Fixes the following crash noted when testing on keymanweb.com, focusing
during the load process on a slow network.
```
helpers.ts:111 Uncaught TypeError: Cannot read properties of null (reading 'vkbd')
at g.pageFocusHandler (keymanweb.ts:125:44)
at sentryWrapped (helpers.ts:87:17)
```
When embedding the On Screen Keyboard into a page, the calculation for
the bounding rectangle only took into account the first offsetParent,
which meant it would get the offset incorrect when there were multiple
layers of offsetParents.
Once I changed to using getAbsoluteX / getAbsoluteY, this also corrects
for scroll with window.pageOffsetY, so we can eliminate that special
case for fixed positioning.
This was observable in the Web Developer test window, where interaction
with the bottom row of the keyboard in touch mode was not working
correctly, in 15.0.210-beta.
The Caps Lock change I made in the previous commit broke
desktop Caps Lock handling. The fix was a little more
convoluted. Turns out that _UpdateVKShift is also pretty
crufty. I simplified it, removing dead code.
Fixes#5964.
It seems that the test for subkey position was incorrect when the
document had been scrolled. Removing the test for `fixed` positioning
resolves the issue.
This fixes an issue that arises when loading the Keyman Developer Server
home page and there are registered keyboards, but no keyboards active.
The floating osk view would throw an exception because it would attempt
to set the title of the view from the active keyboard, which is `null`.
Selection direction was not maintained in mutations, which could have
unexpected consequences. Added support for selection direction to input
and textarea.
The functions `getTextBeforeCaret()` and `getTextAfterCaret()` are named
somewhat incorrectly, as they actually get the text before and after the
active selection (and a collapsed zero-length selection is equivalent to
the caret). It would be worth renaming these in a future refactor.
This PR fixes the unit tests so that caret position is tested correctly
with an active selection -- the caret can be at either the start or the
end of the selection, corresponding with the direction in which the user
originally selected the text. It also fixes the assumptions around the
above named functions for `input` and `textarea` types.
Note that selection interactions are still buggy with prediction
selections; these bugs were present in 15.0.118-alpha and I will tackle
them in an upcoming commit.
Relates to #5853 and others.
Selection management was not working properly with the various
OutputTargets:
1. When there is a non-empty selection, rules have no context -- it's
like new text.
2. Backspace over a selection deletes just the selection.
3. Typing a character replaces the selection, of course, and collapses
the caret to the end of the new text.
4. `hasSelection` is a very strange name for `OutputTarget` descendants.
It doesn't mean "has an active selection" but rather, kinda means
"supports selection internally".
5. Added `isSelectionEmpty` which is used for some of the new selection
rules above.
Note that the `touchAlias` OutputTarget class does not currently support
selection. I hope we can deprecate `touchAlias` with the use of
`inputMode` (#3030) in the future, rather than adding support for
selection.
Relates to #5853.
Two things happened here:
1. Construction of Mocks made an assumption that the selection should
always be deleted (outputTarget.ts:363). However, for NewContext and
PostKeystroke processes, we don't want to change anything.
2. Even if nothing is changed, the transcription would emit what is
in theory a no-op ruleTransform (insert="", deleteLeft=0,
deleteRight=0). But apps would treat this as deleting the selection.
This fix goes a little broader than I would have preferred, but adds a
readonly mode to the transcription and mock model, so that we can
control explicitly when changes are applied to the text store.
If the OSK had a non-integer width or height stored in the OSK cookie,
then the width and height would be parsed incorrectly and the OSK would
fail to be sized correctly on first load; this also caused a script
error and made the OSK impossible to interact with.
Fixes#3620.
Implements the Caps Lock layer support and the double-tap gesture on the
shift key to access it.
The double-tap gesture has been implemented with a view to extension to
support other multi-tap gestures in the future. However, for now, it is
limited to supporting the Shift key, if and only if the keyboard
includes a Caps layer.
The reason for this v15 limitation is that multi-tap on regular keys
would involve either rewinding the previous keystroke (the first tap),
or forcing keyboard developers to consider 'rota' style rules in their
keyboards to support the multi-tap gestures, as we need to make sure
that the first tap is accepted and processed for immediate feedback.
This needs more design, to avoid unnecessary complexity in the keyboards
and/or the rewinding of the keystroke (even though that is conceptually
supported in Keyman Engine for Web already). Basically, we don't want to
constrain the way that a keyboard author may use the multi-tap gesture
by hard-coding the rewind, but neither do we want to make all multi-tap
gestures needlessly complex to author.
The shift key (and other modifiers, potentially in future) needs special
support for multi-tap as the key that is being tapped changes with the
layer change. This is currently managed through recognising `K_SHIFT` in
the key id.
I have tried to follow the `PendingGesture` pattern for multi-tap, and
the gesture itself supports a series of taps, not just a double-tap. The
maximum time to complete the tap series is 125msec * number-of-taps, so
for a double-tap is 250msec.
The changes to support a Caps Lock layer itself were minimal; just
adding the `text.KeyboardProcessor.getStateFromLayer` function and
calling it during `KeyEvent` construction. The remaining changes relate
to the multi-tap gesture.
Minor changes:
* I moved `constructNullKeyEvent` to `KeyEvent` in order to make it
more accessible to other classes.
* The multi-tap gesture does not have a promise to complete, so that is
now an optional member of the `PendingGesture` interface.
This makes two corrections:
1. Removed an obsolete test for screen size based on dpi, which is
unreliable, meant that we had been treating many Android tablets as
phones (domManager.ts).
2. Font scale for Android tablets had some incorrect assumptions around
keyboard height and devicePixelRatio, which I was able to simplify
into a fixed ratio, and was much better across all devices I
tested (however, all emulated in Chrome).
Summary of these changes:
Two targeted improvements; more optimization is possible but IMO this is
a significant enough improvement to go through test:
1. In oskView.ts, avoid calling `VisualKeyboard.refreshLayout` twice
from `OSKView.refreshLayout`. This roughly halves the time spent in
`VisualKeyboard.refreshLayout` for a keyboard switch.
2. In modelManager.ts, we avoid a very expensive `unloadModel` /
`loadModel` sequence from `registerModel` by checking to see if the
model spec we are passed is already registered. This avoids an
expensive callback to Keyman for Android which causes the banner to
'bounce'.
Longer notes:
So `VisualKeyboard.refreshLayout` is called 20+ times during a keyboard
switch. It is a very expensive call. It is the bulk of the time spent
when you press the globe key in Keyman for Android. This seems ripe for
optimization; see the following functions in keyboard.html:
* `setKeymanLanguage` -> `setActiveKeyboard` -> 6 calls
* `enableSuggestions` -> `registerModel` -> 8 calls
* `stateChange` -> 6+ calls
`OSKView.refreshLayout` always seems to call
`VisualKeyboard.refreshLayout` twice: first with `setSize`, and then
again itself.
Keyman for Android and KeymanWeb seem to fight over who is responsible
for selecting models (KMW first calls `loadModel`, then Keyman for
Android calls `unloadModel`, `loadModel`, through `enableSuggestions`).
The fix for rotation (`correctOSKTextSize()`), as called from
`stateChange()`, is expensive as well, because it rebuilds the entire
keyboard from scratch.
Along with that, we have the size of the webview changing as the model
is unloaded and reloaded and lots of back-and-forth between the
KeymanWeb and KMEA which seems unnecessary.
Shouldn't KeymanWeb be responsible for loading and unloading models,
once they are all registered? We should only need to register the models
once at page load time (when config changes, do we just reload the
page?).
Fixes#5950.
If a user starts a longpress gesture and then flicks up, that
immediately brings up the longpress menu, and the timeout for display of
the menu should be cancelled -- otherwise the end result is a key that
stays 'stuck' on.
Fixes#5944.
If a custom OSK, such as sil_euro_latin, calls `keymanweb.KSF()` or
`keymanweb.saveFocus()`, this ensures that the next focus change is
ignored, which prevents the OSK from being dismissed as soon as it is
clicked on.
This is a regression, probably arising in version 10.0.
Note: `_SelectionChange` is no longer used or referenced, so I removed
it as part of this fix.
