Automatic per-merge builds shared one list with deliberately cut releases, and the previous change only made them identifiable rather than separate: both were still prereleases, which is the thing the issue objects to. Labelling was the wrong answer to it.
GitHub has no channel concept, but it has two published states, and there are exactly two kinds of release here. A push-triggered build is now a prerelease; a workflow_dispatch build is an ordinary release and the newest one carries the Latest badge. The Releases page shows two lists instead of one list with two kinds of title in it, and /releases/latest resolves for the first time.
The prerelease flag therefore means "automatic and uncurated" rather than "beta". The product being 内测 is stated in the release notes and in docs/installation.md, which is where a claim about stability belongs; the flag cannot carry that meaning while it is also the only thing separating the channels. docs/installation.md now tells readers to take the Latest badge, describes both channels, and says explicitly that being in the release channel means somebody picked it, not that it is stable.
Already published releases are not touched, so the Latest badge stays empty until the first dispatch after this lands.
Refs MSIME-Windows#167
GitHub has three release states and no notion of a channel, so an automatic per-merge build and a release somebody deliberately cut both land in the same Pre-release list that docs/installation.md presents to users as the 内测版 channel. Before every releasable merge produced a release that did not matter; now it does.
Releases published from a push get 「（自动构建）」 in the title and a note saying they were not curated. Releases published from a workflow_dispatch keep the bare tag as their title, which is what release-please already set. docs/installation.md tells readers which is which and why they share one list.
This is deliberately a labelling change and not a second release track. Giving automatic builds their own tag namespace would need a parallel versioning mechanism alongside the one release-please owns, and stacking that on a pipeline that is currently failing and shipping unsigned artifacts is the wrong order.
Refs MSIME-Windows#167
* 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.
A push to main carrying a fix or a feat now ends with a published installer, with no human merging the release pull request and no manual workflow_dispatch in between.
release-please only cuts a release for a release pull request that was already merged when it ran, so the prepare job now invokes it twice: once to refresh the pull request against the commits this push brought, and once after land-release-pr.sh has merged it, to turn that merge into the draft the rest of the run builds and publishes. Merging before the first invocation would instead release everything except the push that triggered the run.
land-release-pr.sh is check-release-pr.sh plus the merge it deliberately used to skip. It merges with GITHUB_TOKEN, whose pushes do not start workflow runs, so the merge cannot produce a second Release run queued behind this one on the concurrency group; the release is created and built inside the run that merged it. It merges the exact commit CI proved via --match-head-commit, then waits for that merge commit to be reachable from main so the second release-please invocation does not quietly find nothing to release.
resolve-auto-release.sh picks which of the two invocations reported the release to build, preferring the second when both did, and applies the invariants validate-draft-release.sh gives the manual path: the release is tagged at a full commit SHA, and the tag agrees with the version.
workflow_dispatch stays as a repair path for a run that failed after creating its draft.
This spends one of the signing certificate's limited signatures on every releasable merge, which is what the previous manual design existed to avoid. docs/product-release.md records the trade, including why the queueing this can produce is not answered by cancel-in-progress.
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 probe took its include directory from $ServerRoot/MetasequoiaImeEngine/contracts, which is where the engine sat when it was a submodule of Server. It is a submodule of the repository now, so the path does not move with -ServerRoot. A missing include directory is not a configure error, so this surfaced minutes later as C1083 on a header nobody had moved; the script now checks and says so before building.
The local packaging rehearsals had the same stale assumption, looking for MetasequoiaImeTsf, MetasequoiaImeServer and MetasequoiaImeUiHtml beside the installer. Worse, they skipped the build when a neighbour was absent, which in this layout means every build is skipped and whatever binaries are lying around get packaged into an installer you then run on your own machine. The three components are directories of this repository, so a missing one is now a hard failure, and Prepare-PackageFiles.ps1 is passed the directory names it needs.
Both were missed because the sweep that moved the submodules read its own output through head, so a truncated list looked like a complete one. Re-checked by resolving every relative path literal in every tracked file with no filtering.
check-dictionary-contract.py compares scripts/dictionary_product.py against the engine's own copy, so the job needs the submodule's contents and not just its gitlink. The reusable workflow this job replaced checked out submodules; the rewrite dropped that.
The script reported a missing file as "differs from the pinned Engine contract", which describes a drift that had not happened and sends the reader looking for a content change. It now says the file is missing and how to get it.
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.
Preparation for folding Server, UI, UiHtml, Installer, ImeLog and TsfEditControl into this repository. The Windows tip's own sources, build files and local build scripts move under windows/ so every component sits at the same level. Product-level files stay at the root: the lock, the release manifest, the shared CI scripts, the integration tests, the docs and the Engine submodule.
Every other piece of shell in these workflows lives in scripts/ci and the workflow only wires it up. The language probe went in inline, so it did not: twenty lines of PowerShell building an .iss heredoc inside a YAML block scalar, which is the least reviewable and least runnable place to put it.
Same check, now in check-inno-language.ps1, and runnable by hand on a Windows machine. It falls back to the system temp directory when RUNNER_TEMP is unset so that running it outside CI works.
Every release build since the packaging pipeline landed has failed at "Compile the installer", which is why the last eight versions are still unpublished drafts:
Error on line 63 in msime_setup.iss: Couldn't open include file
"c:\program files (x86)\inno setup 6\Languages\ChineseSimplified.isl"
install-inno-language.ps1 derived the destination from `Get-Command ISCC.exe`. On the windows-2025 image Inno Setup comes from Chocolatey, so that resolves to the shim at C:\ProgramData\Chocolatey\bin\ISCC.exe and the file was written to C:\ProgramData\Chocolatey\bin\Languages\. ISPP resolves compiler: against the real installation. The step printed "Installed ... (21516 bytes)" and exited 0, and the build failed two steps later pointing at a path nothing had written to, which is why this survived eight releases.
The installation is now located by a file only it has -- ISPPBuiltins.iss or Default.isl, neither of which sits beside a shim -- checking the uninstall registry key, the two Program Files locations and any ISCC.exe on PATH, and failing with the list it tried rather than falling back to a guess.
ci.yml now compiles a throwaway script whose only content is a [Languages] entry pointing at compiler:Languages\ChineseSimplified.isl. Asserting the file exists at a path this script computed would only restate the script's own belief; compiling is what proves ISCC agrees. The URL and digest pin moved into the script so that check does not need a second copy of them.
The signing certificate allows a limited number of signatures, so a release costs something finite. It should not be spent by merging a pull request.
Until now a push to main was enough to spend one: release-please opened the release pull request, merge-release-pr.sh merged it, the merge commit came back through this workflow as release_created, and the build jobs signed two binaries and an installer and published the result. Nobody had to ask for any of that.
should_publish is now true only for workflow_dispatch, and the package job -- the only one that signs -- states that condition itself rather than inheriting it from four other jobs whose conditions a later edit could quietly change.
Auto-merging the release pull request goes with it, since that is the mechanism that turned a merge into a release, and leaving it would also mint a draft release and a tag on every merge to main that nobody intends to build. What remains automatic is getting a passing check onto that pull request, which is not optional: GITHUB_TOKEN cannot start a workflow run, so without the dispatch in check-release-pr.sh the release pull request has no checks at all and required_status_checks (#121) leaves it permanently unmergeable.
Releasing is now two deliberate steps. Merge the release pull request, which creates the draft release and its tag and builds nothing. Then run the Release workflow against that tag. Dispatching the same tag twice is already refused by validate-draft-release.sh, because a published release is no longer a draft, so a retry cannot quietly spend a second signature.
The lock pins immutable commits, but nothing checked that anybody merged them. Today it pins server 40d85a4 and engine 9a4b0c6, both of which exist only on architecture/product-contracts branches, and a release built from this commit would have shipped and attested to a Server that never reached main. The release job's existing provenance check covers only the TSF commit being released, not the inputs the lock names.
verify-published compares each locked commit against its own repository's default branch and accepts only behind or identical, the two states that mean it is an ancestor. ahead and diverged both mean the commit sits on something nobody merged.
It runs in the release path and deliberately not in product-ci.yml. A pull request legitimately locks branch commits while one change lands across several repositories at once, which is exactly the situation this branch is in; enforcing reachability in pull request CI would deadlock the landing it exists to protect. At release time the judgement inverts, because an input that never reached its default branch is an input nobody reviewed.
The step ran only when release-please reported a pull request as created, which is once. It rewrites
that branch on every later push to main, and a release pull request that could not be merged the
first time is never retried -- main advancing during the build is the normal way that happens, and
the script deliberately leaves the pull request open in that case, so the two together meant it
would sit there forever.
Finding the branch through git/matching-refs and then its open pull request covers both: a rewritten
head gets a fresh build, and a pull request left open gets picked up on the next push. Taken from
MSIME-Linux#38, which caught the same mistake in the equivalent change there.
MSIME-Windows#121 records that CI cannot be made a required check here, because release-please's
pull request never gets one: GITHUB_TOKEN may not start workflows, so the pull_request run is
created in action_required and waits for a human forever.
I previously wrote on that issue that the only way out was a GitHub App or PAT. That was wrong.
workflow_dispatch is the exception GitHub allows GITHUB_TOKEN to fire, and MSIME-Apple's release
automation has been relying on it all along. Checked its release branch: workflow_dispatch runs
exist there and succeed, alongside pull_request runs that do not.
Also checked the two alternatives that looked cheaper, and neither works:
- A push trigger on release-please--** produces nothing. Every run on that branch in MSIME-Linux
is a pull_request event; a GITHUB_TOKEN push creates no run at all. MSIME-Linux#31 merged on
the belief that it does.
- The approval policy is not the cause. All six repositories are already on
first_time_contributors_new_to_github, the loosest setting, and bot-triggered runs are still
gated. Comparing runs, the gate tracks triggering_actor exactly: github-actions[bot] gives
action_required, houko gives success.
So merge-release-pr.sh now dispatches ci.yml on the release branch, polls for the run at that exact
head commit, and waits for it with --exit-status before merging. The checks land on the pull
request's head commit, which is what required_status_checks needs.
It also re-reads main and the pull request head afterwards: main can advance during a build, and
merging then would rebase the release commit onto something that was never tested.
ci.yml gains workflow_dispatch so it can be dispatched at all. The paths-ignore stays: it keeps the
unapprovable pull_request run out of the checks list, leaving only the dispatched run's results,
which avoids two check runs with the same name and different outcomes on one commit.
The prepare job gains actions: write, without which gh workflow run cannot dispatch.
release.yml had grown thirteen inline shell and PowerShell blocks, some of them twenty lines, which made the workflow hard to read and impossible to lint or run outside Actions. Each one is now a file under scripts/ci and the workflow only wires them up.
Two things fall out of the move rather than being separate changes. The two signing steps were near-duplicates and are now one sign-binaries.ps1 taking a path list, used for both the packaged binaries and the finished installer. The pinned Inno Setup language file URL and checksum move to the packaging job's env, where they read as configuration instead of being buried mid-script.
The Server job now checks this repository out into automation/ for its two scripts. That checkout has to come after the MSIME-Server one, because checkout cleans its destination and the workspace root would take automation/ with it.
shellcheck is clean on all six shell scripts. The PowerShell is not syntax-checked here for lack of pwsh, but four of the seven are verbatim from steps that already ran green in run 33953797415.
The Windows installer has been produced by hand: build the TSF DLLs and the Server locally, build the settings page, copy dictionaries out of a developer checkout, then run the msime-installer scripts against a self-signed certificate. This automates that end to end, keeping the same artifact name the releases have always carried, MetasequoiaIME_Setup_v<version>.exe.
The workflow follows the shape of MSIME-Apple's release.yml. release-please owns version.txt and opens a release PR; merging it creates a draft release, which the publish path then builds and fills in. A manual dispatch can rebuild an existing draft tag, and validates that the tag matches version.txt at an immutable commit SHA before anything is built.
Four jobs feed the packaging job: the TSF DLLs for x64 and Win32, the Server Release build, and the settings page. msime-installer needs no changes because the packaging job stages each checkout under the historical directory name its scripts already address, so Prepare-PackageFiles.ps1 and Compile-Installer.ps1 run unmodified. windows-2025 ships Inno Setup 6.7.1, above the 6.6 the scripts require.
Dictionaries are not rebuilt here. They are downloaded from a dict-* release of metasequoiaime/MSIME-Dict and checked against its SHA256SUMS.txt, with the Mozc IPAdic / ICOT / Okinawa notice placed where Prepare-PackageFiles.ps1 expects so it reaches the installed product.
Signing follows the pattern MSIME-Apple uses: sign when the certificate secrets are configured, otherwise publish with an -unsigned suffix and say in the release notes that uiAccess will not take effect. Sign-PackageBinaries-Local.ps1 mints a self-signed certificate for local testing and is never invoked here.
release-please cannot drive the VERSIONINFO block in the resource script, which spells the version both as a comma quad and as a dotted string, so scripts/apply_version.py stamps it from version.txt before the release build configures CMake. It also has a --check mode. The committed resource moves off the stale 1.0.0.1 placeholder to the current 0.0.9.
The manifest starts at 0.0.9 rather than the four-part 0.0.9.2 that semver cannot express, so the next release will be 0.0.10, which still sorts after every existing tag.
Build the TSF tip with /MT so it no longer depends on the host's MSVCP140,
skip SetText(empty) in OnCompositionTerminated after normal EndComposition,
and guard SetText/EndComposition with SEH.
Evan Hu