- reorders don't interact with markers, but must transit them
- add updated tests
- simplify reorder_group::apply(): the 'common prefix' discussion
was out-of-date as std::mismatch is handled by the ldml_processor
and the core context.
- add_back_markers() promoted to SPI so it can be called from reordering.
Fixes: #10516
ICU 73 comes with CLDR 43 which contains various corrections. This means
that we get different results when we minimize `bmf-Latn`. This change
sets the test expectations based on the ICU version and so allows tests
to pass on Ubuntu 24.04 which comes with ICU 74.
Fixes#10520.
The definitions for `struct km_core_action_item`, `struct
km_core_backspace_item`, `enum km_core_backspace_type`, `enum
km_core_action_type`, `km_core_state_action_items()`,
`km_core_state_queue_action_items()`, `km_core_process_queued_actions()`
are now in keyman_core_api_actions.h rather than in the primary
keyman_core_api.h, as they should not be used by engines in general.
The current users of those APIs are the interactive kmx debugger, and
the IMX integration in Keyman Engine for Windows. Both of these depend
on the old action queue model rather than the actions struct.
In a future version, we may refactor these further to make them apply
directly to the KMX processor, and provide access via an interface to
the KMX processor rather than as a general Core API. Once this change is
made, then Core itself will no longer support action queues at all, and
will expect keyboard processors to fill in an action struct (this is
already done for LDML).
Fixes#10547.
In order to prevent confusion, the stubbed and incomplete .js files
generated by the LDML keyboard compiler are no longer emitted for
version 17.0 release.
Fixes#10542.
The `sourcePath` attribute should only be added to a .keyboard_info file
if the file is actually in an expected folder matching a pattern like
`(release|legacy|experimental)/k/keyboard`.
If the path does not match that pattern, then we just omit the field in
the emitted .keyboard_info or .model_info file.
Fixes#10488:
1. Use a consistent `exitProcess()` function so that we always wait for Sentry to do its thing.
2. Make `KeymanSentry.isEnabled()` non-async because it is synchronous.
3. Remove duplicate initialization of `KeymanSentry` in kmc.ts.
When the action from the core is to persist an option there
is no need to write that value back to the core as it has already
been updated, by the core.
When building packages warnings are treated as errors. When building on
Ubuntu and Debian servers we don't have a patched ibus version but still
want to build the packages. So this change outputs a message instead of
a warning when building packages.
