Show an inspector with inputwidget when the user clicks a math
node. The data of the math equation is shown in the
edit box, it can re-render the math tag image when
the inspector is closed, and save the change when
saving the page.
TODO:
* Change the icon from link to math
* Translate title by translatewiki in i18n
* Other further UI improvements
Change-Id: I4d7533af25186cc39cc4bc6a4326d222ffd6db19
This patch creates two files handling math node matching,
which are:
ve.dm.MWMathNode.js - a basic version of toDataElement
and toDomElement functions work
ve.ce.MWMathNode.js - math formulas in VisualEditor are
clickable and render as img tags
Change-Id: Ib909c5fb02e385c88050f42d02d207ab6a97d0dd
Any annotation could have a style attribute which has
a compound effect, and this also prevents roundtrip
errors. Button tool logic should prevent any annotation
from being applied twice which shouldn't be.
Bug: 49755
Change-Id: I8502a55cd2b195d28d0e2ecd63f15de670f80d60
Since we're now only loading the light-weight init on page load,
the section editing wasn't just deferred to after page load (like
it was before), but wasn't happening at all until you clicked
"Edit" (at which point the library loads). It only worked when
going back to "View" after "Edit".
Contrary to tab layout, edit section handling needs to be
accessible both in the top init and in the main target class
because we need to run it both at run time and after the user
has saved a page when we show them the updated page without
refresh. This is why we need to transfer the method at run time
and give the main class access to it as well.
Can't wait for bug 50707 to get rid of this mess...
Bug: 50731
Bug: 49993
Change-Id: Iab9c81222df7f1084179c3643d158374a89ca14b
This re-fixes the fix to the frame fix. fix fix.
Two things changed: Background position flipped to the right properly
(was somewhat off before) and text-alignment flipped as well.
Bug: 49613
Change-Id: I0aa317ef674537d20d7ed64e74eaa3cfe8030c8c
This was broken after 92c38ea removed ve-mw files from the
static loaders but didn't update makeStaticLoader.php so
it was harder to keep the files in sync and a few minor issues
on the documentation pages.
Since the files are still mixed in the non-mw modules the only
way to exclude them is to do what was manually done in 92c38ea:
Filter out paths starting with "ve-mw/" from non-mw modules.
Change-Id: Id58ee89ac18c63c01719dc11ec7a07ddeee3ea0b
Initialisation initialisation? It's time to rename ve.init
to ve.platform and ve.init.Platform to ve.platform.Environment,
but that'll come later.
* Moved support detection and skin set up to separate class-less
file.
* Swapped usage of ve.msg for mw.msg.
* Callback of edit tab now does an mw.loader call to fetch
the actual VisualEditor libraries.
Though mw.loader won't load the same thing twice, we would
bind a callback each time. To avoid instantiating ViewPageTarget
more than once we use a Deferred.
Bug: 50542
Bug: 50608
Bug: 50612
Change-Id: Ic8b0004ab5288fa91bb29d496485b93ffd8d977e
Objective:
* Make the majority of link inspector, which is generic to any annotation, usable for other annotation inspectors
Change-Id: I1f7e9c13537105da7aa0351c9c92e8af5eb5a3f4
Just a proof-of-concept that this would work on Opera. I'm figuring out
what was this for initially...
This reverts commit 30dadb6c8c.
Bug: 47794
Change-Id: Ie4f1334e0beb7ba7d4f4aa3b42464fa219c4778c
!$element.is( ':visible' ) - jQuery .is() returns boolean based on the
parameter.
$element.not( ':visible' ) - jQuery .not() returns a filtered jQuery
collection with elements matching the parameter removed.
Change-Id: Iedf55f3453514ee710bbb6d702bbf9a5570a4a1d
Move all MW-specific files into the ve-mw directory, in preparation
for moving them out into a separate repo.
All MW-specific files were moved into a parallel directory structure
in modules/ve-mw . Files with both generic and MW-specific things were
split up. Files in ve/init/mw/ were moved to ve-mw/init/ rather than
ve-mw/init/mw ; they're still named ve.init.mw.* but we should change
that. Some of the test files for core classes had MW-specific test cases,
so those were split up and the test runner was duplicated; we should
refactor our tests to use data providers so we can add cases more easily.
Split files:
* ve.ce.Node.css
* ve.ce.ContentBranchNode.test.js (MWEntityNode)
* ve.ce.Document.test.js (some core test cases genericized)
* ve.dm.InternalList.test.js (uses mwReference test document)
* ve.dm.SurfaceFragment.test.js, ve.ui.FormatAction.test.js
** Made core tests use heading instead of mwHeading
** Updated core tests because normal headings don't break out of lists
** Moved test runners into ve.test.utils.js
* ve.ui.Icons-*.css
* ve.ui.Dialog.css (MW parts into ve.ui.MWDialog.css)
* ve.ui.Tool.css
* ve.ui.Widget.css (move ve-ui-rtl and ve-ui-ltr to ve.ui.css)
ve.dm.Converter.test.js: Moved runner functions into ve.test.utils.js
ve.dm.example.js:
* Refactored createExampleDocument so mwExample can use it
* Removed wgExtensionAssetsPath detection, moved into mw-preload.js
* Genericized withMeta example document (original version copied to mwExample)
* Moved references example document to mwExample
ve.dm.mwExample.js:
* Move withMeta and references example documents from ve.dm.example.js
* Add createExampleDocument function
ve-mw/test/index.php: Runner for MW-specific tests only
ve-mw/test/mw-preload.js: Sets VE_TESTDIR for Special:JavaScriptTest only
ve.ui.Window.js:
* Remove magic path interpolation in addLocalStyleSheets()
* Pass full(er) paths to addLocalStyleSheets(), here and in subclasses
ve.ui.MWDialog.js: Subclass of Dialog that adds MW versions of stylesheets
ve.ui.MW*Dialog.js:
* Subclass MWDialog rather than Dialog
* Load both core and MW versions of stylesheets that have both
ve.ui.PagedDialog.js: Converted to a mixin rather than an abstract base class
* Don't inherit ve.ui.Dialog
* Rather than overriding initialize(), provide initializePages() which the
host class is supposed to call from its initialize()
* Rename onOutlineSelect to onPageOutlineSelect
ve.ui.MWMetaDialog.js, ve.ui.MWTransclusionDialog.js:
* Use PagedDialog as a mixin rather than a base class, inherit MWDialog
bullet-icon.png: Unused, deleted
Stuff we should do later:
* Refactor tests to use data providers
* Write utility function for SVG compat check
* Separate omnibus CSS files such as ve.ui.Widget.css
* Separate omnibus RL modules
* Use icon classes in ViewPageTarget
Change-Id: I1b28f8ba7f2d2513e5c634927a854686fb9dd5a5
This actually fixes a hack done in a previous commit where an undocumented
surface option was being passed, but not always. And in some cases there
were uses of the popup which didn't really have a surface to pass in.
Change-Id: I08b5dc2dc879f6abd1b16d1d66ac1d8b5624f673
Set a static property on big, small, sup, sub to allow them
to be added multiple times to an annotationSet.
Fix the converter to count out annotations when opening/closing.
Bug: 49755
Change-Id: Ifbede9345a66434022dbd681eada447ab81ab025
Objective:
* Allow browsing, searching and adding documented parameters
Changes:
ve.ui.MWTransclusionDialog.js
* Replace regular text input with ve.ui.ParameterSearchWidget
* Fix uses of $(), this.frame.$$ is correct
ve.ui.Dialog.css
* Change rules for addParameterFieldset to make search widget auto-size vertically
ve.ui.Widget.css
* Add styles for ve.ui.MWParameterResultWidget
ve.ui.MWParameterSearchWidget.js, ve.ui.MWParameterResultWidget.js
* New classes
* Provides a way to search and select parameter for a template
* Displays parameter label, name, aliases and description
*.php
* Links to new files and messages
Change-Id: Ie5dbe8c44ce5d64c4b49b09517fb66cd30dd7304
This fix adds a directionality property to the frame objects so directions
can be inherited and manipulated inside the frame <body> tags.
There are also several corrections to flipping positioning of panels
in the GridLayout and of the icons in fieldsets and labeled elements
so they appear properly (right instead of left).
I've added a 've-rtl' / 've-ltr' class to frame content divs to serve as
selectors for rtl fixes and general language fixes. Most CSS corrections
would likely be fixed when CSSJanus works in iframes, but the selectors
would still be useful.
This sets the stage for directionality inheritance inside iframes,
and fixes most of the problem with the Page Settings window in RTL wikis.
It also fixes all the labels with icons across VE interface.
Bug: 49613
Change-Id: I418e669b0999552167683352acb365a4249ab9cc
Per the TemplateData spec:
* Most message values can be null, though they're never undefined.
* 'deprecated' is actually {boolean|string}.
For us:
* Make #isParameterDeprecated always return a boolean
* Add #getParameterDeprecationDescription to get the explanation
of why a parameter is deprecated
* getParameterDescription doesn't need to pass 'null' for fallback
since that's the default fallback.
And it may return null or string.
* #getParameterSets returns an array of Set objects, not an array
of strings (see spec for more information).
Change-Id: Iee788bc8fa6f0545085a0f154b77b8176f770dd2
ve.ui.Surface.js
* Make local overlay a child of ve-ui-surface and a sibling to
ve-ce-surface elements.
** This keeps local overlays relative to their surface and eliminates the
need for insane z-indexes.
ve.ui.PopupWidget.js
* PopupWidget boundaries are now relative to ve-ce-surface and no longer
protrude out
ve.ce.Node.css, ve.ui.Window.css
* Removal or replacement of insane z-indexes.
ve.ce.FocusableNode.js, ve.ce.ProtectedNode.js, ve.ce.ResizableNode.js,
ve.ui.Context.js
* Translate offsets from local overlay
ve.init.mw.ViewPageTarget-monobook.css,
ve.init.mw.ViewPageTarget-vector.css
* Skin specific z-indexes for global overlay
ve.init.mw.ViewPageTarget.js
* Applied direction specific mw class to ce.Surface vs ui.Surface to
prevent mw content styles from being applied to ui elements.
ve.ui.Dialog.css
* Adjustments to surface inside of dialog so that relative offsets for
local overlays can be properly calculated.
ve.ui.Surface.css
* Explicitly force .ve-ui-surface to be relative so that it's children can
be relatively positioned.
ve.ui.Widget.css
* Removal of unnecessary font-size properties now that local overlay is
sibling of surface.
ve.js
* Added get relative position helper method to translate position offsets
from target parent
Bug: 50241
Change-Id: Ibadce404a2286bc5dcec48f0d9da89004dbbd867
Resize handle event handling was set up to bind on focus and
unbind on document mouse up. The problem is that focus can be
set programmatically and sometimes more than once. This would
cause the image to be properly resized once, but immediately
resized again to 0x0. Now, before binding events to the resize
handles, all resize handle events are unbound.
Change-Id: I22b7902d303b5ca43e132c121d743d6815980fc6
If a block image is resized and undone, there are lots of problems.
* The block image gets the wrong size. When rolling back the
transaction, height and width are passed as strings.
MWBlockImage tries to set the width to: width + 2. A 300px image
became 3002px.
* Resize handles didn't reposition and resize
* Relocatable drag marker didn't reposition and resize
* Focusable highlights didn't reposition and resize.
Change-Id: I8792b567b77da90ce5a2ad132e306705c5706606
Reference lists would never disconnect, and some transactions would cause them to be "updated" even though they don't have a document anymore.
A way to test this is to insert an image in a document with some references. When you click apply, it crashes.
The solution was to make sure we actually disconnect from the internalList and listNode on teardown.
Change-Id: Ieb0354938b68fe5e7965f61f9ccce07cd6fb1120
Added both bidi-isolation and directionality. Since the directionality
has to come from the user interface (and not the wiki dir itself) I had
to read the $('body') direction. The other option could be adding a
mw.config.get().userLanguageDir that picks up the interface directionality
on top of the wiki directionality.
Bug: 50543
Change-Id: I9a00e9545a46f13750ab0d118b23d05573ee1a2e
Objective:
* Enable navigating search results using keyboard up and down keys
* Bubble up query input enter events
Changes:
ve.ui.SearchWidget.js
* Route query input events to emit
* Add handler for keydown events
* Support navigating through the list of results using vertical arrow keys
Change-Id: Ia2e4b27075a8ab2e29a69294e9f7847e8f5a3f83
Issues:
* Selected item is removed, but no select even was emitted
Changes:
ve.ui.SelectWidget.js
* On remove, emit select null when removing a selected item
* On clear, emit select null always
Change-Id: I1a94b0f8a841611b20e09f6c02f37fb9589e85c3
Changes:
ve.ui.SelectWidget.js
* Make getRelativeSelectableItem properly loop around
* Replace confusing and broken getClosestSelectableItem with getFirstSelectableItem since that's the only use case we had for it anyway
ve.ui.PagedDialog.js, ve.ui.LookupInputWidget.js
* Update calls to getClosestSelectableItem to use new method
Change-Id: I2399d01a45c43d1ad663ed6c6de156e796065306
Changes:
ve.dm.MetaList.js
* Auto-initialize index to the end of the list (like the docs claim) even when not providing an offset
ve.ui.MWCategoryInputWidget.js
* Cleanup fix for not offering to move the last category to the end
ve.ui.MWCategoryWidget.js
* Allow pressing cancel on the meta dialog after moving a category to the end without crashing
Change-Id: I45059bda022639f7942f1799a42ecf8a4d962df0
Break setUpSurface into a bunch of parts that setTimeout() each other.
This breaks the initialization of the editor as a whole up into the
following phases:
0) Receive AJAX response, parse HTML
1) Build linear model data
2) Build DM tree
3) Build CE tree, CE DOM, create surface
4) Initialize surface, some UI initialization
5) Launch sanity check, remaining setup, unlock editor
This is a bit hacky right now, we'll probably want to refactor later.
ve.dm.InternalList.js:
* Allow construction without a dm.Document
ve.ui.Surface.js:
* Allow construction based on a dm.Document
Change-Id: Ibf48a7e85cd7376c8f6205ed165ff57e71e34c35
Since format dropdown tool is the only tool where we actually lose
focus to the documentNode, we must restore focus so that firefox
and chrome will display a cursor after converting content branches.
Bug: 50338
Change-Id: I4059b2688565570e0efc21078035775b7aed49e1
As part of this, put the 'help' icon next to the beta icon, make the
text not greyed-out (now that it's an actual action) and provide the
link.
On hover, all three items get underlined, which is irritating, however.
Bug: 50476
Change-Id: Id65968072b7134f5864bbd96acf34fd0c23fe17c
Previous when the selection was collapsed and the cursor was in a block
slug we'd get a JS error thrown. This fixes that, though it could do with
a refactor in time.
Bug: 47947
Change-Id: I596ea9bc553f563e3dc7689c518c7d1319ac1dc3
To trigger a save rejection from the api, set:
$wgSpamRegex = '/spam/i';
and making an edit adding the word "spam" to a page.
Class changes:
* Rename message system in the save dialog from "warning" to
"message" as it will now contain both warnings and errors.
(css class, class property, method names, ..)
Localisation:
* Remove ugly hardcoded and wikitext-requiring "'''Warning:'''"
prefix from the warning message, instead have a message for the
word "Warning" and re-use this in #showMessage for each message
of type "warning" (bolding applied in code instead of in i18n).
* Rename visualeditor-savedialog-dirtywarning to
visualeditor-savedialog-warning-dirty and remove from
VisualEditorMessagesModule.php as it no longer requires pre-
processing from the server.
Clean up:
* Re-alphabetise the order of some messages.
* Clean up duplication and redundant logic in mw.Target#onSaveError
and mw.ViewPageTarget#onSaveError.
Bug: 50350
Change-Id: I3daf631fb0d62ba88e05aa50c77c9940d61395a0
Assumption about toolbar height never exceeding 88px is not correct. At
least in Chrome, after setting in advanced configuration font size to
"Large" instead of "Medium", the toolbar's height is 110px and part of it
is transparent. Because it is transparent user can see content behind it
when scrolled down, which makes for a not really good experience.
Change-Id: Ibdc2410cc4ab29bfe774961fc062d172386975d4
We currently change <ref name="foo">Foo</ref> ... <ref name="foo">Foo</ref>
to <ref name="foo">Foo</ref> ... <ref name="foo" /> , because know
that the second ref tag isn't canonical and so we blank it.
Instead, we now preserve the contents of all ref tags that come after
the canonical one.
Change-Id: I45a51a879271890fe46c4184f1029f12d27af678
<table>\n\n</table> round-tripped to <table>\n\n\n\n</table> because
we would store '\n\n' in both the innerPre and innerPost fields.
Fixed by not setting innerPost if the element is empty.
Change-Id: I0393bfaf9793fdebc8fff72c8760113fa69bb2bd
The converter wasn't setting .annotations on meta items created to
represent empty annotations, which meant that HTML like
<i>Foo<b></b></i> would end up as <i>Foo</i><b></b> in the linmod.
Change-Id: I13d7d9820beeee1e8c3673e08051361d6c6ac4cd
If you had <meta /><b>Annotated text</b> in a wrapper paragraph,
the converter would swap them and output the linear model equivalent
of <p wrapper><b>Annotated text</b></p><meta />.
This happened because the meta item was queued, and annotations didn't
trigger metadata queue flushes. The fix is to trigger a metadata queue
flush whenever we're about to write something that isn't itself queued.
Change-Id: I168abc0392fbec5503271d1653ee5c38518f857d
Objective:
* Fix issue where async behavior of addTemplate caused templates to be added in the wrong order
Bonus:
* Get rid of special adders for transclusion parts, just construct objects outside and use addPart
Change-Id: Ibe579f033873446376d72d3bd1b9f92d9f361de5
* 'captcha' property from ConfirmEdit API is already exposed
in ApiEdit and ApiVisualEditor through the 'edit' property
in our response data.
* Add parameters 'captchaid' and 'captchaword' to ApiVisualEditor
and mw.ViewPageTarget#getSaveOptions. ApiVisualEditor will
forward these to ApiEdit which forwards them to FancyCaptcha.
* We display the captcha through a saveDialog warning.
Bug: 50356
Change-Id: Ia7d2102cba89d00ec8508e846061023b330ece4f
We need this for captcha's. They'll be added and displayed as
a warning, and cleared like all others. But they shouldn't be
wrapped in a paragraph and aren't delivered as html text but
as a jQuery object (can't serialise to static html since we'll
have to keep a reference to the TextInputWidget and what not).
Change-Id: I3734fbd3c6f0270094db014bbc76d065a73ab918
This parameter was documented as "HTTP status text" but this
was incorrectly copied from other events. The actual code
never passed anything other than null.
Now documenting it as API response data (if available, else null)
Change-Id: I88c8c525f3ca53fd47de634aec469be372884cfc
Parameter config is optional (both in the parent class and here),
this subclass forgot to add the fallback.
Though in a plain subclass the local fallback isn't needed
(parent class takes cares of it), in this case it is needed
since the constructor directly accesses config.multiline etc.
Change-Id: I3f9b73efb6c40a29efa221e1654b1247baaca281
This is done by setting renderHtmlAttributes to false, for both block
and inline transclusions.
Bug: 49925
Change-Id: I975cf1de3b273a5189e6edf7db937c4388fa4fbe
Objectives:
* Use template data for templates added to a transclusion using
the list controls
* Cache template data between different instances of the dialog
* Allow placeholders to be replaced with templates asynchronously
* Save http request and abort pending requests on close
Also:
* Fix breakage of (unused) method getParameterAliases
(dont delete aliases)
* Remove unused method getTemplateSpec.
Bug: 49778
Change-Id: I391e51cb900ef5560455d6f3d4d2a8b99ed2b034
When the cursor is programmatically moved, as we need to do
from time to time, Firefox's internal reference for the appropriate
cursor location on nearby lines is maintained, creating an
unexpected experience.
The fix is to wiggle the selection. This tricks Firefox into thinking
that it adjusted the selection natively.
Bug: 48289
Change-Id: I75e7e4467ec1c3faf344709886b9be5b1423cd3e
Rational is that you wouldn't want to highlight something that is not
visible. And, obtaining offsets for hidden slugs in this case does not
return correct values.
Bug: 49818
Change-Id: Ic2818423c26f5bbb2c056c3c94740d822be4f1b8
Ed & Roan:
Disable editing of references of which we are unable to find the
source (e.g. <ref name="x"> without a target, or when the target is
currently nested in something we don't yet process such as inside a
<references> block or a template).
Timo:
Improve UI to not be a regular focusable node where the inspector just
won't show up but add a not-allowed cursor and explanatory tooltip.
James:
Fix messages to refer to VisualEditor instead of "the" VisualEditor.
Change-Id: Ib2bca092ce13c9187fa8b27ad6a6404cae02aea2
The sanity check converts the linear model back to DOM, then
compares this DOM to the original DOM and rejects the deferred
if they are not equal.
The DOM creation has to be done synchronously (before we unlock
the surface), but the actual comparsion can be (and is) done
asynchronously.
To make the UI flow of the save dialog easier we just keep the
save button on the toolbar itself disabled until the sanity
check is done. Though this should finish before the user starts
editing the document (let alone start saving), we do add a class
to the button to indicate a progress cursor.
To simulate a slow sanity check, set the setTimeout in
startSanityCheck to 5000, load VE, make a change, hover the
button, and see it change from disabled + progress-cursor to
enabled after 5 seconds.
To simulate the sanity check failing, change !== to === in
the first "if" in startSanityCheck.
Bug: 47521
Bug: 50067
Change-Id: I04f71fe8e00c6257fbc953cc9de3323e24709b0f
Objectives:
* Split reference dialog (at least for now) an edit and an insert dialog
* Add reference search widget for selecting an existing source, or
choosing to add a new one
* Abstract reference names, don't allow editing them and generate them
when needed
* When editing groups, move the internal item and update all references
to it
* Resolve name conflicts when moving a reference to a new group by
generating a new list key
Bonus:
* Add getNodeGroups method to internal list
* Add getUniqueListKey method to internal list
* Add destroy functionality to ce.node to release events and references
Bug: 49733
Change-Id: Ib244ff6ad9b4cee1decfd9b9e1d3d4e9cdcfb78c
Objectives:
* Make protected CE nodes only add shields when attached to a surface
* Remove hack in reference list node used to work around this issue
Change-Id: I48d4478558b49037b80b6131e1b2e01dc4f9ec47
* Port the class preservation logic from BlockImage to InlineImage
* Add preservation for unrecognized classes
* Add the logic for collapsing spaces in the class attribute in more
places
Change-Id: I26faad7e00ab2f0a0f5d076552e56b32c692ae74
When you selected the "edit" link by tabbing, the links would expand.
In Chrome, you could then tab to the "edit source" link, but in Firefox,
the links would just contract back and you couldn't reach "edit source".
This happened because in Chrome, the "edit source" link is already
considered to be focused when the blur event happens on the "edit" link.
But in Firefox, this is not the case: the blur fires first, and only
after that does Firefox determine what to focus next.
Fixed by waiting 100ms to contract after blur, just like we do on
mouseleave.
Change-Id: I9a38b629ca5e580003d82a3dca8dbf7564486fa0
After a spirited discussion, localOverlay is getting some children.
* localOverlayBlockers - UI elements that are meant to prevent
interaction with the element. This includes Phantoms and
Highlights.
*localOverlayControls - UI elements that are meant to be
interacted with in order to modify the element. This includes
resize handles and drag marker.
*localOverlayMenus - UI elements that should always position
above all other controls. This includes the context menu.
Bug: 50159
Change-Id: Ic69c2ad275389a31c9fbaf47f3665dcdbb7ac2af
Objectives:
* Make options unable to be selected or highlighted when disabled
* Add default styles for selected state
* Fix line-height for option labels
* Use default cursor when field is disabled
Change-Id: I8a535cd6b259b092c2abce27ddb10882cdde6cf5
Objective:
* Make the escape key close dialogs, like pressing the X button
* Auto-focus the iframe on window open
Bonus:
* Add ESCAPE and SHIFT to ve.Keys and use instead of hardcoding numbers
* Use ve.Keys in some other places too
Bug: 49809
Change-Id: Ibf1fce5e24efcd83d9e1465c3cdaac24ff3fb45d
Omitting the offset causes insertMeta() to automatically use the
end offset, so we don't need to compute it in two places. The two
computations were also slightly different.
Change-Id: I55543fdd113a6a986899c093733191df948acb2e
The save dialog has z-index: 3;, which succeeds in overlaying it on the
toolbar in its normal position, but fails once the toolbar starts
floating, because the floating toolbar has z-index: 100;
In practice this meant that if you were scrolled down and the toolbar
was floating, you could open the save dialog just fine, but you
couldn't close it because its controls were below rather than on top of
the toolbar.
Hacked around this by detecting the floating-ness in the toolbarPosition
handler and setting a class on the toolbar tracker accordingly.
There may be a more elegant way to fix this; an actual UI engineer
should figure that out, not me :)
Bug: 50324
Change-Id: I8c6ab1026705d00baa20f115255d0d7e74ee72bf
This ensures that attributes and properties that are supposed to be
stripped on copypaste are actually stripped.
Bug: 49307
Change-Id: I8c90f4a0b33acba6eea3180cc077f8dc440e6e7b
Follows-up I4b9c47fd65a700a:
* Remove unused closingBracketSymbol.
* Moving veSectionEditUri and sectionEditUri inline as it is
only used once. This will allow it to be easily dereferenced
(instead of likely staying in memory due to being claimed by
remaining closures "expand" and "expandSoon" which will
continue to claim and enjoy access to this scope.
Also changed it to use the attr() callback so that we
don't access `$editLink.attr( 'href' )` twice.
* Add various comments explaining this non-obvious approach.
Though it makes sense (eventually) and more elaborate details
are in the original commitmsg, that is no excuse for a 100 line
function without a single comment, especially with all these
things going on.
* Use "shrink" instead of "contract" (seems more common in
context of web UI elements and less ambgiuous in English).
* Remove redundant toString in `new mw.Uri( veEditUri )`.
Aside from redundant, it also had more overhead (serialising
uri object to string and re-parsing). It revoked its ability
to make an efficient clone by taking the mw.Uri object as input
as opposed to having to parse it again. mw.Uri#clone does the
same internally.
* Avoid variable names like $this, that or self. Using a more
descriptive name instead.
* Re-use $closingBracket instead of using last() 3 times which
makes 3 clones of the jQuery object with the same last element
in it.
* Refactor odd swapping of closingBracket-hide and
middleBracket-clone-show.
It now keeps the original opening/edit/closing in tact and adds
devider/hidden/edit-source/outerClosing at the end.
Change-Id: I5f093f2927b769fed0c6d1a40f99e73f9b653b9a
We need to normalise titles so 'user:foo_bar' == 'User:Foo bar', and
we also need to some HTML attribute removal as links from Parsoid
will have href and rel set (again, this should be fixed in by Parsoid
when the do the merging at their end).
Bug: 49985
Change-Id: I5fb5bfc69c344ca4ce4803d7b6116074648a8d7e
They are only run in the MW test runner, where the MW dependencies
are available. Create a ve.test namespace for storing shared
test runners.
Change-Id: I079cb18b1c73614d25a12c5d6afcf0700469e52e
Objectives:
* Associate models with tools, rather than dialogs and inspectors
* Move tool/model association utilities to ve.ui.ToolFactory
* Obliterate the view registry
Notes:
The only special case for leaving modelClasses definitions in place is
for the linkInspector. It uses these for selection expansion.
Because tools can now override the static canEditModel method, we can
dynamically evaluate a model, rather than be restricted to only
comparing classes. This will be useful for disabling editors for models
that are for some reason incomplete or otherwise broken and cannot be
safely edited.
Change-Id: I7adf254990112d90f1f808593a9111afc7a116b5
Because of the z-indexes of major elements of the mono book
skin, and because the overlay containers are appended to the
body, the overlays can't be positioned between the surface and
the toolbar. Before this fix, the overlays are appearing beneath
the surface. This fix will retain proper positioning of the overlays
between the surface and the toolbar for Vector, and will overlay
everything in Monobook.
Later, we will have the overlay container more tightly integrated
with the surface to avoid this stacking problem.
Change-Id: Ibb1553099cc1e35e6a0928a99b584885508ca5b6
Convenient way to check if an object is an instance of one or more classes.
This should probably be moved to oo.js.
Change-Id: I2be55d9501cf62fde006281f1cf4813095b39b41
Re-label the headings in the MWFormatDropdownTool with MW-specific
headings per bug 43334, reduce the size of the dropdown's contents a
little to make the headings less vibrant, and move the H1 option to the
end of the list to further discourage its use.
Note that there are some issues with the underlying structure here and our
ability to split the repos into VE-core and VE-MW will need this to be
refactored.
Bug: 43334
Change-Id: I5a58b4dcebd6ceae0ffcd24f663429f25bdc3db9
Objective:
* Provide quick access to section edit links for both source and visual
editing
Story:
After using this prototype of my mockup, I realized how bad my mockup
was. Hooray for prototyping!
The issues were twofold:
1. Adding down-arrows to the edit links made the page look worse, and
was sure to incite rage and panic throughout the community.
2. The menu was just too heavy. Matmarex made an observation early on
after seeing it, that it wasn't very "Vector", and while I agreed, at
the time I didn't have any better ideas.
Thank you to Matma Rex for prototyping this feature. Aparently there was
also a previous attempt (I13bbb9549). We appreciate your help.
The new design is simple.
* Section edit links look normal
* On hover or focus, the edit source link also appears next to it
To make the two links look separate, we needed a divider. To make the
divider look good we needed to add space around it. To balance the
space, we needed to add space to the brackets. To avoid changing the
view, we needed to only add space to the brackets on hover. To avoid
the text moving around, we needed to make the brackets move away from
the text, rather than the text move away from the brackets. To make
this change smooth, we needed to use transitions. To make the links not
force the heading to wrap in one state but not the other, we needed to
reserve the space, using visibility rather than display. To reserve the
space we had to use closing brackets as spacers, hiding/showing one of
them on mouse enter/leave and leaving the other always hidden. To avoid
the right bracket from getting clipped by the edge of the screen when in
expanded mode, we needed to add a bit of padding to the right side of the
section edit link top level span. To prevent the extra links from
flashing as you move your mouse down the page, we needed to wait 100ms
before showing or hiding them due to mouse enter/leave.
We use negative margins to move the brackets. Animation implemented
using CSS transitions. We bring the pipe divider in from the core
'pipe-separator' message.
To style the brackets independently we needed to wrap them in spans and
add classes to them. Change Id27555c6 in core will make the wrapping
unnecessary, but the two should still get along just fine.
Interestingly, we needed to @noflip the bracket styles because CSS
Janus flipping is triggered on UI language, but the brackets need to be
styled according to the content language.
Changes:
ve.init.mw.ViewPageTarget.css
* Add styles for extra section edit link components
ve.init.mw.ViewPageTarget.js
* Add edit source link, and make it visible when the mouse is over the
heading or either section edit link is focused
*.php
* Links to new messages
Bug: 48429
Change-Id: I4b9c47fd65a700a81c880144247fec524edff7e5
Use the 'all' mode of SurfaceFragment#getAnnotations to correctly
handle the selections which include linked and non-linked text
in the LinkInspector.
Bug: 50208
Change-Id: I1cab7f3cc4fc9589eced01ad38c59fe5b9622a57
This bug caused all references containing complex content (e.g. links
or templates) to be dirty-DOMed and reformatted by Parsoid.
ve.dm.MWReferenceNode.js:
* Parse the original body.html and check if it's semantically equal to
the new value. If so, don't set it.
ve.dm.Converter.js:
* .normalize() the converter output to remove empty text nodes and
merge adjacent text nodes
ve.dm.example.js:
* Update reused reference test to have body.html absent, not empty
* Add a link to one of the reference tests so this bug is triggered
** The link's attributes are ordered specifically so that toDomElements
will reorder them, at least in Chrome (may behave differently in
other browsers)
** This test fails without this fix in place
Change-Id: Idc091a14422fbb117a3d06fc6bb9497768086fc3
When the toolbar animates, the position of the surface changes.
ProtectedNode now listens for the toolbarPosition event and
repositions the phantoms.
I also tried moving the localOverlay to be a sibling of the surface
(instead of child of body), but the same problems persisted.
Bug: 49853
Change-Id: I8ae1c8f66c6083b4ffee7107a78a298126e1064c
Local overlay z-index is positioning overlay items (phantoms, etc)
above the floated toolbar. The overlay container is a sibling to the
toolbar's parent and is later in the DOM. z-indexing is
unnecessary and has been removed.
Bug: 50096
Change-Id: Ib46f66b1d1bea36d761f793e9966e09db6e675ca
It was discovered that nested iframes inside firefox do not fire load
events: http://osdir.com/ml/dev-apps-firefox/2010-03/msg00134.html
Since we no longer care when the iframe is loaded, changing the frame
onLoad method name to load and manually calling it was sufficient.
Bug: 50115
Change-Id: I65c0c404a2a384566701b1c6b1bb978e0b99c90d
Broke reference insertion because it removed
newFromNodeReplacement(), which is still in use.
This reverts commit 1765e39b40
Change-Id: I043997715474ad4850329ff903eb7a8c61c8b453
HTML like <td><span>Foo<!--bar--></span></td> would lose the comment
when converting to linmod, and so we'd drop the comment when converting
back to HTML.
This happened because we were queueing up meta items to possibly move
them and their whitespace out of the wrapper paragraph even if we were
inside of an annotation and there was no way we could move things out
at all. The fix is simply to detect these case and not queue up meta
items in that case.
[12:11] marktraceur Let the comments hit the floor, let the comments hit the floor
[12:11] marktraceur (to the tune of the Drowning Pool "Let the Bodies Hit the Floor")
Bug: 50071
Change-Id: I5a844e2e0655bc1db152b3805718324ca0bb04df
We calculate an overlap offset for when the popup widget has been moved
to avoid going out of the viewport, but that offset is only applied to the
body and not the callout.
Change-Id: Ib9d08d60ccfa6562378640c526faf1e8363abbba
The UI still needs a simplified version of annotation comparison
so it can clear/set annotations, whereas for serailisation we need
to compare selected HTML attributes (and once Parsoid is fixed,
compare all attributes).
Change-Id: I1a5521a3f51ea373a21cfbc9a007a2381c12f4fe
This was a big gaping DOM diff that's been in there since the beginning,
but which we only found today.
Change-Id: Ie4e791b212ddd3210bc1d5c11308452bb57ae3ce
There is still a problem that those images would get inspector for opening media edit dialog
displayed, and that dialog let's edit caption, which does not make sense in this case, however
that should be a separated bug.
Change-Id: I24491979bc7a27d2380924e48cb677b01e9f8f9b
Remove enableAboutGrouping from GCNode and move it to the subclasses
so we can have GCNodes without grouping.
Having matchTagNames there doesn't make sense now that GCNode is a
mixin rather than a base class.
Also removed unused .static.generatedContent property.
Change-Id: I2b5350e70b7829cf009c752d066919e5a3593db0
Also checks for subheadings, i.e. we keep moving forward one sibling
as long as the next sibling is a heading of higher order than the
previous one. If the heading is of the same or lower order then we
assume there is an empty section, so we just stop.
Bug: 49563
Change-Id: I6fcc463aeeae40278f5caf04ae251c8ecfe65472
Build the new set of classes as an array (avoiding the leading space),
and only set it if it differs from what was there before. This avoids
reordering the class attribute.
Change-Id: I863ec8de2e7637f76690c4e0b7db5e153fb7e408
Prevents changing of DOM attributes on unmodified transclusions
as they can clash in the store.
Also use 'original' over 'orig' consistently.
Bug: 50079
Change-Id: Ib13bb206c49b1f5b186e40632a5c109def0f042e
Due to some sort of mysterious race condition (probably deferred reflow)
calling this.$menu.show() set style="" rather than style="display:block;"
because Firefox believed the parent was still hidden.
Work around this by removing display:none; from the context's CSS rules.
We don't need it because we always manually .hide() these things anyway.
Bug: 50073
Change-Id: I29e210aba9a6265d8364ff8ae49408cb4c2428b9
Previously we were just passing a slice of the visible data, instead
of using getFullData.
Bug: 50059
Change-Id: I3bbea49132ef4a720a147ba9b170c39a0c00f711
Previously we would reserialize the JSON blob in data-mw even if we
didn't change it, which potentially reordered keys and caused a DOM
diff.
Bug: 50066
Change-Id: If0a5bcc67d3a172de0e8839cfda11efacfbf36ff
We were previously ignoring this data leading to situations
where a template that can be invoked like {{foo|1=bar}} and
{{foo|thing=bar}} (where the template data documents param
thing with alias '1') will show up in the editor with no
parameter information or label for the 1= call, but will show
up for the thing= call.
Now they are properly aliases so both will appear the same
in the editor dialog.
Bug: 49838
Change-Id: I37ec0e152df905844ac58ed1834fca29dccb4eec
Fixes (follows-up I1b48ef5240, I6daff5c596):
* Invalid html passed to jQuery constructor.
* Use prop() instead of attr() for boolean values.
* Use append() instead of html() when appending nodes instead
of parsing html.
* Rename shadowed variable name clash 'mw' to 'mwData'.
* Fix odd construction where we parse '{}' to create an empty
object.
* Have ve.ce.MWReferenceListNode#update perform changes off-document
in a detached tree.
* Fix deep property access that can fail. mwData is set to
either JSON parse of data-mw attr or empty object.
Accessing mwData.attrs.group needs to be guarded by whether
mw.attrs is indeed set.
* Have `mw` and `about` attribtue in references list roundtrip
(especially mw which can data we aren't editing/re-creating).
* Add missing 'refGroup' property to MWReferenceListNode's
data element (similar to what MWReferenceNode already has).
Change-Id: I67e4f378ccd04e97361d8e58ae57db5353075756
* Use plain text rather than HTML in TextNode
** Bypasses HTML parsing, and doesn't cause regeneration of nodes like
appending to .innerHTML does
** We were only using HTML so we could use entities, so replace those
with \uNNNN sequences
* Use native DOM functionality rather than jQuery
* Inline flushBuffer()
Change-Id: I7c6376b55cc0f1420a01a77b365b073fe1636263
Better for performance (lower level, #get calls #toArray when
called without arguments) and more semantically correct.
Change-Id: I8e615674e51fd148367fd397bd169fa203a240ad
HTML5's drag and drop has a ton of benefits and a couple of
limitations. To achieve a native drag marker, an image tag helper
is used to indicate the size of the relocatable node. Chrome
shades the marker gray natively, Firefox is styled to match.
Change-Id: I755b698a3d968cc7e6ff125109d68ac83fd8a8a2
jQuery's implementation does some sort of traversal, so on large pages
it took 300-400ms just to focus the document node.
Change-Id: I0de95d8800a1aedeb19b27d204dfd8fd08cec533
This is a waste of time, because we hide it anyway. This took
quite a bit of time for long articles.
Change-Id: I1bcfd5b908c89dc2e08d5520e8870b145177ca10
* changes:
Fix incorrectly written CSS that was affecting all table cells on the page instead of cells in mwMetaDialog only
Fix for rendering block images
Follows-up I99acbd1699:
* "Parent method" comment
* Remove redundant slice() call to convert arguments to array,
native JavaScript methods that take array-like arguments such
as Function#apply and Array#slice are both compatible with
the Arguments objects, no need to convert it. Most invocations
already did this right but a few were recently introduced again.
* Removed silly "Document dialog." descriptions.
* Removed a few redundant "@method" tags in the near vicinity
of code I changed.
* Fixed function invocation to be either on one line or
one parameter per line. Having all arguments on one line
but the name + "(" looks confusing as it suggest there
is only 1 parameter. Same as object literals:
so:
{ foo: 1, bar }
or:
{
foo: 1,
bar: 2,
}
not:
{
foo: 1, bar: 2
}
Change-Id: I379bc2b32603bcf90aba9b4cd0112e7f027d070e
* Optimise initialisation by building the tree from the bottom up
and applying the class last.
* Actually document the dialog. Documenting it as "Document dialog"
is useless and confusing. If there is no documentation yet, leave it
empty so it is included in jsduck's optional "no_doc" warning list.
* Remove the redundant "Call" prefix in the parent method description,
we don't do this else where either.
Change-Id: I99acbd1699b2c99a2c9b58f34dc1e07bea58203d
Objectives:
* Introduce new icon for a reference
* Repurpose existing reference icon for references block
Change-Id: I1bfcdbeda3b9730d62d6b264ef72921e93623f7e
At the moment we create an alien meta item if an annotation
wraps contents of zero length, however we also need to handle
nested empty annotations, i.e. where the contents is other
empty annotations (i.e. alien meta items).
As alien meta items simply preserve DOM elements we don't
mind what type of alien meta items they are, so this also fixes
the case of an annotation wrapping another alien meta
e.g. a comment or <meta> tag.
Bug: 48605
Change-Id: I98d88b341efcd6384ee1fda5d6d7e1e1acb107aa
Previously pageExists thought matchingPages was an indexed object when
in fact it was an array. This didn't manifest as a bug as we subsequently
check the normalised URL in the correct manner (using indexOf), but it
did remove the optimisation of not running mw.Title if there was an exact match.
Change-Id: Ic616cbfa0d7ed5447e032dd1f772779db2dc19e5
Until this is fixed by Parsoid, this is a dirty hack that compares
HTML attributes on annotations, excluding data-parsoid. Obviously
we shouldn't have any parsoid specific code so this should be removed
as soon as it is fixed properly.
Bug: 48194
Change-Id: Ibb18b4f653c664e8ab7876498dc8395d878f7aaa
Was previously doing the text slicing before the cluster split
using substring, instead of after using splice.
Bug: 49233
Change-Id: If9c0860e2fe7b01eb0b27aae67c671062799a0fa
The API gives us normalisation information about all titles,
including titles of which it couldn't provide any template data.
Spec storage should only have properties with valid values
(has property means to have a valid value, like database). The
absence of this check was causing it to turn this:
{
pages: {
'Template:Unsigned': { params: { .. } }
},
normalised: [
{ from: 'foo bar', to: 'Foo_bar' },
{ from: 'template: unsigned', to: 'Template:Unsigned' }
]
}
into:
{
pages: {
'foo bar': undefined
},
normalised: [
{ from: 'foo bar', to: 'Foo_bar' }
]
}
Bug: 49493
Change-Id: I0b661744d0388345561897c9631f15e691737031
Previous condition was put in as part of I9ed6f9fb as a
temporary fix for bug 42842. However this code was causing
the popup widget to not appear for items at the start of
the document. It appears the bug it was put in to fix
has since been fixed elsewhere.
Bug: 49421
Change-Id: I8bcdea0f7a6951216bf8368865d23ef6246880ea
The transclusion dialog lets you completely delete all templates
within it. If this happens we should delete the node completely, or
not insert it if it's a new instance.
Bug: 49831
Change-Id: Ic5d4fbf73f3d1be9cf877bc8b98096de9e98991a
This function builds a transaction that takes a document slice and
inserts it back into the document it came from, applying any changes
that were made.
This makes editing document slices simple:
slicedDoc = doc.getDocumentSlice( captionNode );
// Edit slicedDoc using a surface
tx = ve.dm.Transaction.newFromDocumentInsertion( doc, captionNode, slicedDoc );
surface.change( tx );
Specifically, newFromDocumentInsertion replaces the node's contents
with the document's contents (meaning any changes made to the node in
the meantime are lost). It also merges the stores internal lists
of the two documents and remaps indexes accordingly. This means editing
of references inside of references is supported.
This functionality is not specific to slices, and can also be used to
safely insert data from a paste buffer, with internal list data being
transplanted correctly.
ve.dm.MetaLinearData:
* Make merge( [ undefined, undefined, ... ] ) return undefined rather
than [].
ve.dm.Document:
* In getDocumentSlice, store a pointer to the original dm.Document in
the new one, and also store the length of the internal list. This
allows us to figure out which internal list items the two documents
have in common when we insert the modified slice back into the main
document.
* In getMetadataReplace, optionally take the inserted metadata as a
parameter, to allow for operations that insert both data and metadata.
Per Ed's review, rewrite this function to return null rather than {}
if no metadata needs to be replaced.
ve.dm.InternalList:
* Add method to merge two internal lists
ve.dm.Transaction:
* Remove newFromNodeReplacement and replace it with newFromDocumentInsertion.
* In pushReplace, optionally take the inserted metadata as a parameter.
Change-Id: I786ee7bad796aa54bc242993b4de3ad18ad0773e
Without these properties set, the layout is collapsed into a 0x0px square due to the
parent having overflow-x: auto; and no width/height set.
Change-Id: I77aef8afa3327b11f453196badcdfad3f51f7c3d
The check for the toolbar not overlapping the last branch node
'fixes' an edge case where the user has scrolled too far, and
the viewport is shorter than the last branch. A more common
problem caused by this is that if the last branch is very tall
the toolbar disappears whenever you scroll past it.
Bug: 48662
Change-Id: I1c7662f2b6f1ced6f80dec16c6ed69a8cc0c06c8
Removed inherited (and often wrong) documentation, replaced direct
usage of onApplyButtonClick with more standard onClose handling,
and one case of calling the wrong parent method.
Change-Id: I86ed16860e996b42c141a6499eefb9084d759a72
"Template" parts should be labeled with a normalized version of what you would actually write inside a set of curly brackets.
Change-Id: I4b9c541a01da86cb078d38012e5a9fead2fa233c
Objectives:
* Clear this.titles when resetting, otherwise results are shown once and
then blacklisted forever.
* Attach done() before always(), so done() gets called first and not
after always() has dereferenced the request object.
* Remove use of imaginary variables like this.loading, which were never
set, and were referencing even more imaginary methods on other imaginary
objects like this.request.abort()...
Change-Id: Ifbe74346f68b6a91f425b56a2696fd1e46b61e02
If the query is cleared, we should still clear all items and abort any
loading that may be going on. We should still not try and load results
for an empty query though.
Change-Id: If23c6e0469247b1c1c47ab85b9a5a010e8a4fe26
Only remove the namespace prefix if it's in the template namespace,
and prepend a colon for main namespace titles.
Change-Id: Ie910bae172274e36c7c9a675a631a0876ff13ed1
1. It's redundant and ugly.
2. We already have an icon.
3. It's unnatural for anyone used to working with templates.
Change-Id: I821e124158a14679e411324620dc9d7c40252383
Objectives:
* Allow adding content or templates within a transclusion
* Add template placeholder to model and view which resolves to nothing
when saving (thus disappearing if not resolved)
Changes:
*.php
* Add links to new messages and files
ve.ui.OutlineControlsWidget.js
* Organize controls into "adders" and "movers"
ve.ui.Widget.css
* Add styles for adders/movers sections of outline controls
* Make adders appear on hover to reduce clutter
ve.ui.Icons-*.css
* Add icon for "add item"
ve.ui.Dialog.css
* Add styles for add template fieldsets
* Make placeholder items in the outline italic
ve.ui.PagedDialog.js
* Pass adders config to outline controls
ve.ui.MWTransclusionDialog.js
* Add support for adding content and templates
ve.dm.MWTransclusionModel.js
* Add addPlaceholder method
ve.dm.MWTemplatePlaceholderModel.js
* New class, pretty much an empty part
* Using this makes the UI much easier to work with - no need to special
case the outline control for new items
* Because it's not supported specifically in
ve.dm.MWTransclusionModel.getPlainObject, it produces nothing and goes
away naturally on apply
Change-Id: I3478560fb53ba2ccd3fb26bafb6a61e6415565eb
* Fix bug in MWTransclusionNode sometimes passing a NodeList rather
than an array (caused an error because NodeLists can't be .map()ped)
* Use copyDomElements in ce.AlienNode as well
Change-Id: I11c6483d3c062047be80f75a3fac03f13bff9662
* Provide a utility for copying an array of DOM elements into a
different document
* Copy the DOM elements returned in toDomElements(), otherwise weird
issues arise when the same data is converted to DOM twice
Change-Id: Ie927420624f0d4af0692e18d1bc6f952c8013d61
In aaa5ad2 we introduced a feature test. However though we do require
ES5, we don't (yet) use strict mode. It is an easy and solid one to
feature test, but it the unfortunate reality is that IE9 does not fully
support ES5. Strict mode is among the features it didn't (correctly)
implement. As a result commit aaa5ad2 locked out IE9 completely.
Replacing it with a more explicit (and harder to maintain) test that
does pass in IE9. We now need to be very careful about using ES5
features in our code. It has to be in this list here to avoid breaking
in a browser that doesn't support it.
Change-Id: I77b623e15a0791bfaa2cd835e43107cefb3c25f5
Objectives:
* Make get$$ work with empty jQuery selections
* Reset the $items selection when clearing a group
Changes:
ve.Element.js
* Add existing context property to possible solutions
ve.GroupElement.js
* Reset this.$items after detaching them
Change-Id: I73b46649d7ecd5045941086c4e1bb3241c8f4ef3
Added return boolean to ve.dm.Surface#breakpoint to indicate if
and transactions were pushed to the big stack. Use this value in
MWMetaDialog to see if we need to undo on close.
Added tests for ve.dm.Surface#breakpoint.
Bug: 49630
Change-Id: Ieb2e9e361afe057af93c4d374acc85df58bfb4c3
Slightly more efficient way of seeing if the template has changed,
and results in a cleaner set of data element attributes.
Change-Id: I1507520005bfb8a88bfa6038dac5c3b15506425d
matchingPages contains titles with spaces, but .toString() returns titles
with underscores, so normalizing with .toString() is incorrect. Use
.getPrefixedText() instead.
This caused a bug in the link dialog where, if the link was an internal
link pointing to an existing page with a space in its name, the "New page"
section would be displayed but would be empty, and the link's target would
be displayed in the "Existing page" section but colored red.
Change-Id: I59d8b87b996d80da1b25124b942bd2b72ef50845
The difference is that .children excludes non-element nodes (text nodes
and comment nodes). These can't have attributes anyway, so there's
nothing lost by skipping them, and this way we avoid bugs where a
text node split causes the indexes to be off.
(Text node splits are probably due to an interaction between whitespace
preservation and paragraph unwrapping, and aren't necessarily bad. We
just shouldn't rely on indexes into .childNodes)
Change-Id: I905a50e1c299ebafcbd4eaa0f938b06a1b5849ff
