Commit Graph

52 Commits (706ff6fb264bd5f7e8663b76c827fae85f2a2ea9)

Author SHA1 Message Date
mula.liu 706ff6fb26 fix(icons): orange ground and white mark, as the product has always been
Reported: "你改的标注应用图标不对啊,我之前是橙底、白线的风格(参考 squid 的
图标)."

The previous commit inverted the figure/ground relationship. Chasing
consistency with the alpha-only tray mark, the app tile went to espresso
with a warm mark — technically consistent geometry, and it passed every
contrast check I ran, but those checks were pointed at the wrong axis.
Consistency has two of them here, geometry and light/dark, and the one a
returning user actually recognises is "orange tile".

Now the app icon is literally the tray mark with a colour turned on:
`#FF6B35` ground, which is the orange this project's own master already
carried, and `#FFF4EA` mark, the cream the app's panels are drawn on. The
plus stays a punched hole showing the ground rather than a third colour,
so both icons are one drawing on two different grounds. Two colours also
removes the failure mode the last version needed a rule to defend.

Tray PNG is byte-identical again; the .ico ladder (16/24/32/48/64/256) and
.icns entry set are unchanged, and every tracked PNG still matches its
committed dimensions.

Co-Authored-By: Codex <noreply@openai.com>
2026-11-19 21:49:05 +08:00
mula.liu 06de22b840 release: prepare Shift 0.1.13
Versions and release docs for the two installers.

Numbered 0.1.13 rather than 0.1.12 because 0.1.12 was staged by a commit
that never shipped: the channel still serves 0.1.11, and this batch has
grown well past what that commit described.

Six reported issues:

- 系统小图标改为黑白单色模板图,由系统着色。
- 应用图标与托盘共用一份几何,但在 Dock 与界面里保持彩色。
- Windows 一键安装运行组件必然失败,四个独立原因藏在同一处。
- 钉在屏幕上的图片新增可用的放大/还原按钮。
- 「最近使用」的收藏行带上常驻指示器。
- 新固件在主面板顶部提示,点一下跳到设置 → 软件更新。
- 参照 nex_squid 降低闲置开销:四处采样节奏加上最近列表只解码视线内的图片。

Verified on this host: 191 Rust and 51 frontend tests, `cargo clippy -D
warnings`, and a GUI pass against a signed feed advertising 9.9.9 — the
banner appears on both the probe launch and a later same-day relaunch with
the feed unreachable, so the notice really is restored from disk.

The Windows install fix still wants a real Windows host: the exact prose a
dismissed UAC prompt prints, and whether each package accepts
`--scope user`, are handled defensively rather than verified.

Co-Authored-By: Codex <noreply@openai.com>
2026-11-19 21:29:30 +08:00
mula.liu c8a5d41eca docs: document the release banner and the shared icon geometry
Two ledger sections, the README paragraphs for the banner, and the 0.1.13
version row.

「升级提示跨启动存活」records the constraint that is invisible in the code:
the probe is throttled to one a day, so a notice held only in memory is
visible on the morning a release is found and invisible on every launch
after — including the launch right after an update installs. It also
records why a restored notice is re-judged rather than trusted, since
remembering a claim is how you end up advertising the version the user is
already running.

「所有图标共用一份几何」records that the consistency the request asked for is
geometric, not chromatic: the tray mark must be flat and alpha-only
because the shell tints it, and the app mark must be coloured because
nothing tints a dock tile. What made the old icon read as another product
was the gradient, the shine, the shadow and six pastel wedge tints — not
the presence of colour.

Co-Authored-By: Codex <noreply@openai.com>
2026-11-19 21:28:31 +08:00
mula.liu 8a4c9d2b0d feat(icons): one mark, drawn twice — tinted for the shell, coloured for the dock
Reported: "应用的图标问题,要和当前的单色图标保持一致设计" — and clarified:
consistency of *design*, but colour in the panel and the system.

So consistency comes from geometry and filling rules, not from the palette.
`scripts/build-icons.py` now defines the ring once (OUTER/INNER/HUB/GAP/
PLUS_ARM/PLUS_BAR) and emits both marks from it: the tray keeps the alpha-only
flat form because the shell tints it, and the app icon keeps the same figure
in three flat colours drawn from the UI itself — `#FFB25C` wedges and the
`#f04f24` accent the app's own buttons use, on an espresso tile that is also
what shows through the plus punched out of the hub.

What the old icon actually got wrong was not that it was colourful. It was a
gradient tile with a radial shine, a drop shadow and six pastel wedge tints,
which is a different visual language from a crisp flat mark no matter what
hue it wears. Hence the rule: at most three flat colours, and no gradient,
gloss, shadow or per-wedge tint.

The menu-bar icon is byte-for-byte unchanged; the generator reproduces the
committed PNG exactly, so this commit alters only the coloured surfaces. The
hub-inside-a-lighter-ring orientation is deliberately not inverted, because
flipping which end of the mark is dark would read to existing users as a
different product rather than a redraw.

`tauri icon` regenerates the whole tracked set, and the .icns entry set and
.ico size ladder come out identical to the committed ones, so nothing in the
packaging contracts moves.

Co-Authored-By: Codex <noreply@openai.com>
2026-11-19 21:27:01 +08:00
mula.liu e1fb18bc6a feat(updates): announce a new release where the user is looking
Reported: "当有新固件时,在主面板顶部要显示有升级信息。点击跳转到设置的软件更新."

The banner is cheap; the reason it did not exist is that nothing on the main
panel could know about a release. `useUpdateFlow` was mounted only inside
Settings > 软件更新, so closing that panel dropped the only copy of the state.
Mounting a second copy is what the hook already supports — it reads the status
Rust remembers and subscribes to the broadcasts Rust already sends, so neither
view issues an extra network check, and the banner survives closing Settings.

The notice is persisted, which is the part without an obvious answer. The probe
runs at most once a day, so a remembered status that lives only in memory makes
"available" visible on the morning it is discovered and invisible every launch
after — including the launch straight after an update installs, exactly when the
user is most likely to look. Measured, not theorised: the first launch showed no
banner, and ageing the probe stamp was the only way to make one appear.

Remembering a claim means being wrong about it later, so a restored notice is
re-judged rather than trusted. A version that no longer leads is dropped, since
the user is already running it; `ready` keeps its promise only while the
installer file is really on disk, and otherwise says `available`, which is still
true and merely means downloading again. Restore runs before `setup_tray` so the
tray row cannot contradict the banner, and a copy told to stop checking restores
nothing, because that switch means "don't tell me".

Two banners that used to be mutually exclusive now stack. An update notice and a
broken screen-recording permission are different news, and chaining them meant a
user who fixed the first never learned the second existed.

Co-Authored-By: Codex <noreply@openai.com>
2026-11-19 21:25:15 +08:00
mula.liu a95aebe1d0 docs: record the measured idle-cost A/B behind the sampling changes
The section quoted the sidecar's own micro-benchmark, which is not a
number anyone can reproduce. Replaced with a per-process A/B against the
installed 0.1.11 over 60 s idle: the sidecar drops 140 ms to 40 ms, and
that row is the only one that can be attributed cleanly, because the
sidecar has no window and so cannot be confounded by whether the main
window is open.

The main-process row is explicitly not offered as "the discovery thread
saved 40%": repeated runs of the same build swing by about 40 ms, the same
order as the difference itself. Saying so is cheaper than letting the next
person quote a coin flip.

Co-Authored-By: Codex <noreply@openai.com>
2026-11-19 18:24:58 +08:00
mula.liu 9ba619c352 release: prepare Shift 0.1.12
Versions, release docs and the ledger's version row for the two installers.

Both artifacts were built from this tree with the release key baked in, so
either one can verify the signed channel: `Shift-0.1.12-macOS-universal.dmg`
(universal, `Signature=adhoc`, `codesign --verify --deep --strict` and
`hdiutil verify` both clean, no notarization) and
`Shift-Setup-0.1.12-win-x64.exe` (cross-compiled, PE machine `0x8664`
checked, the platform's own frontend confirmed embedded).

Four reported issues, and the fifth request:

- 系统小图标改为黑白单色模板图,由 `scripts/build-tray-icon.py` 重绘。
- Windows 一键安装组件:winget 的良性退出码、未钉源导致的挂死、缺失的提权
  路径,以及装完后进程内 PATH 仍是旧值,四个原因都在同一处。
- 钉在屏幕上的图片新增可用的放大/还原按钮。
- 「最近使用」的收藏行带上常驻指示器。
- 参照 nex_squid 降低闲置开销:伴生进程、Windows 手势监视、发现线程与主窗口
  时钟四处采样节奏,外加最近列表只解码视线内的缩略图。

The Windows install fix still wants a real Windows host: the exact prose a
dismissed UAC prompt prints, and whether each package accepts
`--scope user`, are both handled defensively rather than verified.

Co-Authored-By: Codex <noreply@openai.com>
2026-11-19 17:52:32 +08:00
mula.liu ef68121aeb docs: record the release's constraints in the ledger and the READMEs
Three ledger sections, the 0.1.12 version row, and the user-facing lines
they describe.

「Windows 一键安装运行组件」is the fullest of them: four independent defects
hid in ten lines, and each one reads as "安装失败" from the UI, so the
section records which signal decides the outcome (exit code first, prose
only to break a tie) and why the attempt ladder starts on the command line
0.1.11 already sent.

「钉图窗口的放大与还原」and 「闲置时的取样节奏」record constraints that are
invisible in the code and easy to simplify back into a bug: sizing off
`monitor.size()` instead of the work area, mixing logical and physical
pixels between `set_size` and `set_position`, making a silent Shift copy
deaf in order to save 0.03%, and restoring `default_window_icon()` on the
tray because the colour wheel looked friendlier in the source.

Co-Authored-By: Codex <noreply@openai.com>
2026-11-19 17:52:01 +08:00
mula.liu 7e3bf5efc1 perf: stop sampling when nothing is looking
Reported: "参考nex_squid的代码,进一步降低软件本身占用的系统资源." The
pattern worth copying out of nex_squid's `useSystemSnapshot` is that a
poll stops when nothing can see its answer. Shift had four places that
polled for the life of the process instead.

- `native/macos/drag-bridge.swift` — a separate tracked process polling at
  30 Hz forever, and allocating a CGEvent each pass to read a cursor
  position nothing wants while no button is held. A/B against the shipped
  0.1.11 binary: 20 ms CPU per 6 s idle is now 0 ms, while 12 s of holding
  the button still costs 30 ms either way, so a gesture is sampled exactly
  as finely as before and an idle Shift stops touching the CPU. It runs at
  10 Hz until the button goes down.
- `src-tauri/src/drag_bridge.rs` — the Windows gesture watch slept 33 ms
  unconditionally. An *enabled* ring keeps 33 ms, and not from
  conservatism: the loop has to observe the press to measure the drag
  threshold from where the button went down, so slowing it does not notice
  a gesture late, it never records it. Only a switched-off ring can afford
  a quarter second, since the one thing it waits for is the switch.
- `src-tauri/src/share.rs` — the discovery loop. Measured on the 0.1.11
  bundle: 0.22% CPU with ShiftDrop on, 0.03% with it off. Switching the
  feature off means "stop announcing", never "be deaf" — a silent copy
  still hears beacons so another machine can still find and send to it, and
  expiry must keep running or an hour-old peer pretends to be live forever.
  So the saving comes from waking less often while silent and from skipping
  bookkeeping when there is no peer to expire, not from probing less.
  Announce and sweep intervals are untouched, so discovery stays as fast.
  `a_silent_copy_still_knows_and_still_loses_peers` pins that distinction.
- `src/components/MainWindow.tsx` — the 30 s clock ticked for the life of
  the app while Shift sits parked in the menu bar with this window hidden,
  re-rendering a list nobody could see. It now exists only while the page is
  visible, and an empty library is not worth a render at all.

Co-Authored-By: Codex <noreply@openai.com>
2026-11-19 17:50:57 +08:00
mula.liu 680b4f159e fix(pin): give a pinned image a working enlarge control
Reported: "钉在屏幕上的图片放大按钮无效." There was no enlarge button to
be ineffective — the pin window shipped exactly two actions, copy and
close, so the third press a user went looking for did not exist.

One button now covers both directions: fill the screen, or return to the
size the image was pinned at. An enlarged pin that cannot be shrunk back
is no longer usable as a reference, and a second control for that would
spend hover space on the less common half of the gesture.

Sizing and placement are the whole feature, and both are pinned by tests
because each is correct on the machine that builds it and wrong elsewhere:

- the bound is the monitor's work area, not its size, or the window grows
  under the menu bar and a borderless window cannot be dragged back onto
  itself;
- `set_size` is told in logical pixels and `set_position` in physical
  ones, which on a 2x display is the difference between filling the screen
  and filling it twice from the wrong corner;
- size is set before position, because on a mixed-DPI desktop resizing can
  move a window onto another monitor and change its own scale factor;
- growth is capped at 3x, since the preview is already capped at 1040 px
  and past that the button delivers mosaic, not detail;
- the window grows from its centre and is clamped inside the work area, so
  a pin parked against a screen edge grows inward rather than off.

A fresh pin always opens at the small size. The window is reused rather
than rebuilt, so without that a user who enlarged once would find every
later pin arriving full-screen and soft.

Co-Authored-By: Codex <noreply@openai.com>
2026-11-19 17:49:30 +08:00
mula.liu 36d98044cc fix(ui): show a favourite without a hover, and decode only visible thumbnails
Two changes to the same row, so they land together.

Reported: "操作列表中 收藏的文件 前面增加一个指示器." The star that *sets*
a favourite lives in the row's action cluster, and that cluster is
`opacity: 0` until the pointer arrives, so in a list whose entire purpose
is outliving the files it names, a favourited file looked exactly like an
ordinary one until you hovered it. The mark now sits on the leading corner
of the thumbnail, on its own disc so it stays readable over a picture that
happens to be dark, and `pointer-events: none` keeps it from intercepting
the row's open or drag-out.

While the row was open: `loading="lazy"` only defers a decode, it never
releases one. Measured in WKWebView across 250 real screenshots, 13 images
were decoded at rest and the count climbed monotonically to 136 scrolling
to the bottom without ever dropping, because the whole list is laid out in
one scroller and every image is fetched on mount. Rows are now gated on an
IntersectionObserver rooted on the list itself — an ancestor's clip is
applied before the viewport root is consulted, so measuring against the
viewport would make the margin do nothing — holding roughly 20 live bitmaps
across the same scroll. A row that scrolls away gives its bitmap back, and
the kind glyph stands in, which is the face the row already showed before a
thumbnail arrived.

Co-Authored-By: Codex <noreply@openai.com>
2026-11-19 17:48:28 +08:00
mula.liu 47f8bc69d9 fix(runtime): make the one-click component install work on Windows
Reported: "在windows机器上实测一键安装组件会失败." Four independent
defects hid inside the same ten lines of `install_target_package`:

- `winget`'s benign codes — 0x8A15002B already installed, 0x8A150016 no
  upgrade applicable — were read as failures, so a machine that already
  had the tool was guaranteed to be told the install failed.
- `--source` was never pinned, so on a host that also carries msstore
  `winget` could block on a source agreement `--silent` never answers,
  and the 15-minute deadline turned a hang into the same failure.
- Machine-level packages (Poppler, FFmpeg, ImageMagick, Typst) install
  into Program Files and need admin, and Shift is an unelevated GUI
  process. There was no elevation path at all.
- Post-install verification could not succeed even when the install did:
  a process keeps the environment block it was launched with, so the
  WinGet Links directory written to the registry never reached the live
  `which` lookup.

Attempts now run unscoped, then `--scope user`, then elevated machine
scope, and success is decided by whether the binary is on disk after the
process PATH has been re-read. The four outcomes a user can act on —
declined prompt, installed-but-not-on-PATH, no package source, timeout —
each say so in both languages instead of "安装失败".

Co-Authored-By: Codex <noreply@openai.com>
2026-11-19 17:46:52 +08:00
mula.liu 287b3af09d fix(tray): let the system draw a black/white mark, not the colour wheel
Reported: "收到系统小图标时采用黑/白单色图标." The tray was handed
`app.default_window_icon()`, the 512 px gradient wheel, and `tray-icon`
scales whatever it is given to 18 pt: on the menu bar and in the Windows
notification area that is a blob, not a mark.

The new icon is drawn flat and alpha-only by `scripts/build-tray-icon.py`,
in the shape of the ring the app itself paints so the two read as one
product, and `icon_as_template` lets macOS tint it black or white per the
menu bar. Windows has no tinting to ask of, so it gets the same glyph at a
size it can actually render.

Co-Authored-By: Codex <noreply@openai.com>
2026-11-19 17:45:07 +08:00
mula.liu 85de468bc3 fix(drop): 拒绝改由光标表达,不再往按钮上叠禁止图标
按钮处理不了这次拖拽时保持 disable,鼠标光标换成禁止符号;能服务的光标仍是
AppKit 画的「+文件」。页面上那一层 .drop-ban 叠加层删掉了。

macOS 上外部文件走原生拖放通道,而 wry 的 draggingEntered:/draggingUpdated:
固定回报 NSDragOperation::Copy,页面改不动那条通道的光标。新增
src-tauri/src/drop_cursor.rs:页面把判定用 set_drop_refusal 交下来,原生侧在
运行时替换 wry 私有类的这两条方法(类名带 crate 版本,按前缀扫描),把 Copy
收窄成 NSDragOperationNone。两次都先调用 wry 自己的实现再改返回值——那次调用
正是 over/dragover 的投递入口,拒绝时提前 return 会让页面再也收不到指针移动,
一次拒绝就粘到整段手势结束。回报 None 时既没有 performDragOperation: 也没有
draggingExited:,所以拒绝期间盯一次鼠标松开,用 drop://cancelled 收起面板状态。
Windows 不需要这套:OLE 的回报跟随页面设置的 dropEffect。

实测:PDF 从「最近使用」和从访达分别拖到「文字识别」,两处都是按钮禁用+光标
禁止、无 hover、松手静默;中途拖到能服务的按钮立刻恢复并正常弹出圆环。
2026-11-07 12:43:06 +08:00
mula.liu 935de413d8 docs: 收录 2026-11 全面评审的修复与约束
评审轮改动此前只在本地:更新链路的下载与校验收紧、分享接收的失败上报与
「完成后在访达中显示」、截图/录屏/文件读取的边界处理,以及新增的
ARCHITECTURE / SPECIFICATION / SECURITY / CONTRIBUTING 四份文档和
publish-release、CI 的调整。 ENGINEERING-NOTES 补上「更新、分享与读取边界」
一节,把这些不应被简化掉的约束写进台账。

这些改动已随 0.1.11 构建物发布,此提交只是让源码与产物在服务器上保持一致。
2026-11-07 12:42:35 +08:00
mula.liu cb758c6a03 fix(drop): disable a spent button and mark it, so both drag paths agree
A file from the file manager arrives through the native drag channel, where wry fixes the macOS drag operation to Copy, so the cursor could never carry the refusal. The answer now lives on the button: it is disabled for the length of the gesture, shows a prohibition sign over its own label, and takes no hover feedback. A Recent row also stops being reported as 'no files' by the native enter, which had lit all three buttons for a file two of them could not serve.
2026-11-07 02:34:44 +08:00
mula.liu f625832856 fix(drop): show a forbidden cursor for spent buttons 2026-11-07 01:15:10 +08:00
mula.liu 3057c65d27 fix(drop): stop the drag-end fallback from eating the drop it waits for
0.1.11 made a dragged row do nothing on every button. The release added to
`dragEnd` landed the gesture immediately, and on macOS the native drop is
reported *after* the button is up: the fallback consumed the gesture with
`dragEnd`'s own position - which Chromium reports as the origin, not as where
the pointer stopped - and the drop that followed was read as a finished
gesture and thrown away. Three buttons, no feedback, on both platforms.

The fallback now waits, and only claims the gesture when it can actually name
a button. A release that aims nowhere stands back, so a real drop that arrives
later still runs, and `landed` keeps the two from both acting. A position of
(0, 0) is treated as no position reported, which is what it is on Chromium, and
answered with the last position the drag moved through.

A row dragged out of the window now carries the file rather than the row. Left
alone the engine photographs whatever is under the pointer, and a Recent row is
a thumbnail, a name, an age, a folder path and five hover buttons - a strip of
the list, which reads as the whole list coming along. The ghost is the file: the
picture the list already decoded, or the extension when there is no picture,
with the name under it, matching the figure the ring draws for a file it is
about to convert.

Rebuilt both 0.1.11 installers from this tree.
2026-11-07 00:53:09 +08:00
mula.liu be84d5640f release: prepare Shift 0.1.11
Versions, release docs and the ledger's version row for the two installers.

Both artifacts were built from this tree with the release key baked in, so
either one can verify the signed channel: `Shift-0.1.11-macOS-universal.dmg`
(ad-hoc signed, `hdiutil verify` valid, no notarization) and
`Shift-Setup-0.1.11-win-x64.exe` (cross-compiled, PE machine `0x8664` checked,
frontend confirmed embedded).

The Windows drop fix needs a real Windows host before it can be called done:
this forge has no second machine, and the two delivery paths that went silent
are exactly the ones only WebView2 exercises.
2026-11-06 23:14:57 +08:00
mula.liu 2bf63f2366 fix(drop): let Windows finish a row dropped on a button, and spend every button alike
A row dragged from Recent onto any of the three buttons did nothing on
Windows, and reported itself as a refusal. Both delivery paths can stay
silent on that platform, so the gesture now ends in `dragEnd`, which always
arrives and carries a point to hit-test.

The two paths are silent for two separate reasons. A Recent row is a real
file drag, and WebView2 registers its own OLE drop target on the window, so
the DOM drop never reaches the page; that target then reads only `CF_HDROP`,
which an in-page drag does not carry, so `enter_is_valid` stays false and the
native `over`/`drop`/`leave` are withheld too. `landed` keeps the fallback an
execution rather than a second one, and a release outside the window runs
nothing, because the hit-test has slack around every button and the row was
meant for Explorer.

An idle ring also stops getting in the way. It is borderless and always on
top, so one left open from a previous drop sits over the buttons; macOS
passes mouse events through its transparent pixels and Windows hands the
gesture to whichever HWND is under the pointer. Starting a row drag now puts
such a ring aside, while a ring owned by an in-flight Shift-drag gesture keeps
waiting for the release that decides its fate.

The availability rule stops being OCR's alone. `lib/dropCapabilities` answers
one question - can this button serve these files - and the button's spent
state, its candidacy while a drag is in flight, and the drop itself all ask
it, so Convert spends for a `.key` the way OCR spends for a PDF, on both
platforms, whichever drag source carried the file. Tools never spends: its
ring ends in Open, Reveal and Send. An unknown payload is not a refusal,
because spending on an empty path list would disable a gesture that works.

Two other reports this round were not about the drag at all:

- A file received over the LAN now reveals itself only when "show results in
  Finder" is on, the way a finished conversion already behaves, and a batch
  reveals once at its last file rather than once per file.
- Missing runtime components install as one request. Installs run one after
  another because the package managers hold a single lock, and one package
  answering two rows - Poppler for both PDF rows on Windows - is installed
  once. The panel re-reads the machine afterwards either way, since a batch
  that stopped at its first failure may have landed everything before it.
2026-11-06 23:13:53 +08:00
mula.liu 84c4056a0d feat(ui): spend the OCR button instead of explaining it, and unify how the app speaks
Two changes, one goal: the same gesture and the same news should look and read the
same way wherever they happen.

Dropping a file OCR cannot read used to be answered after the fact - the drop ran,
refused, and popped "文字识别仅支持图片格式". The user had to do the wrong thing
first and then be told. Now the decision happens when the drag starts: MainWindow
records the files in flight (`dragPaths`, from the native `enter` payload or from a
Recent row's `dragstart`) and marks the button spent for the whole gesture with
`is-drop-unavailable`, so the refusal is visible before the pointer is committed and
releasing on it is inert and silent. Both drag sources were verified in a production
build, including the reverse case: an image still lights the button and opens the
editor. `over` carries no paths and `dataTransfer` is unreadable during `dragover`,
so the paths have to be registered at the start - a later refactor that reads them
from the move event gets nothing.

The audit that followed found the same class of problem in the messaging, and worse:
the app said one thing three ways.

Prompts were the largest gap. A Rust-authored sentence reached the screen verbatim,
so a Chinese interface answered Convert with `Saved 1 file.`, and macOS's first-run
write-permission failure was one long English line. Rust now names the outcome or the
failure (`outcomeSaved`, `errCaptureBlocked`) and the interface words it; `translate`
still returns an unknown key unchanged, so genuine technical detail - `ffmpeg exited
with status 137` - is reported as it is rather than flattened into a generic line.
`feedback.test.ts` scans the Rust sources and fails on any code without wording in
both languages, and on any outcome the worder does not know; both directions were
checked by breaking them. It also asserts the two dictionaries list identical keys,
which was silently false before.

Durations were arbitrary: the identical share result showed for 2.6s in the ring and
3.6s in the main window, one file's `processing` message lived 12s in one branch and
120s in another, and generic errors used 2200/2400/3000/3200/3800. `FEEDBACK` is now
the only ladder, and `hold` is reserved for tasks that report progress and must not
expire mid-task.

Error text had three normalisation idioms, two inside the same file; `messageFrom()`
is the only one left. Share failures reported twice at once - a banner saying "could
not send" plus a toast naming a different reason - and now report once, in the panel,
with the real reason. The editor's failure notices wore a green check and lasted 1.4s;
a clipboard refusal was silent, and an OCR failure printed into the text box as though
it had been read from the image.

Disabled controls were dimmed at nine different opacities across the app, three
buttons were disabled in JS with no CSS at all, and several `:hover` rules answered a
disabled control. `--dimmed` plus `cursor: default` and `:not(:disabled)` hovers is now
the single treatment, and the OCR drop-target state joins it. Every transient message
is a live region, three unlabeled icon buttons are named, Escape closes the settings
sheet and the first-run panel, the main toast is raised above every sheet of its
window, and nine raw px font sizes went back onto the type ladder.

Out of scope, deliberately: the accent is still ~87 literals rather than one variable,
and the muted warm greys are still written many ways. Both need a visual pass, not a
search and replace.
2026-11-01 16:18:20 +08:00
mula.liu 75405d84c4 release: prepare Shift 0.1.10
Version and release docs for the two installers.

The ledger's drop-gesture section now says which release did what: 0.1.9 removed
the action-ring fallback and added the explaining toast, 0.1.10 removed the toast
and the panel's own drag animation. Both were written as one change until the
published 0.1.9 tree was checked, and it still had the toast and the is-dragging
rule.
2026-11-01 12:48:57 +08:00
mula.liu 45325f0a6c feat(ui): quiet the drop gesture, and refuse files OCR cannot read
Four changes to how the main window answers a drag.

The tray row said "识别图片文字…" where the window says 文字识别, and the ellipsis
made it read like a preview rather than a command. It now carries the same words
as the button, in both languages.

OCR refuses a file it cannot read instead of handing it to the format ring. The
fallback was meant to be helpful - "the format ring is more useful than a failing
OCR run" - but from the user's side it looked like the button they dropped on had
ignored their file and opened something else. A drop on OCR is now either a read
or a refusal that names the reason.

Dragging an outside file and dragging a Recent row are now the same interaction.
The panel used to scale, recolour and swap its heading when a file entered the
window from Finder, while a row dragged from the list did none of that, because
an in-webview drag raises no native drag events. The panel no longer animates at
all: the is-dragging class, its two rules, and the transition they drove are
gone, and feedback lives only on the three buttons, which both paths already
share.

Releasing off a button no longer pops the toast. 0.1.9 added it to explain why
nothing had run, but it fired at the pointer, in a place the user had not chosen
to aim at, and the buttons already show where a drop lands. Off a button,
nothing happens, silently, on both paths.

The version comparison test gains the patch-component case: 0.1.10 is the first
release whose patch number reaches two digits, and the rule was only asserted for
the minor position.
2026-11-01 12:48:36 +08:00
mula.liu b524b69fc8 release: prepare Shift 0.1.9
Version, the release docs and the engineering ledger for the two installers.

The ledger gains a section on the Runtime components panel, because the bug it
records was invisible rather than broken: a dropped `return` made macOS answer
the document-PDF row with a Pandoc pipeline it never uses, so the panel reported
a missing component with no way to install it. The section states the invariant
that a row may not read as missing unless it also offers a way forward, which is
the part a future refactor is most likely to lose again.

README sections that described dropping files as "opens the action ring" now
describe the aimed rule, and the capture-behaviour option is gone from the
Settings tour.
2026-11-01 12:04:03 +08:00
mula.liu e6818e029e feat(ui): drop onto a button, and read the panels again
Three changes to the main window, requested together.

A file drop is now aimed. The three buttons are the only drop targets and the
button you release on is the action it runs; a release anywhere else on the
panel runs nothing and says so, instead of opening an action ring nobody asked
for. This replaces the old fallback in both delivery paths - the DOM handler and
the native webview drop - because two routes to the same gesture must not have
two different rules. dropFiles therefore takes a required action rather than an
optional one, and the actions ring is still reachable from the Shift-drag
gesture and from a file type with no convertible formats.

Recent rows can be dragged back onto those buttons. A row leaves Shift as a real
file drag, so one gesture can be reported twice, once by the DOM and once by the
native events, and which arrives first depends on the engine; recentDrag.landed
lets only the first through. The hint under the buttons names the affordance,
which is otherwise invisible, and all three buttons light up as candidates for as
long as any drag is in flight - a native one via dragging, an internal one via
rowDragging, since a row drag inside the webview does not raise native events.

Type sizes are on one ladder now (--text-3xs to --text-hero) instead of being
invented per rule, which is how the settings drawer drifted down to 7-8px. 92
declarations under 12px became one. The drawer is wider, its tabs are pills, and
the runtime, share, setup and radial surfaces moved onto the same scale.

The capture-behaviour setting is removed: the CaptureAfter enum, its field, its
command and its UI. A finished capture always shows the result card, falling back
to the editor when a card window cannot be created. Stored settings that still
carry captureAfter load unchanged, because the struct does not deny unknown
fields.

Also dropped 23 i18n keys no surface used, one dead re-export in history.ts, and
a stale .home-drop p rule that was outranking the new hint and sizing it as body
copy. tsconfig now turns on noUnusedLocals, noUnusedParameters and
noFallthroughCasesInSwitch, which is what keeps the dead imports from coming
back; the tree already passes all three.
2026-11-01 12:02:24 +08:00
mula.liu 23973cee8f fix(runtime): report the bundled renderer for document PDF on macOS
0.1.8 turned the renderer branch of document_pdf_component into a bare expression,
so the component it built was discarded and every Mac fell through to the Pandoc
row. That row requires a LaTeX or Typst engine macOS never runs, so Settings ->
Runtime components read 'Pandoc PDF pipeline: missing' with no Install button,
because document-pdf has no macOS install target.

The two platforms now have their own function, like every other platform-specific
row in the file, so the fall-through is a compile error rather than a silent one.
The Pandoc half is still real on macOS - it turns a picture-bearing DOCX into HTML
for the renderer - and it keeps its own row and install target.

Two guards cover it: the row must stay bundled on macOS, and no row may read as
missing without either an install entry or a non-external source. The second
reproduces the reported symptom exactly when the fix is reverted. The matching
PDF-engine error now names what the panel actually shows on each platform.
2026-11-01 11:59:42 +08:00
mula.liu 0aa929b2f5 build: bake release key into macOS packages 2026-10-27 18:49:21 +08:00
mula.liu df35111326 release: prepare Shift 0.1.8 2026-10-27 18:45:32 +08:00
mula.liu d9bb079ec2 docs: record discovery, conversion and macOS packaging regressions 2026-10-26 20:24:21 +08:00
mula.liu 708bb5b0b1 v 0.1.7 2026-10-26 20:11:20 +08:00
mula.liu 35cca5987c release: 0.1.6 — discover peers across filtered Wi-Fi 2026-10-26 19:53:35 +08:00
mula.liu 579fc6218a release: 0.1.5 — device discovery survives an upgrade
An upgrade emptied the "Send to a device" list, with nothing in the UI to say why.
share.rs was not the cause: the last commit to touch it is b732b91, which only
recorded received files in the history library.

The cause is the macOS Local Network grant. Builds are ad-hoc signed, so every
rebuild is a new code identity, and the grant is keyed to that identity rather than
to the bundle id - 系统设置 now holds five Shift rows, one per build this machine has
run. The declared NSLocalNetworkUsageDescription was missing, so a new identity was
never asked and simply received no group traffic: beacons still went out, direct
unicast was still answered, only the list stayed empty.

That declaration is bc5d3b8. Both READMEs and the packaged macOS note now name the
permission and say what a denial looks like, because unlike Screen Recording it
produces no error - just an empty list. The Developer ID item in M5 records that
this has now actually cost a user a feature, not merely "may reset approvals".

Also ships the Updates panel button pair from 22ff93a, which was merged after 0.1.4
and never published.

Gates: cargo fmt --check clean, cargo test --lib 152 passed / 2 ignored, pnpm check
clean, pnpm test 38 passed. Channel verified by downloading both artifacts back
from the raw endpoint and re-hashing against the manifest, and by the ignored feed
test run with the release key. 55,504,368 B DMG and 24,521,557 B installer.
2026-10-26 19:14:05 +08:00
mula.liu bc5d3b80d4 fix(share): declare the local network usage description
Discovery stopped working after an upgrade without any change to share.rs: the
last commit to touch it is b732b91, and only to record received files in the
history library.

What the upgrade does change is the code identity. Builds are ad-hoc signed, so
every rebuild is a different binary to macOS, and the Local Network grant is keyed
to that identity rather than to the bundle id - 系统设置 now lists five separate Shift
rows, one per build this machine has run. The copy installed by the upgrade has an
identity macOS has never been asked about.

It also has no way to be asked: NSLocalNetworkUsageDescription was never declared,
and macOS shows no local network prompt without it. The app then simply does not
receive the group traffic, silently. The panel confirms the symptom is not specific
to Shift's port - this Mac received zero mDNS and zero SSDP packets while its own
beacons looped back fine, and unicast to a peer still works, which is how the two
other copies were found by hand.

Declaring the string lets the prompt appear for a new identity. The durable fix is
the stable Developer ID identity already tracked under M5, which stops the grant
from being lost on every single build.
2026-10-26 18:54:25 +08:00
mula.liu ed4acee2e4 docs: record the published 0.1.4 channel
Both artifacts were downloaded back from the raw endpoint and re-hashed against
the manifest rather than read from the local build output, and shift-update.sig
verifies over the manifest's exact bytes against the release public key.

The retired .pkg left the branch with this publish: the tree is rebuilt from
artifacts/ on every run, so dropping it from the build was enough.
2026-10-26 17:36:47 +08:00
mula.liu 48b24acc2f fix(release): derive the public key from either key file
A raw ed25519 private key is 32 bytes and so is its public half, so
'pubkey KEY_FILE' could not tell them apart and printed the private bytes as a
public key. Baking those into a build produces a key no signature can verify
against, and build.rs accepts any 64 hex characters, so the mistake is silent at
build time and shows up on a user's machine as an update check that cannot verify
the manifest.

Anything not named *.pub is now read as the private key it came from and expanded,
so both documented spellings of the command yield the same key.
2026-10-26 17:29:59 +08:00
mula.liu 334082cd9b release: 0.1.4 — figures survive every conversion
Three conversion fixes since 0.1.3, all found with the same two documents:

- DOCX -> PDF kept the text and dropped the pictures, because AppKit's OOXML
  reader returns an attributed string with no attachments at all. An image-bearing
  DOCX now goes through Pandoc into HTML before the same printer, which keeps
  headings, tables and figures. Image-only DOCX files still take the page-scan
  path, and a machine without Pandoc falls back to the old text-only result
  rather than failing.
- Markdown -> DOCX/RTF/HTML lost the same figures for a different reason: a
  Markdown image is a path, Pandoc has to open the file to embed it, and it
  resolves that path against its own working directory, which for a running Shift
  is nowhere near the document. It now gets --resource-path of the source folder.
- A picture that did arrive printed as a cropped fragment. The layout sizes an
  attachment from the density header in the file - 72 dpi when it carries none -
  and ignores CSS, so a 1600 px screenshot asks for 1200 points on a 595 point
  page. The renderer fits oversized attachments to the text column, scaled and
  not resampled. Markdown and HTML documents with wide images are covered too.

The Updates panel's install and discard buttons are one matched pair now, sized to
the panel's own 12px scale instead of the dense 8.5px rows of the Runtime tab they
were borrowed from, and the row drops the discard control while an install is
counting down.

Gates: cargo test --lib 152 passed / 2 ignored, pnpm check clean, pnpm test 38
passed. Artifacts verified from the built bundles, not the build log: the DMG
carries 0.1.4, x86_64 + arm64, Signature=adhoc, and its bundled doc-render places
a wide image at 499pt inside the page; the Windows exe carries the feed URL and
the release public key.
2026-10-26 17:24:22 +08:00
mula.liu 22ff93a5fc fix(updates): make install and discard read as one decision
The row under a verified download took both controls from the Runtime tab, where
rows are dense: .runtime-install is 27px tall at 8.5px text and .settings-link is
a bare 8.5px text link with no box at all. In the Updates panel, whose own status
text, artifact chip and version pill are 12px, that produced an orange pill beside
floating text - the pair did not look like one choice, and the text link did not
read as a button you could press.

They get their own .update-action now, sized to the 12px scale the panel already
uses, and both take the same box so the eye groups them. The commit variant wears
the app's existing primary gradient, so the action that does something is the loud
one and Discard stays the quiet alternative.

"放弃" also stops competing at the moment it stops meaning anything: once the
handoff script is written the app is counting down to exit, so the row collapses
to the single control still in play. update_clear only resets the status, so this
is presentation, not a saved deletion.

updateInstalling was a whole sentence inside a fixed-height pill, which is what
made that button stretch unevenly against its neighbour. It is "正在安装…" now;
the panel's own closing paragraph already says Shift will restart and interrupt a
running conversion, so the button does not have to repeat it.
2026-10-26 16:22:34 +08:00
mula.liu 9950607cd4 fix(markdown): keep images when a document with relative figures is converted
Same class of bug as the DOCX one, different cause. A Markdown image is stored as
a path, so Pandoc has to open the file to embed it - and it resolves that path
against its own working directory, which for a running Shift is nowhere near the
document. 智飞空地一体化低空应用平台技术方案.md references images/system-architecture.png
and images/production-deployment-topology.png; both came back as "Could not fetch
resource", and Pandoc replaced each figure with its alt text. The DOCX was 22 KB
with no word/media entry at all.

Passing --resource-path of the source's own folder fixes DOCX, RTF and HTML in one
line. It is additive rather than a replacement: Pandoc still searches the working
directory too, so a figure that resolved before still resolves - verified with two
images of the same relative name in two different folders, both of which landed in
the output.

md -> pdf already worked, because Pandoc leaves the path untouched there and hands
the HTML to the renderer with a base URL pointing at the document's folder; only
the routes where Pandoc must read the bytes itself were affected. md -> html was
the quieter case: --self-contained promises one portable file, and without the
figures inlined the single HTML opened as a broken image icon.

keeps_markdown_images_when_converting_to_word pins docx and rtf. It fails against
the previous build with "docx ... lost the picture", and the RTF check reads the
pngblip marker rather than assuming a zip.
2026-10-26 15:43:59 +08:00
mula.liu 19602b79f0 fix(docx): keep figures when a DOCX is printed to PDF
Two bugs, both visible in 平台技术方案(带图测试专用).docx: one figure, eleven
tables and 357 paragraphs, whose PDF came out as a text-only document.

First, AppKit's OOXML reader returns the text of a DOCX and silently drops every
inline drawing - an attributed string with zero attachments for a document with a
picture in it. That is not a rendering loss, it is a parse loss, so no margin,
print setting or layout change can recover the figure: by the time the page is
drawn, the picture is simply not in the attributed string. A DOCX whose body
contains an image is now converted by Pandoc into HTML with its media extracted
beside it, and the HTML - which Pandoc reads faithfully, headings, tables and
pictures - goes to the same printer. Image-only DOCX files keep the existing
page-scan path, which runs first, and a machine without Pandoc falls back to the
old text-only behaviour rather than failing.

Second, once pictures could arrive they printed as a cropped fragment of
themselves. The layout sizes an attachment from the density header inside the
file, assumes 72 dpi when there is none, and ignores both bounds and CSS width: a
1600 px screenshot at 96 dpi asks for 1200 points and lands on a 595 pt page.
Measured across PNG pHYs and JPEG JFIF, the placed width is exactly
pixels / dpi * 64. A plain NSTextAttachment does honour bounds, so the renderer
now replaces any oversized attachment with one that fits the text column; the
print operation scales it to the page with the text, landing at 499pt, which is
the printable width the body text already uses. Nothing is resampled, only the
size the layout asks for changes. This also fixes Markdown and HTML documents
with wide images, which had the same overflow.

Both halves are pinned by tests, and both fail against the previous build:
keeps_pictures_when_rendering_a_text_docx_to_pdf sees no image at all, and
fits_a_wide_picture_onto_the_page reports a picture placed 1066.77pt wide on a
595pt page. The second reads the Form XObject's /BBox, so it needs no content
stream decoded and no pixel compared.
2026-10-26 14:50:52 +08:00
mula.liu 58949bebd2 docs: macOS ships the DMG alone, the pkg is retired
The .pkg existed to hand a fresh machine a copy that was not quarantined, so it
opened with no prompt at all. The DMG has now been verified to work on a fresh
machine - drag to /Applications, one approval on first launch - which is the same
one approval the DMG has always needed. So the second container bought nothing
and is dropped: one artifact per platform is less to keep straight, and nothing in
the channel changes.

What the docs say instead of recommending the package: install the DMG, allow the
first launch once. In-app updates still never prompt, because Shift's own
downloader sets no quarantine flag. The macos manifest key stays a DMG for the
reason it always did - the in-app handoff mounts it with hdiutil.

scripts/macos-pkg.sh stays in the tree and RELEASE.md says plainly that it is no
longer part of a release, so the next reader does not wonder whether a publish is
missing it. The 0.1.3 branch still holds the retired package because removing it
means re-publishing; it has no manifest key, so nothing can be served it.

The packaged note is rewritten for the same reason: it told a reader to download a
file that will no longer be built, and its verification section checked the
package instead of the image.
2026-10-26 10:47:48 +08:00
mula.liu f2a1ff6d52 docs: record the macOS packaging scheme next to the artifacts
The ad-hoc signature, the two containers and their different jobs, and how to
verify a download were explained only in docs/RELEASE.md. A user who downloads
Shift-macOS-universal.pkg from the releases branch never sees that file, so the
same write-up now also ships as artifacts/README-macOS-universal.txt, beside the
per-arch Windows notes that already travelled with the channel.

Every claim was checked against the shipped artifacts rather than carried over:
the .pkg payload is Signature=adhoc, pkgutil --check-signature reports no
signature for the package itself, and shasum -c passes on the published set. The
file is force-added to the repository even though artifacts/ is gitignored, so
the scheme is documented where a contributor reads it.
2026-10-26 10:40:33 +08:00
mula.liu 4ed236bd9e docs: correct why older builds opened
The commit before this one claimed releases that opened fine "were ad-hoc signed,
predating the config entry". They were not: every DMG this channel has published,
back to the first 0.1.0, carries Authority=Shift Local Signing, because Tauri selects
a keychain identity itself when signingIdentity is unset.

What survives is the verified part — a self-signed identity that cannot chain to a
trusted root elsewhere, and an ad-hoc binary that gets an approvable prompt instead of
a hard refusal — so the build change stands and only the archaeology is corrected.
2026-10-26 09:48:18 +08:00
mula.liu 42d6424e9f build: sign macOS ad-hoc so other machines can open the app
The published DMG was refused on a fresh Mac as damaged/malware, with only
"Move to Trash" — no 仍要打开, no pane entry to approve. That is not the
"unidentified developer" box, and it is worse than shipping unsigned.

Cause: `signingIdentity: "Shift Local Signing"`, a locally created self-signed
certificate. On the build host it looks healthy — `codesign` reports valid on disk
and satisfies its Designated Requirement, `spctl` even accepts it — but the
certificate lives only in this keychain, is self-issued and CA:FALSE, so it cannot
chain to a trusted root on any other machine. Sequoia turns an unchainable
signature on a quarantined download into a hard refusal.

Releases that opened fine predates that config entry and were ad-hoc signed, which
gets the softer prompt the user can clear (right-click → 打开, or 允许 in 隐私与安全性).
Set signingIdentity to "-" explicitly so this cannot drift back, and rebuild all
three macOS/Windows artifacts from it: app is `Signature=adhoc`, the DMG carries no
signature of its own, and the pkg payload is ad-hoc too.

Docs corrected: the earlier explanation in "macOS packaging" blamed the DMG
container, which was wrong — the container still matters for first install (an
installed app is not left quarantined, so a .pkg opens with no prompt), but the
malware verdict came from the certificate.
2026-10-26 09:44:53 +08:00
mula.liu e9760b8514 release: publish a macOS .pkg so a fresh machine can open Shift
Builds are signed with the self-signed "Shift Local Signing" identity and are not
notarized, so Gatekeeper calls Shift an unidentified developer. Dragging the app out
of the DMG hands that verdict to the app: a fresh machine refuses the first launch
until the user approves it in 系统设置 → 隐私与安全性.

`scripts/macos-pkg.sh` wraps the built bundle with pkgbuild and the channel now
publishes `Shift-macOS-universal.pkg` alongside the DMG. Verified rather than
assumed: installing a quarantined copy of the package leaves the installed app with
only `com.apple.provenance` — no `com.apple.quarantine` — and `codesign --verify
--deep --strict` still passes, because pkgbuild neither re-signs nor strips the
payload (CDHash identical across built bundle, DMG copy and pkg copy: 3b7fac6c…).

The package is unsigned because `productsign` needs an Installer Signing certificate
and this build has only a code-signing identity, so the approval does not vanish — it
moves to the download, once, before anything is installed. `installer -pkg … -target
/` needs no approval at all.

The `macos` manifest key stays a DMG on purpose: `update_install`'s handoff mounts the
image with hdiutil, so pointing the key at a .pkg would break in-app updates.
2026-10-26 09:22:39 +08:00
mula.liu c98d554a88 release: 0.1.3 — the Recent list survives an upgrade
The Updates tab drops the "installs the ... artifact" chip, and the state label
now leads the status row instead of floating right, since the version it used to
sit beside moved into the header.

The file list no longer disappears on upgrade:

- The library already lived in the app data directory, outside the bundle, so an
  in-app update or reinstall never touched `history.json`. What it never had was a
  bridge from the store 0.1 used: that list lived in the webview's own storage,
  which a reinstall does not survive, so the first launch of a newer build showed
  an empty Recent list over files that were all still on disk.
- `history_import` merges those rows once, on the first launch after an upgrade.
  An entry the library already lists keeps its own timestamp, the kind is
  re-derived from the file rather than taken from the older store's guess, and the
  legacy key is dropped only after the merge succeeds so a later upgrade cannot
  resurrect rows the user has since removed.
- `HistoryEntry.kind`/`.at` gained serde defaults, so a row from the older store
  degrades to a usable entry instead of failing the whole import.

Verified on this machine that the legacy `shift.recent` key is still present in the
installed app's webview store, so the import has rows to adopt.

Gates: cargo fmt --check clean, `cargo test --lib` 149 passed / 2 ignored,
`pnpm check` clean, `pnpm test` 38 passed, `pnpm build` ok.
2026-10-26 04:02:31 +08:00
mula.liu 15498482fa release: 0.1.2 — install the update in place, and lead the tab with the host
Updates now finish without leaving the app. After the SHA256 gate, macOS replaces
its own `.app` from the downloaded disk image and Windows runs the same NSIS
installer with `/S`; both from a detached helper that waits for this process to
exit and then opens the new copy.

- `update.rs`: `bundle_path` (is this a bundled run?), `can_replace` (is the
  enclosing folder writable — checked before quitting, not after), and the two
  handoff scripts. Every macOS failure path still `open`s the bundle it had, so a
  failed swap leaves Shift running rather than closed.
- `windows-installer.nsi`: `${Silent}` skips the "Shift is running, close it?"
  box, which would otherwise abort an update nobody is there to answer.
- `useUpdate.ts`: a failed install reports its own error and keeps the verified
  download, instead of downgrading a "ready" status to "error".

The Updates tab now leads with two lines above everything else: system,
architecture and running version, then the feed URL. The header no longer repeats
the version in the status row.

Verified end to end on macOS: a 0.1.1 bundle swapped by the published 0.1.2 image,
`codesign --verify --deep --strict` clean afterwards, image detached, and the
busy-image path shown to reopen the old copy.

Gates: cargo fmt --check clean, `cargo test --lib` 146 passed / 2 ignored,
`pnpm check` clean, `pnpm test` 34 passed, `pnpm build` ok.
2026-10-26 03:41:47 +08:00
mula.liu b732b91ea6 release: 0.1.1 — update channel, settings/UI trims, macOS + Windows x64 only
Update path
- `src-tauri/src/update.rs`: manifest fetch, semver compare, optional ed25519
  signature check against a key baked in at build time, then a size + SHA256
  gate before an installer can run. `UpdateStatus` now reports the artifact key
  it resolved so the UI can name what it will install.
- Settings → Updates panel shows the feed, the check/download state machine, and
  the local OS, architecture and release artifact for this machine.
- `scripts/publish-release.sh` + `scripts/release-key.py` publish installers and a
  signed `shift-update.json` onto an orphan `releases` branch.

Scope trims requested for this release
- Settings: the Radial rings tab is gone, so ring orders are read from
  `settings.json` with no in-app editor for now.
- Main window: the update entry in the header bar is gone; updates live in
  Settings → Updates only.
- Release targets are macOS universal and Windows x64. ARM64 and Linux stay
  buildable and are no longer published; `docs/RELEASE.md` records the decision.

Gates: cargo fmt --check clean, `cargo test --lib` 142 passed / 2 ignored,
`pnpm check` clean, `pnpm test` 34 passed, `pnpm build` ok.
2026-10-26 03:06:46 +08:00
mula.liu b96bb03bbd 修复bug 2026-10-25 18:23:08 +08:00
mula.liu c6eff4e131 修复bug 2026-10-25 12:24:05 +08:00
Shift 3227e7c2d9 hotkeys: duplicate copy stays alive and reports dead hotkeys; live state derives from instance ownership 2026-10-21 10:18:19 +08:00