- CPTextField, CPTokenField, CPTextView: the PLATFORM(DOM) gates
around _inputElement (superclass and override) and _cumulativeOffset
are removed; the methods now exist under both targets with runtime
CPDOMAvailable guards returning nil (object methods) or a zero
point (struct method) when no DOM is present
- contract change: under CommonJS these methods change from absent to
present-and-inert; all callers are browser event paths never
exercised by the theme blender or nib2cib
- zero preprocessor directives remain in these three files
- work-in-progress towards complete removal of CPP directives
Merges 71 commits from main. Headline content: the Tests/Manual
integration test suite revived and reorganized after long disuse
(renames, new applications, updated cibs), restoring a large
integration validation surface. Functional fixes include replacing
unload-based termination with pagehide (back/forward cache
compatibility) and hardening CPTextField's input-element handling.
Conflict resolutions (local versions retained):
- AppKit/CPTextField.j: main's input-element logic carried, with this
branch's preprocessor-removal sweep re-applied on top. Main's
conditionals are re-expressed as runtime CPDOMAvailable guards; the
_inputElement method-definition gate is retained.
- AppKit/Platform/CPPlatform.j: the DOM platform implementation stays
inlined under the PLATFORM(DOM) gate. The split layout breaks the
legacy build: the include fragment is compiled standalone by the
source walk and cannot resolve its superclass. Main's pagehide
termination fix is ported into the inlined section; the restored
AppKit/Platform/DOM/CPPlatform.j fragment is deleted.
Verified: legacy build green through Foundation, AppKit and Tools;
Tests/Manual suite passing.
- CPTextField.j: 14 PLATFORM blocks resolved; mechanical translation
to runtime CPDOMAvailable guards (wrapped, compound, early-return,
one runtime if/else); one site used a nonstandard directive
spelling (#if PLATFORM (DOM), space before the parenthesis),
normalized and resolved with the rest
- The method-definition gate around _inputElement remains, out of
scope for this change
- work-in-progress towards complete removal of CPP directives
There was a problem introduced by pull request #2600 when trying to select and copy text in a selectable text field.
The problem only occurs when the text field is in a modal window.
It was caused by the modal run loop that is present as an eventListener on the CPApplication when in modal mode. It will
force a bail out from the sendEvent: method in CPApplication when trying to handle the event. As the check for key
equivalent is moved down by the #2600 pull request it will never be allowed to handle a copy command from the browser
menu or key equivalent.
The solution is to allow the text field to handle the event if it is selectable. This will also allow the text selection
to be altered from the keyboard with keys like shift-left/right/up/down arrow. Combinations like shift-option arrow also
work for whole word selection.
When a textField is used in a tableView, setValue forThemeAttribute
should refer to state CPThemeStateTableDataView instead of
CPThemeStateNormal as #2744 is now more accurate on states.
Previosly, we unflagged ```CPTextFieldInputDidBlur``` and ```CPTextFieldInputResigning``` before
the call to removeChild. This resulted in DOM exceptions in Chrome under certain conditions.
See https://stackoverflow.com/questions/21926083/failed-to-execute-removechild-on-node for
why we need to unflag ```CPTextFieldInputDidBlur``` and ```CPTextFieldInputResigning``` only after removing the element.
This apply to token fields also but with the corresponding flags. Also token fields now have the same behavior as the text field when bluring etc.
Without this change, clicking the done button in iOS Safari would dismiss the virtual keyboard but then it would immediately come back (as we forced a refocus).
The fix is to no longer force a refocus since it turns out we don't need to anymore given another previous fix.
Previously, when double-clicking on a table cell to edit it, there was a possibility of a race condition. A refresh of the display was requested, then the run loop was passed through once. But it was possible for the display refresh to not be queued when the run loop was passed through, in which case the refresh would cancel the editing. Or at least I think that's what was happening. ;-)
This commit (hopefully) eliminates the race condition by synchronously refreshing the layout and display.
Made some miscellaneous formatting fixes as well.
daboe01
David Richardson
Martin Carlberg
Alfred Wärnsäter
Didier Korthoudt
Didier Korthoudt
daniel
Alexander Ljungberg
Alexandre Wilhelm
Didier Korthoudt
Aparajita Fishman