docs' open_file and export rows and the
CLI's run/convert usage listed only some formats. The CLI usage now prints
the writable extensions next to the readable ones, both generated from the
format table (export_extensions), the export tool points to its generated
format enum, and the docs name EPS, DXF, EMF, WMF, SVGZ and templates.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A line that is not a JSON object with a string `method` (an HTTP request
line, other non-JSON text, a JSON array, invalid UTF-8, a line over 4 MiB)
now gets one error reply and the connection is closed, so a browser's
cross-origin fetch can't smuggle a command in its body. Connections are
capped at 16. Same transport as LightCraft (storytold/lightcraft#129) and
FilmCraft (storytold/filmcraft#123).
Follow-ups from the review of #205.
- Names that are user, file or system data are shown as they are: menu
items for recent files, saved views, recent fonts and Type > Font,
user swatch and graphic style libraries, saved perspective presets,
custom workspaces (and Reset <name>) and plug-ins no longer go through
the catalog (menus::shows_a_name). widgets::dropdown_names,
menu_item_name and dim_name are the non-translating forms of the shared
widgets, for the panels and dialogs that list such names.
- The command palette builds its entries (shown label, search text) once
per language and plug-in set instead of translating ~700 labels every
frame while a query is typed.
- The locale probe (reg.exe / defaults) runs on a background thread at
startup instead of on the first frame.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Strings in code stay English and are the lookup keys; `crates/ui-egui/src/i18n`
maps them to display text at render time from one `*.tsv` catalog per language
(`LANGUAGES` registry, `tl!`/`t`/`tr`/`tr_ctx`/`tr_id`/`tn`/`fmt`). Command ids,
menu paths, `ui.menu.list`, the control channel, the CLI, MCP, status messages
and undo labels never see translated text, so agents, scripts and the existing
tests are unaffected (under `cargo test` the UI is always English).
Wired: in-window menus and the macOS native menu (rebuilt on a language change),
dynamic labels (Undo X, Reset X, Last Effect: X as templates), the command
palette (matches English and translated text), dock tabs, panel tooltips, the
toolbar (category names now cut by characters, not bytes), the dialog frame, the
shared widgets, the Preferences dialog, the shortcut editor, the hint and status
bars, the Home screen, and the literals of every panel and dialog.
Preference `interfaceLanguage` (User Interface › Language): `auto` or a code.
`auto` follows `VECTORCRAFT_LOCALE`, `LC_ALL`/`LC_MESSAGES`/`LANG`/`LANGUAGE`,
the macOS preferred languages and, on Windows (which sets no `LANG`), the user
locale from `HKCU\Control Panel\International`. `zh-TW`/`zh-HK`/`zh-MO`/
`zh-Hant-*` resolve to `zh-hant`; Simplified locales stay English until a
`zh-hans` catalog exists.
`zh-hant.tsv` (~2,850 rows) is clean-room, in the vocabulary used in Taiwan;
terms shared with PhotoCraft reuse its wording, vector terms use everyday
vector-drawing vocabulary. Tests: catalog well-formedness (placeholders,
ellipses, command ids, no Simplified characters), locale resolution, and two
coverage tests that fail when a menu string, panel/tool/preference label,
dialog button, hint-bar fragment or `tl!` literal has no row in a language
marked `complete_menus`. `VECTORCRAFT_I18N_DUMP=file cargo test -p
vectorcraft-ui-egui dump_source_strings` lists the strings for translators.
Also: two clippy findings under rustc 1.97 fixed in passing.
Integrated with main (squashed onto it, 42 commits on): the Czech interface
(#213) moves from its Rust table to `i18n/cs.tsv` on the registry (all 556
rows, Czech plural rule, its menu-coverage and glyph tests kept); Find Font's
resolution notes and the canvas context menu's labels are translated and
covered by the catalog tests (`menus::CONTEXT_LABELS`); the glyph test checks
CJK text only when the craft-fonts build input is embedded (#220 moved the
Japanese fonts out of this repository); a language saved by an older version
carries over to `interfaceLanguage`; Windows' locale lookup runs reg.exe by its
full path.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The server answered tools/* and two fixed resources and nothing else, and
resources/templates/list returned an empty array. This adds the rest of the surface a
client can actually use, against 2025-06-18.
- prompts/list and prompts/get: five workflows (poster, icon-set, recolor,
trace-and-style, export-set) with templated arguments. A missing required argument
is -32602 and names itself.
- resources/templates/list, and resources/read for the templates: object/{id},
command/{id}, effect/{id}, swatch/{name}. Reading one thing at a time instead of
vectorcraft://document/json, which on a real document is thousands of lines an agent
pays for again on every change.
- completion/complete for prompt arguments and template variables, ranked prefix
first, capped at 100 with total and hasMore. The values are read live, so a suggested
format, effect id, command id, trace preset, swatch or object id is one the engine
has.
- logging/setLevel and notifications/message. Nothing installed a log::Log before, so
the engine's log::debug! calls went nowhere; this is the sink, silent until asked.
Installing it is vectorcraft-cli's job rather than the library's, so an embedder
that already has a logger keeps it.
docs/mcp.md gains a Protocol section with the capability table and a Not implemented
section: subscriptions, progress, cancellation, sampling, elicitation and Streamable
HTTP all need the transport to interleave reads, which this one does not — handle_line
answers a line with a reply, so it neither reads an incoming cancellation nor can write
progress before the reply it belongs to.
62 tests (50 unit, 12 property/fuzz). protocol_props.rs gained five: prompts survive
junk arguments, completion/complete never panics and always answers at most 100
string values with a consistent hasMore, every template reads or fails cleanly, and
prompts/completions/templates are swept with junk params while asserting the document
comes back byte-identical.
ROADMAP: one bullet under Automation in Shipped so far and the changelog line. The
parity table and the gap list are untouched — MCP is agent automation, so feature
parity does not move.
Typing with an input method (the macOS Japanese IME and the like) used to
insert the raw keys (romaji came in as ASCII): egui's Ime events weren't
handled.
- Tools gain `ime_preedit` / `composing` / `ime_caret`; the Type tool lays
the marked text out in place inside its typing session, underlined (the
clause being converted thicker), and commits it through `text_input`.
Character ranges from the IME convert to byte ranges in one place. A
cancelled composition leaves no empty undo step; an empty clear with
nothing marked never deletes the selection.
- The UI sends Text and Ime events to the tool in order, sends IMEOutput
every frame while text is edited (what lets egui-winit enable the IME
and place its candidate window at the caret, or the converting clause),
and holds back keys, Enter/Escape, clipboard, single-key shortcuts and
menu commands (Undo, Cut, Select All… also from the native menu) while
composing. A click, a tool switch or a style change ends the
composition with the marked text kept as typed; on macOS the IME is
told to discard it (NSTextInputContext.discardMarkedText) so it doesn't
come back in the next input.
- Undo during a typing session also ends the Type tool's session, so its
caret and marked text follow the document.
- Tests: tools (char → byte ranges, underlines and candidate position),
engine (compose → commit is one step, cancel leaves no step, a click
commits, Undo mid-composition is safe), ui-egui (synthetic Ime events:
compose, commit, ⌘Z held back, IME output, click and app switch).
Checked by hand on macOS 27 with the Japanese IME (romaji input, convert,
commit, cancel, Undo/Redo after commit, click away, Latin mode, app
switch). Not checked: Windows and Linux IMEs, the web build.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The MSI Icon id had no .ico/.exe extension, which Windows Installer needs for
advertised shortcuts and ARPPRODUCTICON, so both showed a blank icon. Release
builds of the desktop app are now GUI-subsystem (no console window; the CLI stays
a console program), the shortcut/summary description names what VectorCraft does, and
packaging-lint rejects an Icon id without an extension. Found in printcraft#57.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
`vectorcraft-cli commands | head -1` panicked and exited with status 101:
thread 'main' panicked at library/std/src/io/stdio.rs:1168:9:
failed printing to stdout: Broken pipe (os error 32)
println! and print! panic when stdout is closed. Output now goes through
outln! and out!, which end the program with status 0 on a broken pipe (the
reader has what it wanted) and report any other write error on stderr with
status 1.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The first window opens at 1440 x 900 when the screen has room for it around a
taskbar or Dock, and maximized (with a size that fits once un-maximized)
when it doesn't, as on 1600 x 900, 1366 x 768 or 150% scaled 1080p screens.
The window's size, position and maximized state are saved with the UI
preferences (UiState::window) and restored at the next launch, moved fully
onto a monitor that is still connected and maximized if it no longer fits.
Root cause: main.rs created the window at a fixed 1440 x 900 at the system's
default position, never checked it against the monitor and never remembered
it. The new apps/vectorcraft/src/window.rs places the still-hidden window
from the monitors winit reports (CreationContext::winit_window) and tracks
its geometry every frame.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
bflatastic
David Whiting
Oscar Sebastián Mellado Bustos
imachus
Gert Pellin
Eliezer Hernandez
Fahri Meral
njmueller
Luskish
WildfootW
Brandon Thomas
midasdf
japbcoelho