* fix(server): publish candidate pages as immutable snapshots
Global::candidate_ui is a plain global with no lock. The IPC worker rebuilds and destroys its vectors while the UI thread paints from them, and the handoff is an asynchronous PostMessage, so the worker returns to its loop and pops the next task without waiting. A second keystroke, or an async english.db candidate arrival, could free the page_views buffer and its strings while FillItemsFromUi held a reference into it. ShowItemContextMenu reads the same vectors from a mouse callback with no message handoff at all.
The worker now copies each finished page into an immutable snapshot and publishes it at the single point where the rebuild completes; every UI-thread reader loads that snapshot once and reads only its own copy. PostMessage is deliberately left alone: the worker also serves UI-originated tasks, so a synchronous handoff would invite a cross-thread deadlock.
* fix(server): bound the aux and diagnostic pipe reads
Both listeners serve a single pipe instance inline and polled a NOWAIT handle for as long as the process ran, breaking only on a read that succeeded or failed with something other than ERROR_NO_DATA. A client that connected and never wrote held the only instance forever, so the loop never returned to ConnectNamedPipe and no later client was accepted, permanently losing the langbar right-click channel and the terminal-deactivation teardown fallback.
Give both loops the same deadline the hello reads already use and go back to accepting when it expires.
* fix(server): snapshot the voice config and replace config.toml atomically
GetConfiguredVoiceInput handed out a reference to a global that LoadImeConfig rewrites from the IPC worker thread, clearing and refilling two std::map members and around twenty strings. The voice control thread and the low-level keyboard hook, which copies the struct on every keystroke, walked those nodes as they were freed. LoadImeConfig now parses into a local and publishes it in one step under a process-local shared_mutex, and the getter returns a snapshot by value. The file lock is deliberately not reused for readers: it is a named cross-process mutex with a five second wait, and taking it in a keyboard hook would stall typing system-wide.
WriteConfiguredValue also reopened the live config.toml with ios::trunc, so a crash mid-write destroyed the user settings and could leave the file unparseable for good. It now goes through WriteFileTextAtomically, which this file already uses elsewhere. That makes the directory entry swap atomic; it does not flush, so it is not a durability guarantee.
* fix(ipc): restrict the server pipes to the owning account
The pipe DACL granted Everyone FILE_ALL_ACCESS. That bit set includes FILE_CREATE_PIPE_INSTANCE, so any local account could add an instance to a name this server already owned and answer part of the connections in its place. The names are machine-global with no per-user component, and the TSF client opened them with no SQOS, so a squatter could also impersonate the connecting thread.
Build the descriptor from the process owner SID and admit only LocalSystem, that account, and app containers with connect-only rights so UWP text input still works; fail pipe creation outright when the SID cannot be resolved rather than falling back to something permissive. Pass FILE_FLAG_FIRST_PIPE_INSTANCE on the first instance of each name so a squatted name fails loudly, and request SECURITY_IDENTIFICATION on the client side.
Authenticating the peer process after connect is a larger change to the handshake and is not attempted here.
* fix(tsf): guard the null focus paths and unadvise the compartment sink
_CandidateChangeNotification treated ITfThreadMgr::GetFocus and ITfDocumentMgr::GetTop as if they could only fail through a FAILED HRESULT. Both return S_OK with a null out-pointer when nothing has focus or the document stack is empty, and the results were dereferenced unconditionally, so an external UILess host calling Finalize while focus moved away crashed the host application. The sibling _RequestCancelComposition in the same class already checks both.
CLangBarItemButton::CleanUp only unregistered the compartment inside the SUCCEEDED branch of a CoCreateInstance, then deleted the sink regardless. This TIP registers COM-less activation support, where that call does fail, and the sink was freed while TSF still held it advised. Unadvising never needed the thread manager, so it now runs unconditionally before the deletes.
* fix(ui): drop the cached visuals when the scene is replaced
Window caches focused, hovered and captured Visual pointers into the scene tree, and SetScene destroyed that tree without clearing them. Toggling the input mode rebuilds the floating toolbar scene, so the next mouse move dispatched a virtual call through a freed object.
Clear all three before the assignment, and release the capture only when this window actually holds it: an adopted HWND may never have called SetCapture, and capturedVisual_ is set even when it did not.
* fix(settings): scope external skin CSS through the CSS parser
injectScopedSkinCss concatenated a skin toolbar.css into an @scope block as text. Block nesting is textual, so a leading close brace in the skin ended the block early and the rest styled the whole settings window. The host validates only the skin file name, never the body, and candidatePreviewCss interpolated unvalidated colour strings into rule bodies the same way.
Create the scope rule empty and re-insert every rule parsed out of the skin as a child of it, so nothing lands outside however the braces are balanced, and accept a colour only in the plain notations we actually render.
* fix(release): gate the dictionary source commit on its default branch
verify_published iterated only data["repositories"], but the dictionary block pins a source_commit exactly like they do. The manifest check merely proves the release agrees with the lock about that commit, which a branch commit satisfies just as happily, so a dictionary built from an unmerged branch shipped with the provenance the gate exists to refuse.
Run the dictionary through the same loop under its own name, resolving its repository from the lock. A retired dictionary repository that no longer answers is reported rather than failing the release: its published artifacts are immutable history, and being unable to re-check them is not the same as finding them unmerged.
Ports metasequoiaime/MSIME-Windows#153 onto the consolidated repository. That pull request was written against the six-repository layout: it edited product-ci.yml, which no longer exists, and re-pinned Server, UiHtml and Installer, which are directories now.
Three of its pinned commits had not reached this repository at all. The consolidation took each source repository at its main, and Server, UiHtml and Installer each landed two more commits afterwards, so dropping the pins as merely obsolete would have silently dropped the work the pull request exists to consume. The two substantive ones are ported here: Server drops its vendored miniaudio.h and builds against the engine's MetasequoiaIme::VoiceCapture, and the installer collects helpcodes from the engine. UiHtml's was an engine gitlink bump, which one submodule resolves on its own.
The engine is pinned to d0dc0c2b rather than to the b7c799ef that pull request named. b7c799ef predates 403b4a4d, which is what this repository pins today and is the commit that added pagination to the settings dictionary contract, so taking the pull request's pin verbatim would have reverted a contract main already depends on and stripped offset, limit and hasMore back out of the generated web bindings. d0dc0c2b is the engine's main and is the same commit the dictionary release declares as its source, so the code and the data now name one commit.
The helpcodes moved into the engine, so MSIME-HelpCode stops being a product input: the gitlink that pins the contracts pins the tables too. product-lock.json is down to the engine and the dictionary release, and the dictionary itself now ships from MSIME-Engine. Its manifest states which commit built the data and whether that tree was clean, and the lock verifies both, so matching digests can no longer stand in for a source nobody can rebuild from. The retired MSIME-Dict repository stays valid for the one tag that shipped from it and nothing else.
The release job that packages now checks out submodules, because the helpcodes it collects live in one.
The emoji, keyboard and handwriting panels embed the GUI framework's icon through a relative path that was written when msimeui was a submodule under server/. It now lives one level further out, at the repository root.
Found by the resource compiler rather than by the sweep that moved the submodules: that sweep matched CMake, JSON, PowerShell, Python, shell and YAML by extension, and .rc was not on the list. Re-checked by resolving every relative path literal in every tracked file regardless of extension; these three were the only ones left that did not resolve.
The product README was under windows/ after the move, but it was never a component readme: it is the download page, the feature guide and the contribution guide for the whole product, so it goes back to the root and windows/ gets a real component readme.
AGENTS.md at the root now carries the repository map, the component boundaries that used to be cross-repository rules, and the release process, since the scripts that process drives live at the root. The component files keep what is genuinely theirs and say which paths they are relative to. References that named sibling repositories now name sibling directories, and the lock documentation records that Server, the pages, the GUI framework and the installer are no longer inputs to pin.
The product lock existed because five repositories had to be pinned to reproduce one product. Four of them are directories now, so a commit of this repository already pins them and there is nothing left for the lock to say about them. What it still records is what the product consumes from outside: the engine submodule, the helpcodes and the dictionary release.
CI gains a job per component instead of a reusable workflow that checked out locked commits and rebuilt them. The release workflow no longer stages five sibling checkouts under historical names and no longer carries a second sparse checkout for its own scripts; Prepare-PackageFiles.ps1 already took the source directory names as parameters, so it is told the new ones, plus where the product-wide notice now lives.
Two checks the removed cross-repository pins used to provide are replaced in kind: the pages are verified against the engine submodule's web contract in CI rather than by locking UiHtml and Engine together, and the engine bump triage that Server carried moves here and now also says so when a bump leaves the lock behind.
Windows, Server and UiHtml each carried their own gitlink to MSIME-Engine, all three pinned to the same commit, and Server carried a gitlink to MSIME-UI that is now the ui/ component. Git only reads the root .gitmodules, so the nested ones had to go regardless.
The engine and the two third-party dependencies now live once at vendor/MetasequoiaImeEngine, vendor/opencc and vendor/cpp-pinyin. Server reaches them through MSIME_VENDOR_DIR and builds ui/ as an out-of-tree subdirectory; the Windows tip's relative contract includes gained the extra level; the ui-html contract sync and the Server env/table generators follow the engine to its new location.
NormalizeClipboardText caps an entry at kMaxChars with text.resize(). wchar_t is UTF-16 on Windows, so that cap counts code units, and a cut landing between a high and low surrogate leaves a lone high surrogate. The result is not valid UTF-16, and it is what gets serialised into the history store and read back.
It takes a non-BMP character — an emoji, or a supplementary-plane Han character, both of which this IME can produce — sitting exactly at the boundary. The entry then carries an unpaired surrogate for as long as it stays in the history.
The cut now steps back one unit when it would split a pair, dropping the incomplete character instead of half of it. Entries that do not end on a surrogate are unaffected.
MSIME-Linux does not have this: its normalize() truncates with g_utf8_offset_to_pointer, which is codepoint aware. The two implementations otherwise agree exactly, including kMaxItems, kMaxChars and the promote-to-front policy.
src/d2d/candidate_window_d2d.{cpp,h} is 528 lines that no target compiles: it appears in no add_executable source list in CMakeLists.txt, and nothing outside the directory names candidate_window_d2d or CandidateWindowD2D. The d2d1 in CMakeLists.txt is the Windows system library, not this directory. It was last modified on 2025-06-25.
It is an earlier candidate window, superseded twice over: the shipped candidate window is the WebView2 page from MSIME-UiHtml, and the native GUI work now lives in the msimeui framework. Keeping a third, unbuilt implementation in the tree means every candidate window change invites the question of whether this one needs the same fix, and the answer is always no because it never runs.
The history keeps it if the Direct2D drawing is ever wanted as a reference.
The engine pin moved 55839c8 -> 127019f in the previous commit, and that range contains 86c5fa7 "fix(quanpin): enumerate alternative segmentations for four-syllable input". QuanpinDictionaryOnlyTriesMultipleSegmentationsForAtMostThreeSyllables asserted the behaviour that fix deliberately changed, so it failed here and, through the product lock, in MSIME-Windows' locked-Server integration job as well.
The engine is the authoritative layer: quanpin/quanpin_dictionary.cpp declares kMaxSyllablesForMultipleSegmentations = 4 and gates enumeration on segments.size() <= that. Four syllables now enumerate alternatives; five still do not. The test says exactly that, with a tbl_6_x fixture row so the five-syllable case would have something to find if the bound ever moved silently.
Evan
