The bulk of them were left behind when the keystroke latency logging was
removed: PerfTimer objects and *ElapsedMs doubles that nothing reads any more
(C4189). Dropping them also takes the QueryPerformanceCounter calls back out of
the keystroke path. The timers that still feed DebugTsfKeyLatency (Ipc.cpp,
KeyEventSink.cpp, KeyHandlerEditSession.cpp) are untouched.
The rest, all behaviour-preserving:
- C4463: the four CompositionProcessorEngine configuration flags were one-bit
BOOL fields, so every `= TRUE` stored -1. They are plain BOOLs now.
- C4456: renamed the shadowing inner locals (pLI -> pItem, focusToken ->
expectedFocusToken).
- C4100: commented out the parameter names that are part of a TSF interface but
unused in the implementation.
- C4702: dropped the unreachable `return 0;` after FreeLibraryAndExitThread,
which is __declspec(noreturn).
- C4102: removed the two vestigial NoPresenter labels and the empty branches
that used to jump to them.
- C4267/C4244: made the narrowing conversions in the pipe write and the offline
key handler explicit.
- C4996: SharedConfigPath now uses GetEnvironmentVariableW instead of the
deprecated _wgetenv.
x64 and x86 Release both build clean; ctest passes in both.
Word does deliver the bare-Shift release to the key-event sink; what it
does not deliver is a fresh GetKeyState() snapshot alongside it. Hosts
that call into ITfKeystrokeMgr outside the message dispatch see the
modifier state as of the last message the thread pulled, so
Global::PureShiftKeyUp/IsShiftKeyDownOnly -- both derived from
GetKeyState() -- can still describe the previous keystroke. The release
match then silently fails even though OnTestKeyUp and OnKeyUp both ran.
Match on the released virtual key plus _shiftHotkeyArmed instead.
The latch already means "this Shift went down alone and nothing else has
been pressed since" (_TrackModifierHotkeyArming disarms on any other
key-down) and it self-clears, so TestKeyUp/KeyUp cannot double-toggle.
Arming reads GetAsyncKeyState(), which reports the real keyboard rather
than the host thread's queue. Same treatment for bare Ctrl.
This is how Weasel and WindInput handle it -- neither installs a
keyboard hook for the toggle -- so revert the hook back to mintty-only.
mintty is a genuinely different failure: it offers no bare modifier
key-up to the sink at all, so there is no event to match against.
70930df stopped eating the bare-Shift release so JetBrains double-Shift
keeps working, and moved the toggle into OnTestKeyUp because TSF does not
call OnKeyUp after a FALSE test result. Word never completes a bare-Shift
release through ITfKeyEventSink at all, so neither callback runs and the
CN/EN toggle was lost there.
The mintty workaround already solves exactly this shape: a thread-local
WH_KEYBOARD hook watches the real bare-Shift press/release and feeds it
back into the deferred hotkey path, with _MarkBareShiftHandled() latching
presses the key-event sink already toggled so hosts that do deliver the
release never toggle twice. Install that hook for every host instead of
only mintty.exe, and rename its members off the mintty-specific naming.
Also log the three bare-Shift release paths under [issue47] -- key-up was
previously uninstrumented, which is why the Word diagnostic log could not
distinguish "release never arrived" from "toggle queued but dropped".
* chore: settle line endings in .gitattributes
Nothing declared line endings, so each file kept whatever the contributor's core.autocrlf produced when it was added: 382 text files ended up LF and 50 CRLF. The CRLF ones are the hazard. Editing one on macOS or Linux writes LF lines into it, and clang-format rejects a file that mixes both -- which is how `quality / Workflow validation` went red on KeyStateCategory.h, where three lines out of a CRLF file had arrived as LF. #309 converted those three lines back and left the other 49 files loaded.
The repository now declares LF and the working trees are renormalized to match, so the failure mode is gone rather than deferred to whichever of those files is edited next.
The four binary types are declared rather than left to git's heuristic, which decides from the first 8000 bytes alone. user_dict.dat is caught by that sample today only because the 40-byte seed is almost entirely NUL.
Verified: the 56 renormalized files are byte-identical to their old contents once CR is stripped, no text file in the index still carries CRLF, and scripts/format.sh --check passes.
* chore: skip the line-ending renormalization in blame
The commit before this one rewrote every line of 56 files without changing a character of them, which is what this file exists to hide.
Two independent failures kept the v0.6.5 installer from being built after
the pipe probe was fixed.
The package job died in "Verify the release commit is on main" with git
exiting 128 on dubious ownership: runner instance 2 leaves _work owned by
SYSTEM while the job runs as `runner`. actions/checkout is unaffected, so
the first plain git call is what fails, and only jobs scheduled onto that
instance see it. Mark the workspace as safe in the three self-hosted jobs.
Neither vcpkg manifest carried a builtin-baseline, so vcpkg resolved ports
against the registry at HEAD and every build depended on a live fetch of
microsoft/vcpkg. When that fetch degraded, the ports simply vanished:
Fetching registry information from https://github.com/microsoft/vcpkg (HEAD)...
fmt does not exist
utfcpp does not exist
Build TSF x64 failed this way twice and passed twice on the same runner,
which is what an unpinned registry looks like. Pin both manifests to
04a9d8e5, the commit vcpkg-lock.json already records as resolved.
Two host-timing paths could hide or displace the candidate window between
ordinary keystrokes. Both get much more likely while the machine is busy,
which is when the flicker is reported.
TS_E_NOLAYOUT from ITfContextView::GetTextExt means "not laid out yet, ask
again", but _GetTextExt reported it as a valid measurement at INVALID_Y.
Server then parks the host off-screen for that frame and the next successful
measurement pulls it back. Chromium-based hosts answer TS_E_NOLAYOUT
whenever their renderer is behind, so under load this repeats per keystroke.
Keep the anchor the session already has instead, once it has one; composition
start still falls back to the sentinel rather than reusing a stale anchor
from another document.
TF_LC_DESTROY was treated as candidate-session teardown for every process
except a hardcoded Telegram.exe. Hosts that transiently destroy and recreate
their context view mid-composition therefore got a HideCandidate between
keystrokes, which clears the live composition on the Server side: the window
blinks and the pending candidate can no longer be committed. Gate on a live
composition instead of a process name. Commit, cancel, focus loss, and
presenter cleanup all end the composition first, so they still tear the
session down through their own paths.
Elevated setup left high-IL config.toml that medium-IL Settings never
rewrote on machines hit by the ANSI path bug. Make the data directory
writable, migrate ACP-mangled leftovers, rebuild unparseable files, and
keep remaining asset paths on wchar_t.
Fixes#242
std::filesystem::path(std::string) and path::string() convert through the
system ANSI code page, so a non-ASCII (e.g. Chinese) user profile path on a
non-UTF-8 (GBK 936) system was corrupted and every config read/write failed
with "设置保存失败"; on a code page that cannot represent the characters
path::string() threw, crashing the Server at startup in a watchdog restart loop.
Never route config paths through the ANSI narrow encoding:
- build g_config_path and the legacy voice path from the wide accessors;
- read config by opening the wide path + toml::parse(text) instead of
toml::parse_file(path.string()) (LoadImeConfig, MigrateLegacyVoiceInputConfig);
- build and open the TSF shared config path as a wide std::filesystem::path.
Add a regression test that round-trips config read/write under a Chinese
profile path.
fanlusky