fix(core): fix keydown/up handling for LDML keyboards #️⃣ by ermshiperete · Pull Request #15609 · keymanapp/keyman

ermshiperete · 2026-02-20T17:18:13Z

For LDML keyboards this change fixes the emit_keystroke flag so that it always has the same value for the KeyDown and the KeyUp event. This fixes some problems with stuck keys. Previously we would set emit_keystroke=TRUE on KeyDown but emit_keystroke=FALSE on KeyUp for frame keys. This caused Linux to never see the KeyUp event, resulting in a stuck key.

Also add unit tests that verifies that the actions that we get after calling km_core_process_event are what we expect.

Follows: #15656
Fixes: #15569
Fixes: #15550

User Testing

Test on Linux

SUITE_BASICTEST

GROUP_X11 Test with Gnome Shell and X11
GROUP_WAYLAND Test with Gnome Shell and Wayland
TEST_ONEKEY_COMPLIANT:

Install ldmlonekey.kmp.zip keyboard (rename to ldmlonekey.kmp). This defines a keyboard where only 1 will output anything.

Open the text editor or gedit and type different keys with and without Shift, Control, and Alt/RAlt. Only the 1 key (without any modifier) should produce an output, all other (regular) keys shouldn't output anything. Enter should output a new line, and Backspace should work as well. The usual shortcut keys with Ctrl should also work: Ctrl-C copy, Ctrl-V paste, Ctrl-Backspace delete word, Ctrl-a select all text.
TEST_ONEKEY_NONCOMPLIANT: Try the same in Chromium, e.g. in the textbox of https://keymanweb.com/.

SUITE_BUGSFIXED

TEST_15569: verify that bug(linux): 'stuck' Enter key with ldml keyboards on Wayland #15569 is fixed
TEST_15550: verify that bug(linux): ldml keyboard desynchronizes and can crash when bksp is used in non-compliant app with decomposed cluster on Wayland #15550 is fixed
TEST_ADHOC: using the Khmer (or some other) LDML keyboard, verify that typing works as expected using Wayland.

keymanapp-test-bot · 2026-02-20T17:18:27Z

User Test Results

Test specification and instructions

✅ SUITE_BASICTEST:

4 tests in 2 groups PASSED

✅ GROUP_X11: Test with Gnome Shell and X11
2 tests PASSED
- ✅ TEST_ONEKEY_COMPLIANT (PASSED) (notes)
- ✅ TEST_ONEKEY_NONCOMPLIANT (PASSED) (notes)
✅ GROUP_WAYLAND: Test with Gnome Shell and Wayland
2 tests PASSED
- ✅ TEST_ONEKEY_COMPLIANT (PASSED) (notes)
- ✅ TEST_ONEKEY_NONCOMPLIANT (PASSED) (notes)

🟥 SUITE_BUGSFIXED:

✅ TEST_15569 (PASSED) (notes)
🟥 TEST_15550 (FAILED) (notes)
✅ TEST_ADHOC (PASSED) (notes)

Retesting Template

Test-bot: retest SUITE_BUGSFIXED  TEST_15550

Test Artifacts

Developer
Keyboards
- Test Keyboards
macOS
- Keyman for macOS
- Keyman for macOS (old PRs)
Web
- KeymanWeb Test Home
Windows

This documents the state of `km_core_actions.emit_keystroke` for different keys pressed. Also some cleanup in other docs. This documents the state after merging #15609 (for ldml keyboards) and NN (for kmn keyboards). Follows: #15609 Build-bot: skip Test-bot: skip

core/tests/unit/km_core_process_event.tests.cpp

rc-swag · 2026-02-26T04:51:32Z

core/src/ldml/ldml_processor.cpp

+
+  if (!ldml_state.is_key_down()) {
+    if (!ldml_state.get_state()->backspace_handled_internally()) {
+      ldml_state.emit_passthrough_keystroke();


I was trying to imagine if there was a scenario where we got a key-up for a Backspace without first getting a key-down. This would be considered bad input anyway. Passing the backspace to the app would probably be acceptable.

I'm not sure I understand why this is needed only for LDML and not for KMN keyboards.

We will need it for KMN keyboards as well. However, I didn't get to that yet. I thought I'd fix LDML keyboards first in this PR since we get crashes with LDML, and then in the next PR tackle KMN keyboards.

core/tests/unit/km_core_process_event.tests.cpp

Meng-Heng · 2026-02-26T08:09:44Z

Test Results

SUITE_BASICTEST:

Test Specs

Ubuntu Noble 24.04.4 LTS
GNOME Version 46
Keyman v19.0.204-alpha 15609-6795

GROUP_X11: Test with Gnome Shell and X11

TEST_ONEKEY_COMPLIANT (PASSED):

Launch Text Editor
Result: Work as expected
- Key 1 produce 1 on the default layer
- Able to use SHIFT, CTRL, TAB, BACKSPACE, ENTER, and ARROW KEYS
- Typing with SHIFT, CAPS LOCK produces the uppercase alphabet letters, symbols, and signs.
- CTRL shortcut keys work

TEST_ONEKEY_NONCOMPLIANT (PASSED):

Launch Chrome -> keymanweb.com
Type in textarea
Result: Work as expected (details the same as above).

GROUP_WAYLAND: Test with Gnome Shell and Wayland

Test Prerequisites

Enable Wayland with sudo nano /etc/gdm3/custom.conf

TEST_ONEKEY_COMPLIANT (PASSED):

Launch Text Editor
Result: Work as expected
- 1 produce on the default layer
- Able to use SHIFT, CTRL, TAB, BACKSPACE, ENTER, and ARROW KEYS
- Typing with SHIFT, CAPS LOCK produce the uppercase alphabet letters
- CTRL shortcut keys work

TEST_ONEKEY_NONCOMPLIANT (PASSED):

Launch Chrome -> keymanweb.com
Type in textarea
Result: Work as expected (details above).

SUITE_BUGSFIXED:

Test Specs

Ubuntu Noble 24.04.4 LTS
GNOME Version 46
Keyman v19.0.204-alpha 15609-6795
Non-compliant app: Chrome -> Keymanweb.com
Compliant app: Text Editor

TEST_15569 (PASSED):

With the LDMLOneKey and Khmer Angkor keyboard
In the Wayland Windowing system, the Enter key stops producing an infinite number of newlines in non-compliant and compliant apps.

TEST_15550 (FAILED):

Non-compliant app --> keymanweb.com --> click in textarea to type
Use the Test_BKSP keyboard
Type abc d/, hit bksp
Verified: Backspace does not stick.
Result: A backspace deletes the character with its combining acute.

TEST_ADHOC (PASSED):

With Khmer Angkor and Test BKSP keyboard
Typing work as expected in compliant and non-compliant apps.

rc-swag

lgtm

mcdurdin

I will need to continue review on this next week but had some questions to start with.

mcdurdin · 2026-02-27T12:56:47Z

core/src/ldml/ldml_processor.cpp

+
+  if (!ldml_state.is_key_down()) {
+    if (!ldml_state.get_state()->backspace_handled_internally()) {
+      ldml_state.emit_passthrough_keystroke();


I'm not sure I understand why this is needed only for LDML and not for KMN keyboards.

mcdurdin · 2026-02-27T12:57:38Z

core/src/ldml/ldml_processor.cpp

+  // Look up the key
+  bool found = false;
+  const std::u16string key_str = keys.lookup(ldml_state.get_vk(), ldml_state.get_modifier_state(), found);
+
+  if (!found) {
+    ldml_state.emit_passthrough_keystroke();
+  }


Does this match the logic in key_down?

yes, that part is exactly the same.

mcdurdin · 2026-02-27T12:58:28Z

core/src/ldml/ldml_processor.cpp

-  this->vk             = v;
-  this->modifier_state = m;
-  this->is_key_down    = i;
-  this->event_flags    = e;
+  this->_vk             = v;
+  this->_modifier_state = m;
+  this->_is_key_down    = i;
+  this->_event_flags    = e;


renaming like this should be in a separate refactor PR please -- it makes it harder to follow the logical changes here and introduces complexity for patching into other versions

You're right, sorry. Do you want me to still split this PR?

Yes please can you split the refactor/rename into a separate PR?

Sorry, doing that required a force-push for this PR.

mcdurdin · 2026-02-27T13:00:02Z

core/tests/unit/km_core_keyboard_api.tests.cpp

      this->keyboard = nullptr;
    }
-    }
+  }


better to keep whitespace-only changes out of a change like this. OK if the whitespace is in an affected area but otherwise it doesn't belong and unnecessarily expands the scope of patches.

mcdurdin · 2026-02-27T13:00:14Z

core/tests/unit/kmx/meson.build

  input: kbd_src,
  command: kmc_cmd + ['build', '--debug', '--no-compiler-version', '@INPUT@', '--out-file', kbd_obj]
 )
-  test('imx_list', imx_e,  depends: kbd_log, args: [kbd_obj] )


ditto on whitespace in this file

core/tests/unit/meson.build

mcdurdin · 2026-02-27T13:01:10Z

core/src/state.hpp

+    // This is used to track whether the backspace key was handled internally
+    // during the keydown event. This is needed so that we can return the same
+    // value from the keyup event. Only used when processing KM_CORE_VKEY_BKSP.


This should be a javadoc style comment. I still don't really understand why it needs to be handled separately in Core.

Backspace is the only key that we sometimes handle internally (if we have enough context) and sometimes not. By the time we get the keyup event the context already got updated and so we have no way of knowing whether or not the keydown handled it internally. Therefore this flag.

Note that things do work without knowing because for backspace we always returned the right value for emit_keystroke, but it will cause the emit_keystroke values for keydown and keyup to be out of sync.

Gotcha thanks! Perhaps we can capture some of that detail in the comment?

Comments addressed. Do you still want me to split out the whitespace and refactoring changes to a separate PR?

For LDML keyboards this change fixes the `emit_key` flag so that it has the same value for the KeyDown and the KeyUp event. This fixes some problems with stuck keys. Previously we would set `emit_key=TRUE` on KeyDown but `emit_key=FALSE` on KeyUp for frame keys. This caused Linux to never see the KeyUp event, resulting in a stuck key. Also add unit tests that verifies that the actions that we get after calling `km_core_process_event` are what we expect. Fixes: #15569 Fixes: #15550

Co-authored-by: Marc Durdin <marc@durdin.net>

docs(core): add keyhandling doc This documents the state of `km_core_actions.emit_keystroke` for different keys pressed. Also some cleanup in other docs. This documents the state after merging #15609 (for ldml keyboards) and NN (for kmn keyboards). Follows: #15609

github-project-automation bot added this to Keyman Feb 20, 2026

github-project-automation bot moved this to Todo in Keyman Feb 20, 2026

github-actions bot added core/ Keyman Core fix labels Feb 20, 2026

keymanapp-test-bot bot added the user-test-missing User tests have not yet been defined for the PR label Feb 20, 2026

ermshiperete changed the title ~~fix(core): fix keydown/up handling for LDML keyboards~~ fix(core): fix keydown/up handling for LDML keyboards #️⃣ Feb 20, 2026

keymanapp-test-bot bot added this to the A19S23 milestone Feb 20, 2026

ermshiperete force-pushed the fix/core/15569_ldml branch 2 times, most recently from f67f988 to c0b6426 Compare February 23, 2026 08:12

Base automatically changed from test/core/minkbd to master February 24, 2026 16:40

This was linked to issues Feb 24, 2026

bug(linux): 'stuck' Enter key with ldml keyboards on Wayland #15569

Open

bug(linux): ldml keyboard desynchronizes and can crash when bksp is used in non-compliant app with decomposed cluster on Wayland #15550

Open

keymanapp-test-bot bot added has-user-test user-test-required User tests have not been completed and removed user-test-missing User tests have not yet been defined for the PR labels Feb 24, 2026

ermshiperete requested review from mcdurdin and rc-swag February 24, 2026 17:01

ermshiperete marked this pull request as ready for review February 24, 2026 17:02

ermshiperete mentioned this pull request Feb 25, 2026

docs(core): add keyhandling doc #️⃣ #15629

Merged

Meng-Heng self-assigned this Feb 26, 2026

rc-swag reviewed Feb 26, 2026

View reviewed changes

core/tests/unit/km_core_process_event.tests.cpp Show resolved Hide resolved

rc-swag reviewed Feb 26, 2026

View reviewed changes

core/tests/unit/km_core_process_event.tests.cpp Show resolved Hide resolved

keymanapp-test-bot bot added user-test-failed and removed user-test-required User tests have not been completed labels Feb 26, 2026

Meng-Heng removed their assignment Feb 26, 2026

rc-swag previously approved these changes Feb 27, 2026

View reviewed changes

rc-swag mentioned this pull request Feb 27, 2026

bug(windows): Ctrl+Bksp with LDML keyboard only deletes one character prior to insertion point, not word #15489

Open

mcdurdin previously requested changes Feb 27, 2026

View reviewed changes

keyman-server modified the milestones: A19S23, A19S24 Feb 27, 2026

ermshiperete requested a review from mcdurdin March 2, 2026 10:25

ermshiperete and others added 2 commits March 2, 2026 17:27

fix(core): address code review comments

67468c4

Co-authored-by: Marc Durdin <marc@durdin.net>

ermshiperete changed the base branch from master to refactor/core/cleanup March 2, 2026 16:34

ermshiperete mentioned this pull request Mar 2, 2026

refactor(core): cleanup #️⃣ #15656

Open

fix(core): clarify and extend comment

9836070

ermshiperete force-pushed the fix/core/15569_ldml branch from 7ec8297 to 9836070 Compare March 2, 2026 16:43

ermshiperete requested a review from rc-swag March 2, 2026 16:48

Uh oh!

Conversation

ermshiperete commented Feb 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

User Testing

SUITE_BASICTEST

SUITE_BUGSFIXED

Uh oh!

keymanapp-test-bot bot commented Feb 20, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

User Test Results

✅ SUITE_BASICTEST:

🟥 SUITE_BUGSFIXED:

Test Artifacts

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Meng-Heng commented Feb 26, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Test Results

SUITE_BASICTEST:

Test Specs

GROUP_X11: Test with Gnome Shell and X11

GROUP_WAYLAND: Test with Gnome Shell and Wayland

Test Prerequisites

SUITE_BUGSFIXED:

Test Specs

Uh oh!

rc-swag left a comment

Choose a reason for hiding this comment

Uh oh!

mcdurdin left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

ermshiperete commented Feb 20, 2026 •

edited

Loading

keymanapp-test-bot bot commented Feb 20, 2026 •

edited

Loading

Meng-Heng commented Feb 26, 2026 •

edited

Loading