This change makes it possible to compile even when the updated
ibus version is not installed. Of course ordered output won't work
in that case, but at least it will compile and the rest of Keyman
will work.
Fixes#7774.
This reverts commit acab920de8.
I'm confused - now we're getting the same warning again - claiming
that `size_t` is defined as `long unsigned int` and so we have to
use `%lu`!? Is this related to the platform we're compiling on?
This time, the warning we get is when compiling for x86_64:
```
engine.c: In function ‘get_current_context_text’:
engine.c:224:15: warning: format ‘%u’ expects argument of type ‘unsigned int’, but argument 5 has type ‘size_t’ {aka ‘long unsigned int’} [-Wformat=]
224 | g_message("%s: current context is:%u:%lu:%s:", __FUNCTION__, km_kbp_context_length(context), buf_size, current_context_utf8);
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
| |
| size_t {aka long unsigned int}
engine.c:224:40: note: format string is defined here
224 | g_message("%s: current context is:%u:%lu:%s:", __FUNCTION__, km_kbp_context_length(context), buf_size, current_context_utf8);
| ~^
| |
| unsigned int
| %lu
```
Reverting the previous fix for now.
This refactoring makes use of the already defined strings.
It also simplifies setting the values by directly specifying an
UTF-16 string.
Motivated by the changes in PR #7667.
The previous method name `ibus_keyman_engine_commit_string`
suggests that this method is overridden from ibus-engine class.
However, while it is true that ibus-engine has a
`ibus_engine_commit_string` method it can't be overridden. This
change renames the method to make it clearer that this is an internal
method that won't be called by ibus.
Both `ibus_keyman_engine_reset` and `ibus_keyman_engine_focus_in`
are methods that get called by ibus. Having `ibus_keyman_engine_reset`
call `ibus_keyman_engine_focus_in` makes it much harder to see in
log files which methods get really called by ibus.
This change replaces the call of `...focus_in` with `reset_context`,
which is what `...focus_in` does as well. The only other thing that
`ibus_keyman_engine_focus_in` does in addition to that is calling
`ibus_engine_register_properties` - which shouldn't be necessary on
a reset (besides that we don't seem to do anything with the one
status property we register).
The previous changes didn't work when we output a character +
deadkey. In response to outputting the character we get a
`ibus_keyman_engine_set_surrounding_text` call in which we
called `reset_context` which removed the deadkey marker. This change
fixes this problem by comparing the surrounding text with the
context we already have. Only if it's different we reset the
context.
This change always gets the context from the surrounding text
if `reset_context` gets called. This helps when the user puts the
IP after a certain character. Previously we basically lost the
context, with this change we restore the context so that it's like
the user just typed the previous characters.
We forward a fake event for both keydown and keyup. This change
will set the flag on the fake event so that we get keydown and keyup.
This is probably not technically necessary, but it helps in
debugging because you can then see that both fake events still
belong to the same keypress.
If we're dealing with an older ibus version that doesn't have the
necessary patches we fall back to the old behaviour. Otherwise
we use the new output reordering.
This change implements a commit queue which allows to control the
order of the output. This ensures that any backspace we generate
will be processed before the character we're adding.
Requires changes in ibus (surrounding text fix (see #7072) and
prefilter change)
Fixes#1489, #4028, #4029, #4030, #4505, #5510, #6639
Previously `process_unicode_char_action` checked for
`action_item->character` being a surrogate. This should never happen
because `action_item->character` is a 32-bit character. This change
removes the unnecessary check and the `keyman->firstsurrogate` field.
Fixes#6385.
The implementation of `ok_for_single_backspace` checked the current
action_item multiple times. However, we already know that that
item is `KM_KBP_IT_BACK` because the only place this method gets
called is from the `process_backspace_action` method. So basically
this method always returns `FALSE`. "Fixing" this method to check
the next action_items makes things fail.
Removing this method since it doesn't do anything except burn
processing cycles...
Fixes#5614.
This change modifies the `run-tests.sh` script so that tests
run on both Wayland (if supported) and X11 unless
`--no-wayland` or `--no-x11` is specified.
This change also adds the new argument `--wayland` to the
`ibus-keyman-tests` executable which runs the tests under
Wayland.
Note that running Wayland tests is only possible in Ubuntu 21.10+
because it requires a version of `mutter` that supports the
headless option which got added in mutter 40.
This allows to use Keyman with the Wayland display server.
One thing that doesn't yet work completely with this change is
caps-lock.
Partially implements #4273.
This change adds integration/regression tests for ibus-keyman. It
re-uses the unit tests that we have for core.
Also modifies one test so that the context doesn't use characters
used in the keyboard. This makes it easier to setup in ibus-keyman
tests.
This will allow to use the array in unit tests.
# Keyman Conventional Commit suggestions:
# - Consider appending the text ". Fixes#5613" to your commit message.
This fixes typing `n>` with sil_ipa keyboard which no longer output
the expected `ŋ` because pressing the shift key reset the context.
For modifier key presses we now exit `KMX_ProcessEvent::ProcessEvent`
early.
This change also removes a redundant line from engine.c. Clearing
the context happens in `reset_context()` again if necessary.
Fixes#5591.
Fixes#5488.
This updates Keyman Core, corresponding tests, and Keyman Engine for
Linux to support deletion of markers through an action, ensuring that
the action queue does not desynchronize with the context.
Don't re-set context if it changed because the cursor location was
wrong. When the user switches to the EuroLatin keyboard in the
middle of the line, we're getting a wrong cursor location. This will
be fixed with the next keypress, but it leads to the context string
suddenly having more characters added to the front. If we re-set the
context because it's different now we will loose the information we
previously added to the context.
We now compare the end of the string. If the new context string ends
with the old context string we assume that the cursor position got
updated and don't re-set the context.
This fixes#4678.
