Skip to content

macOS: IME text committed after the marked text is cleared isn't lost (winit#4650) - #676

Merged
echelon merged 2 commits into
storytold:mainfrom
cyfung1031:fix/winit-4650-macos-ime-commit
Oct 10, 2026
Merged

echelon merged 2 commits into
storytold:mainfrom
cyfung1031:fix/winit-4650-macos-ime-commit

Conversation

@cyfung1031

@cyfung1031 cyfung1031 commented Oct 10, 2026 •

Copy link
Copy Markdown
Contributor

Summary

On macOS, text committed through IME is lost in PdfCraft's text fields when the input method clears its marked text before inserting. This happens with:

  • Apple Korean, for the key that ends a syllable: typing 한 then 5 (or ;) gives 한 without the 5.
  • Input methods that end a composition with unmarkText before insertText (incorrectly handle insertText (macOS) rust-windowing/winit#3925).
  • A commit made outside a key press (a candidate clicked with the mouse, dictation): the next key, such as Backspace, is swallowed.

Port of rust-windowing/winit#4650 (reviewed head 565e504d43), the 0.30 backport of winit#4651, which is merged on winit master (5ea1b78d29). It also includes the part of winit#4748 (reviewed head 41e556b432) that fixes #4651's out-of-key-press case. The other macOS PRs in this batch build on this one.

PdfCraft impact

IME input in every egui text field on macOS: Search, form fields, the JavaScript console, comments, the command palette.

Root cause

vendor/winit/src/platform_impl/macos/view.rs:

  • insertText: reports Ime::Commit only while marked text is left (hasMarkedText()). Apple Korean calls setMarkedText("한"), insertText("한"), setMarkedText(""), insertText("5"), so the 5 arrives with no marked text: it is neither committed nor sent as a key, because keyDown: treats the key as consumed by IME.
  • After any commit the view is left Committed. Outside a key press, no keyDown: follows to clear the marked text and reset the state, so the next key is swallowed.

Change

  • pending_commit: set by setMarkedText, cleared by a commit and when IME is disallowed. insertText commits when a composition was started, rather than when marked text is left. Other insertText calls (ordinary typing) still reach the app as KeyboardInput, which sign: validate signature dicts and post-signing changes against signed revision (#3998) #4576's broader approach would have broken.
  • in_key_down: a commit outside keyDown: clears the marked text and returns to the ground state instead of swallowing the next key (from M7: an added text box or picture dragged onto another page moves there (#4655) #4748).
  • validAttributesForMarkedText lists the underline and marked-clause attributes; some input methods (落格输入法) don't compose until it is non-empty. Upstream spells the underline key "NSUnderlineStyle", but AppKit's NSUnderlineStyleAttributeName is "NSUnderline" (printed on this Mac), so the AppKit value is used. The marked-clause key matches.

Not included

How the macOS changes were checked

A scratch winit app (not committed) built against the vendored winit, first unpatched, then with each patch. It picks an input source on its own input context (NSTextInputContext.selectedKeyboardInputSource), posts key events made from CGEventCreateKeyboardEvent to itself, so AppKit routes them through keyDown: and interpretKeyEvents, and the real input method composes them. It prints winit's Ime and KeyboardInput events, and AppKit's NSTextInputClient calls from winit's own trace spans. Machine: macOS 14.7.1 (23H222), Apple Korean 2-Set, Japanese (Romaji), Cangjie, ABC. Afterwards the input source was set back to ABC.

The scenarios are: Korean syllables followed by Space, a digit, ;, Return and Backspace; Space with nothing composed; Japanese 日本 and あ; Cangjie 日; ABC typing; a held key with auto-repeat; IME disallowed and allowed again mid-composition; commits made outside a key press; and unmarkText followed by insertText. How egui then uses these events was read from egui-winit 0.36.2 (KeyboardInput text becomes Event::Text; Ime::Commit inserts into the focused TextEdit).

Before → after (only these events changed; every other scenario gave identical events):

Scenario Before After
Korean 한 then 5 Commit("한"), the 5 lost Commit("한"), Commit("5")
Korean 한 then ; Commit("한"), the ; lost Commit("한"), Commit(";")
setMarkedText, unmarkText, insertText("final") nothing committed Commit("final")
Japanese か, commit outside a key press, then Backspace Backspace swallowed KeyboardInput(Backspace)
ABC, commit outside a key press, then b b committed through the stale marked text KeyboardInput("b")

Unchanged: ABC typing (KeyboardInput with text, no Ime::Commit), held keys, Japanese 日本/あ, Cangjie 日, Korean syllables with Space, Return and Backspace.

apps/pdfcraft/tests/winit_macos_ime_commit.rs fails if a re-vendored winit loses the patch (each assertion fails on main).

Automated checks

  • The vendored winit built on its own for aarch64-apple-darwin, x86_64-pc-windows-msvc and x86_64-unknown-linux-musl (cargo clippy --lib, lints capped to warnings): the warnings are the same as on main for all three (the one existing clippy::iter_kv_map in linux/x11/event_processor.rs).
  • cargo check --locked -p winit inside the PdfCraft workspace, with PdfCraft's own Cargo.lock, on macOS and on the tree with all nine fixes merged: clean.
  • The new test was compiled and run against this branch (passes) and against main (fails); rustfmt --check and clippy -D warnings on it are clean.
  • Not run locally: the full cargo test --workspace and workspace clippy (the machine is short of disk space for a full build). CI runs them on Linux and Windows.

Risk

Kept as upstream: after a composition cancelled to empty (setMarkedText("")), the next character inserted by the input method arrives as Ime::Commit rather than as KeyboardInput; egui inserts the same text either way. Clearing pending_commit there instead would drop the commit of input methods that clear the preedit right before inserting (the bug fixed here).

Revert and removal

This change is independent of the other winit PRs and can be reverted on its own; three PRs in this batch build on it. Drop it, together with the test and README item (6), once PdfCraft uses a winit release that includes rust-windowing/winit#4651/#4650 and the out-of-key-press part of #4748.

Merging alongside the other winit PRs

This is one of nine fixes to vendor/winit sent together: #672 (Wayland IME after a capability change), #673 (Windows Korean IME caret), #674 (Windows maximized start-up), #675 (X11 input context focus), #676 (macOS IME commit state), #677 (macOS stale composition), and, building on #676, #678 (macOS emoji picker), #679 (macOS Korean double space), #680 (macOS first Korean key). Each adds one numbered item to the single winit row in vendor/README.md, so they conflict with each other on that one line, and nowhere else (each test is in its own file). The item numbers (2)–(10) are fixed in advance: to resolve, keep every item in numeric order. All nine were merged locally in one tree (only that line conflicted); the combined vendored winit builds for macOS, Windows and Linux with the same clippy output as main, and all nine tests pass on it.

🤖 Generated with Claude Code

… (winit#4650)

winit 0.30.13 reports insertText as Ime::Commit only while marked text is
left, and after any commit it swallows the next key. Real paths where that
loses text:

- input methods that end a composition with unmarkText, then insertText
  (rust-windowing/winit#3925): the committed text was dropped;
- Apple Korean 2-Set, for the key typed after a syllable: on 한 then 5 it
  calls setMarkedText("한"), insertText("한"), setMarkedText(""),
  insertText("5"), so the 5 (or ;, or any key that ends the syllable) was
  dropped;
- a commit made outside a key press (a candidate clicked with the mouse,
  dictation): no key follows to reset the state, so the next key (for
  example Backspace) was swallowed.

vendor/winit now commits when a composition was started (pending_commit,
set in setMarkedText, cleared by the commit and when IME is disallowed);
other insertText calls still reach the app as KeyboardInput. A commit
outside keyDown clears the marked text and returns to the ground state.
validAttributesForMarkedText lists the underline and marked-clause
attributes, which some input methods need before they compose.

This is rust-windowing/winit#4650, the 0.30 backport of storytold#4651 (merged on
winit master), plus the part of storytold#4748 that fixes storytold#4651's out-of-key-press
case (storytold#4748's other parts are not included). One correction: upstream
spells the underline attribute "NSUnderlineStyle", but AppKit's
NSUnderlineStyleAttributeName is "NSUnderline" (printed on macOS 14), so
the AppKit value is used.

Checked with the real input methods on macOS 14.7.1 (a scratch winit app fed
key events through AppKit): with the patch 한+5 and 한+; commit the 5 and ;,
unmarkText-then-insertText commits, and Backspace after a commit outside a
key press is delivered; ABC typing, held keys, Japanese (日本, あ), Cangjie
(日) and Korean composition give the same events as before. A test fails if
a re-vendored winit loses the patch.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@echelon
echelon merged commit ec6c904 into storytold:main Oct 10, 2026
5 checks passed
@echelon

echelon commented Oct 10, 2026

Copy link
Copy Markdown
Member

Merged, thank you! Committing when a composition was started rather than while marked text remains (winit#4650/#4651) stops input methods that unmarkText first from losing text, and the real-IME testing on macOS 14.7.1 is excellent. Patch (6). CI has no macOS job, so I cross-compiled vendor/winit for aarch64-apple-darwin before landing each macOS PR.

If you haven't already, come join us on Discord: https://discord.gg/artcraft

krichelj pushed a commit to krichelj/pdfcraft that referenced this pull request Oct 11, 2026
… (winit#4650) (storytold#676)

winit 0.30.13 reports insertText as Ime::Commit only while marked text is
left, and after any commit it swallows the next key. Real paths where that
loses text:

- input methods that end a composition with unmarkText, then insertText
  (rust-windowing/winit#3925): the committed text was dropped;
- Apple Korean 2-Set, for the key typed after a syllable: on 한 then 5 it
  calls setMarkedText("한"), insertText("한"), setMarkedText(""),
  insertText("5"), so the 5 (or ;, or any key that ends the syllable) was
  dropped;
- a commit made outside a key press (a candidate clicked with the mouse,
  dictation): no key follows to reset the state, so the next key (for
  example Backspace) was swallowed.

vendor/winit now commits when a composition was started (pending_commit,
set in setMarkedText, cleared by the commit and when IME is disallowed);
other insertText calls still reach the app as KeyboardInput. A commit
outside keyDown clears the marked text and returns to the ground state.
validAttributesForMarkedText lists the underline and marked-clause
attributes, which some input methods need before they compose.

This is rust-windowing/winit#4650, the 0.30 backport of storytold#4651 (merged on
winit master), plus the part of storytold#4748 that fixes storytold#4651's out-of-key-press
case (storytold#4748's other parts are not included). One correction: upstream
spells the underline attribute "NSUnderlineStyle", but AppKit's
NSUnderlineStyleAttributeName is "NSUnderline" (printed on macOS 14), so
the AppKit value is used.

Checked with the real input methods on macOS 14.7.1 (a scratch winit app fed
key events through AppKit): with the patch 한+5 and 한+; commit the 5 and ;,
unmarkText-then-insertText commits, and Backspace after a commit outside a
key press is delivered; ABC typing, held keys, Japanese (日本, あ), Cangjie
(日) and Korean composition give the same events as before. A test fails if
a re-vendored winit loses the patch.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
Co-authored-by: Brandon Thomas <echelon@gmail.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants