We weren't unbinding these handlers at all, and so the 'ok' or 'cancel'
handlers could run multiple times for one button click, and even worse,
you could get in a situation where clicking 'ok' in one confirm dialog
would also run the 'ok' handler for the other one. This happens because
the ConfirmDialog instance is recycled by the WindowSet.
The way the unbinding is done is ugly; we should either consolidate the
'ok' and 'cancel' events so we can use .once(), or come up with some other
way to automatically unbind the handlers.
Bug: 65557
Change-Id: Iabf0c0d0229add09cc775358fc5a4e5ae783db04
"Yes, switch" isn't constructive; make it primary instead.
"No, cancel" isn't destructive; make it neutral (no flags) instead.
Change-Id: I841cbed4a81eaed679a8c7da89942c6b030a1217
"Discard edits" isn't constructive, and "Continue editing" isn't destructive.
Instead, mark "Discard edits" as destructive, and make the continue
button neutral (no flags).
Change-Id: I7648555ad47be698e75b5019d7738b0afb8611aa
Per TTO on bug 51655, the implementation of confirm() in most browsers is crappy and we
shouldn't use it.
Change-Id: I755085a253c05958e4b50af57b19dab90f2f0fb6
Follows-up e3be4a6. Object properties default to undefined, no
need to check existance first. Looks like like an "isset()" in
PHP for preventing E_NOTICE.
Change-Id: I594b23e6caf1e17d6d5d37e6a5fd81152e78b3a6
jshint:
* Update to grunt-contrib-jshint v0.10.0 (jshint v2.5.0).
* Remove coding style options covered by jscs.
* Enable new option "freeze" (prohibits changing native prototypes).
http://www.jshint.com/blog/new-in-jshint-oct-2013/#option-freeze
* Re-order to match http://www.jshint.com/docs/options/
jscs:
* Update to grunt-jscs-checker v0.4.4 (jscs v1.4.5).
* Format .jscsrc file in a more spacious way and order the
properties less arbitrarily (using the jscs's readme order).
* Enforce more details of our coding style
* Get rid of the unsable "sticky" operator rules which have been
deprecated in favour of using other rules instead that are able
to enforce this more accurately.
- disallowLeftStickedOperators: Remove deprecated rule.
* Ternary covered by requireSpacesInConditionalExpression.
* Rest covered by requireSpace{Before,After}BinaryOperators.
- requireLeftStickedOperators: Remove deprecated rule.
* Comma covered by disallowSpaceBeforeBinaryOperators.
- requireRightStickedOperators: Remove deprecated rule.
* Logical not (!) covered by disallowSpaceAfterPrefixUnaryOperators.
See also If46b94ce1, Ib731f11b1 and I0b0cadbc5 in oojs/core.
Also:
* Update grunt-contrib-watch to latest upstream version.
Change log at https://github.com/gruntjs/grunt-contrib-watch/blob/v0.6.1/CHANGELOG#L1-L17
Change-Id: I6c5a34afea8b05a3dca617897c192594df06ca90
Relies on:
* I292fb34d in OOjs UI to add the confirmation dialog
** I67329820 in MediaWiki core to use the messages added in OOjs UI
** I38f5bb63 in VisualEditor core to register the confirmation dialog
Bug: 50955
Change-Id: I98f9a03d780556b360b57c018c05a27cc1b3862e
Usually we can just send 0. Sending anything else makes PostEdit think we're restoring an old revision.
Bug: 65269
Change-Id: I60454a7a4ea3f6c7cef4c707da3016dd0ec29b88
These changes are to accomodate the design for the mobile/tablet
version of VisualEditor which uses an icon rather than a label
for the drop-down button.
Change-Id: I1086ed4a84ae4061fcc79cc7f587657232c5d5df
Three 'minor' points:
* You have to declare even hidden preferences. Whoops.
* There's no such thing as an "optionsToken", use "editToken".
* You need to POST action=options API calls.
Ahem.
Change-Id: I9c4358107af7bcfca157bd014de49882914e990c
For logged-in users, we can a preference instead of a cookie. This way it is
also preserved between browsers and when cookies are cleared.
Keep using cookies for logged-out users, except if the beta welcome dialog
has been suppressed using the one-off GET parameter 'vehidebetadialog'.
Bug: 55551
Change-Id: Ica9e5a92841fec003ce4a21d740a9bc6ff3da9c7
We don't have a FOUC on the appearing of the 'edit' link. That
one is handled quite intelligently:
* Via the stylesheet that is also loaded in noscript mode, its
(hidden) appearance is already predetermined. So as soon as
those elements are seen by the browser they style correctly
for users without JavaScript (display: none).
* This same stylesheet also hides it for users with JavaScript
but where VE is not available (e.g. due to browser support).
While ve-not-available is added very early on (before
document ready), it could in theory cause a short FOUC, but
that's okay. We simply don't know that VE isn't supported
until then. We optimise for the common case (JavaScript
enabled, VE available), while still ensuring that it is
always hidden in noscript, and is hidden as soon as possible
when VE turns out not to be available.
For some reason, one small detail (the little bit of whitespace
added inside the brackets), was left out of this and was
implemented by adding the class 'mw-editsection-expanded' to them
from a document ready handler.
* First step, get rid of the script that adds this class and
use ve-available instead. That means they're styled
correctly much earlier (we add the class to <html> before
document ready). This can still cause a brief FOUC, though
in most cases they're correct from the start.
* Step two, make brackets expand by default for script users,
and let ve-not-available reset it. This way, like with edit tabs,
a FOUC will never happen for ve-available. And even for
ve-not-available, a FOUC is rare since we add it before
document ready via <html> look-ahead styling.
There was still a brief reflow jump because of negative margins
between two paint events. One was undoing the other at a later
time. These negative margins are a remnant of when we were doing
animations (follows-up I4b9c47fd65a70). They were added to reduce
reflows and content shift, but were now actually causing them.
Removed "padding-right" from mw-editsection, and negative margin
from the brackets.
Also:
* Don't add inline 'style="direction: ltr;"' on every single
editsection throughout the DOM. This was the only operation we
were doing unconditionally. While I doubt the need of it in
general, we can at least allow MediaWiki to do it right, and
only add the override if needed. This saves quite a few DOM
operations.
Change-Id: I7a729edc2cd4a66ebc0ad6935ffd02cb9b948bff
There was a slideDown() call, but this didn't do anything since
toolbars are visible and in the DOM by default.
As a temporary hack, hide it synchronously after creation and
then do the slideDown still.
This could ever so briefly cause a flash, though that didn't
happen in my testing.
This makes the experience smoother when we initialise the surface.
In particular the moment where we swap #bodyContent for our Surface
(which should look visually almost identical), before this change
it was still a bit of a flash since the Surface version has a
toolbar on top, and thus instead of swapping smoothly, we hide
content and show a similar piece of content that has an incompatible
offset from the top.
Bug: 64751
Change-Id: Id94974ba71fd887ce494d7b2b16ec62d43b18575
Update VE core submodule to master (109a3cf):
* 109a3cf ve.ce.Surface: Remove font-size: 1em;
Follows-up I5818dce6265.
* When applied to ve-ui-surface, it applies to both of its children,
namely ve-ce-surface and ve-ui-surface-overlay. The latter
(containing things like context menu and inspectors) should not
inherit styles intended for rendering of wikipage content by
the skin.
* Move mw-content-{ltr,rtl} from an outer parent to a slightly
deeper parent of the actual content. It's now on documentNode.
Still high enough to apply to all content. This was to ensure
it remains a child of mw-body-content, like in MediaWiki core.
* Updated padding/margin hack again. Since now the font-size
applies on the surface, this once again means that now 1em
on the surface != 1em on the documentNode, so we need the
calculation again. Brought it back and this time documented
what it is based on.
Bug: 64599
Change-Id: I9ee4582a6a0ec9e128f5c36a6dbf2adbd0b216a1
Also get rid of checking for NS_MEDIAWIKI explictly and use
MWNamespace::getRestrictionLevels instead
Bug: 50783
Change-Id: I5986ddb9b6f17e4a2aca12dbb551cce4a6cfd663
* font-size and line-height differed between read mode and
edit mode. This makes them match by using the same css class
as MediaWiki core uses for the body content.
Depends on I66c56c57 in MediaWiki core.
* Document padding / margin hack on surface, and sync the values.
In edit mode, the text was being nudged to the right by a few
pixels.
* Bump the wfUseMw call as this is a symptom of how VisualEditor
now depends on MediaWiki 1.23.
Bug: 64599
Change-Id: I5818dce626545c4cbc8bd84fd551dc08baedbe32
Don't unselect article tab when loading VE, do unselect when restoring normal view mode.
Bug: 49407
Change-Id: I4b6e5c898a8af2b404151bba46359dc4bfbd739e
MathCaptcha just extends SimpleCaptcha, so its output is fine to show as
text. Doing that because I'm not sure how to render TeX and this is a
reasonable fallback
Also tidy up the order of some message entries in my last commit.
Bug: 64328
Change-Id: I98312f61471667e7c4dcf715295f85642c31a688
