Tests: Increase robustness of a draggable `revert` test
One of the changes is increasing `revertDuration` from 200 to 300. For some
reason, when jQuery 3.1 is used in an iframe (which TestSwarm uses) the timings
are lower than expected and `setTimeout` too slow and the check for animation
happens too late.
Tests: Workaround a jQuery 1.8 issue with one dialog test
The test "#8958: dialog can be opened while opening" had special behavior in
jQuery 1.8 and the patch from commit 7adb13ac7c30a7ba33f1db256952be77b9e7a310
broke those assumptions. Adjust the test so that it passes again.
Tests: Workaround IE issues in qunit-assert-domequal
In IE, `option` elements may have different initial `option` colors.
They may initially all be transparent, but later the selected
option gets a blue background with white text; we now ignore it.
The logic of `qunit-assert-domequal` was also fixed to use the same
method of fetching styles in all browsers; IE used to get a legacy
one meant for IE <9 due to a mistake in the performed check.
Tests: Fix selectmenu width tests in Firefox with jQuery 3.0 & 3.1
jQuery 3.0 & 3.1 used `getBoundingClientRect()` in its `width`/`height`
calculations and that causes minor differences in fractional width computations.
Allow a tiny delta in tests to fix those tests breaking in Firefox.
Dialog: Fix IE tests about interactions with other dialogs
The workaround triggering a second delayed focus to a tabbable element was
applied in jQueries 3.4 & 3.5 to fix IE issues. It also needs to be applied
in jQuery 3.6 which this commit adds.
This will no longer be needed in jQuery 3.7 and jQuery 3.x-git tests are
already passing because of that.
Selectmenu: Fix selecting options following hidden ones
Change a2b25ef6caae3e1a272214839b815a6387618124 made options with
the `hidden` attribute skipped when rendering. However, that makes
indexes misaligned with native options as hidden ones maintain their
index values. Instead, don't skip hidden options but add the `hidden`
attribute to the respective jQuery UI elements as well.
Build: Fork vendors to remove QUnit deprecated API usage
Changes:
* add `tests/lib/vendor/**/*` to `.eslintignore`
* move `qunit-composite` to `tests/lib` so that we can modify it
* move `qunit-assert-classes` to `tests/lib` so that we can modify it
* move `qunit-assert-close` to `tests/lib` so that we can modify it
* replace `assert.push` with `assert.pushResult`
* remove usage of `QUnit.extend`
The `set-output` workflow is deprecated in favor of using environment
variables, see:
https://github.blog/changelog/2022-10-11-github-actions-deprecating-save-state-and-set-output-commands/
Also, workaround issues with QUnit Chrome bridge: the Chrome bridge
from `grunt-contrib-qunit` is now getting injected into every single
iframe, including an empty one that has no intention of running QUnit
tests. Since that bridge requires QUnit, it fails with an error
in such cases. Workaround the issue by wrapping the bridge in
another function that bails early if QUnit is not defined.
Self-closing tags are reported by newer versions of the htmllint
Grunt plugin. They also don't make sense in our HTML files
since they are not XHTML-compliant and they run in HTML mode
anyway.
Kenneth DeBacker [Thu, 30 Mar 2023 11:05:41 +0000 (07:05 -0400)]
Tests: Test only valid values for Datepicker defaultDate, min/maxDate
The docs say that valid periods when using string value and
period pairs as relative dates are "y", "m", "w", and "d"
https://api.jqueryui.com/datepicker/
Checkboxradio: Don't re-evaluate text labels as HTML
If you generate a Checkboxradio from a checkbox/radio with a label that
contains encoded HTML, e.g. `<em>test</em>` this will work fine
at first. If, however a refresh is triggered on that instance (explicitly or
e.g. by turning it into a `Controlgroup`), the previously escaped HTML will
now be evaluated.
If the label was created based on some user input, this could lead to
unexpected code execution even though the initial output was escaped.
Unfortunately, we cannot update all of them to their latest versions as some
of them now require Node.js 12 and we need to be able to build on Node 10 as
that's what our Jenkins runs.
Widget: Optimize attachment of the _untrackClassesElement listener
jQuery UI 1.13.0 changed the logic attaching the `_untrackClassesElement`
listener in the `_classes` widget method; one of the side effects was calling
`this._on` for each node that needed the listener. That caused a severe
performance degradation for large comboboxes as each `_on` jQuery UI call
causes a jQuery `add` call that calls Sizzle's `uniqueSort` underneath.
Instead, collect the nodes that need the listener and then, outside of the loop,
create a jQuery object out of them and attach the listener once. That's still
slower than the jQuery 1.12 version but only slightly: 936 ms to 1.03s on a very
large list on a recent MacBook Pro, compared to ~30 seconds before this patch.
Ben Mullins [Wed, 5 Jan 2022 10:35:34 +0000 (05:35 -0500)]
Autocomplete: Rewrite with a delay instead of appending the live region
This fixes the issue caused by https://bugs.jqueryui.com/ticket/9357.
We now empty the live region instead of appending to it, and we do so
after a brief timeout so the live region isn't updated on every mousemove
event or when quickly traversing through options.
Due to the fact the widget factory code is now in strict mode, the check for
being called without using the `new` keyword started breaking if you save the
widget constructor to a variable before calling it:
```js
var customWidget = $.custom.customWidget;
customWidget( {}, elem );
```
as then `this` is undefined and checking for `this._createWidget` crashes.
Account for that with an additional check.
Sortable: Fix positioning when moving a Draggable item into a Sortable
PR gh-1793 removed setting `this.offset.parent` in the Draggable
`refreshPositions` method which broke position calculations when moving
a Draggable item into a connected Sortable. restore that assignment.
Josep Sanz [Mon, 8 Nov 2021 10:53:45 +0000 (11:53 +0100)]
Tooltip: Don't crash on empty content
Commit 1f2011ece removed a `try-catch` around triggering the `remove` handlers
in the `jQuery.cleanData` override. The `try-catch` was meant for old IE but it was
also catching an error coming from the tooltip `remove` handler depending on
being able to find a relevant tooltip. The `_find` method returns `null`, though,
when the tooltip cotent is empty.
Instead of restoring the `try-catch`, handle the `null` case in the `remove` handler.
Fixes gh-1990
Closes gh-1994
Co-authored-by: Claas Augner <github@caugner.de> Co-authored-by: Michał Gołębiowski-Owczarek <m.goleb@gmail.com>
A fix from 43254468de7d69b5422e667ba7ebbe864fc34a63 introduced a rule setting
`background-image` of a blank icon to none. However, the selector used for that
rule had lower specificity than another one: `.ui-button .ui-icon` which caused
dashes being shown over the icons.
We needed to increase the specificity of the `.ui-icon-blank` rule past the
above selector and past `.ui-button:hover .ui-icon`. We're doing it by repeating
the class name three times.
Effect: Define the jQuery variable before jQuery Color gets imported
We need to create a local jQuery because jQuery Color relies on it and the
global may not exist with AMD and a custom build (trac-10199). This worked
in UI 1.12 but stopped in 1.13 as jQuery Color is now sourced as an AMD module
and the variable started being defined after jQuery Color code. To restore the
proper order, move the variable declaration to a separate small module loaded
before jQuery Color.
Build: Don't save temporary npm dependencies to package.json
Past npm versions required the `--save` flag to save anything in `package.json`
when installing packages but newer ones do this by default. Pass `--no-save` to
restore the original behavior in the `release-test.js` script.
projects / jquery-ui.git / log