loadFail can result in a retry, in which case it isn't
approraite to reject this promise. Also many of the code
paths call 'tryTeardown' which itself will reject the promise.
Bug: T238332
Change-Id: I366662847304d8ecf79d5899b2804dded67ee999
* Add a postWikitext method and split out postContent
from postHtml
* Move saveSuccess handling into postContent promise
* Connect promise directly to saveComplete instead
* Pass whole response.visualeditoredit object, instead
of splitting into variadic arguments for saveComplete.
* [DEPRECATION] Make serialize return the postHtml promise
and deprecate passing a callback.
Change-Id: I905737515578000b2b87214c92e8b9fe9e82f6b7
New changes:
be8235e82 Localisation updates from https://translatewiki.net.
6f43a6c8d Remove MW-specific code for setting up section editing
Local changes:
* Bring in section editing logic from VE core
Sets attachedRoot iff there is only one SectionNode in the whole document.
Change-Id: I15b5ebf3848482ef6df6d19114d26a1b1d4a3b13
If the paction=serializeforcache request fails, we were erroneously
converting it to a successful result with no value, which later causes
an exception, because since 381b58585c
other code expects the result to an an object.
The bug was introduced in 2015 in 07001001be,
but until that recent change it would only cause a 'badcachekey'
error, which was handled correctly later.
Change-Id: Ie1ffc8c3e616a7d296f2186fb17eaf039971a44f
With this patch, the toolbar slides into place place nicely after
scrolling again, but it still occasionally flickers during the scroll.
* window.innerHeight is now smaller or something, and we have to
twiddle the scroll position by a larger value.
* document.body.scrollTop no longer works for setting or getting the
scroll position, so use different methods.
* requestAnimationFrame() now generates an insufficient delay to make
scrolling happen, so use setTimeout() instead. We actually have to
add a nonzero delay there, otherwise the toolbar sometimes doesn't
animate like it should or flashes in random places on the screen.
This delay is bad because the user can't start scrolling again
during that time, but I think we can live with that.
Bug: T233470
Change-Id: I6c40ee8ce5994e12eadb085bbffd120ef160d4ee
New changes:
4af3f84f7 Mark surface as "showAsDeactivated" when opening a window
79eb0e4e5 ve.ce.Surface: Guard against focusing a un-initialized surface
4124c275e [BREAKING CHANGE] ve.ui.TargetWidget: Construct a real target inside the widget
Local changes:
* Use new target widget
* Remove calls to deprecated methods
* 'surfaceReady' event was upstreamed
Bug: T236400
Change-Id: I765d657c172d96c3b2e2ae5998083e4926a31f15
As of commit c65ed0e7a8ac5f32a3a6e4cb2760dae03e4fca22 in MobileFrontend,
it uses errorformat=html queries (the same as we do), so we no longer
need to massage the responses to make it happy. The same commit also
turned parseSaveError() into a no-op, so we can remove that as well.
Change-Id: I4f0109ce120ebf94e5709d47d775a8178ce216fa
Previously, the ve-mw/init/ directory contained two kinds of files:
those that were used when initializing VE, and those that may be
loaded even if VE is not going to be initialized at all. The latter
kind must not use the `ve` global variable.
After moving those files to ve-mw/preinit/ we can enforce this with
.eslintrc.json in that directory. This would have prevented T228684.
(Technically they merely must not use `ve.init`, and may use `ve`,
but that's harder to enforce. We should instead move the few non-init
methods out of `ve`: now, track, trackSubscribe, trackSubscribeAll).
Also, group some files under ve-mw/init/: targets/ now (only)
contains ve.init.mw.Target and its subclasses, apiresponsecache/
now contains ve.init.mw.ApiResponseCache and its subclasses.
Bug: T228684
Change-Id: I945249a27f6a0fa10a432d5c5dc57bc7e0461fd8
In order to get the desired padding, the `em` values were pre-multiplied
by the parent font-size, which was assumed to be `0.875em` for normal
surfaces and `(13.3333/16)em` for wikitext surfaces. (That should have
been `(13/16em)`, by the way, but that's a negligible difference.)
Unfortunately, the font-size for wikitext surfaces is actually `13px`.
Unlike `em`, `px` values are not affected by custom browser font-size,
so when one was set, the final padding was all out of whack.
Use `rem` units instead to respect the custom browser font-size,
without getting problems due to the parent element font-size.
We don't need separate rules for .ve-ui-mwWikitextSurface now.
Bug: T222217
Change-Id: Ib7ffbf09d5aa23fddb894aa3b081ec993ddcee2d
The call to selectFirstContentOffset() below would re-activate the
surface and (attempt to) show the keyboard if the surface was
deactivated, but not shown as deactivated.
This might have worked correctly by accident before
I39fe44eee8eab7129340bcff796b6b9b3a59a398 in VE core.
Change-Id: I500309cc0aa8cd794175ae683a17c2614fd58cc9
Some post-save scrolling would try to access the view before the
handlers were cleaned up.
Bug: T232347
Change-Id: I30433ef027c52d541351972f8ebb09fe6d45e436
As of I6c043e039fbef62a56f475b0dc365e171ab7bf59, this method is never
called. It was previously used to update the size of the toolbar when
context menus were displayed inside it, but they are now displayed
elsewhere.
Change-Id: I53030de1203a7f0d75780ae796bbb10082d5ef7a
As of I2f2495ab6c10116a6660f4361e49272cb95b988a in MobileFrontend,
the overlay never uses the default spinner.
Change-Id: Ie0aa624e33a5bd21fc20459697cca175d9de5606
The good thing is that every time our CSS overrides get less crazy.
See 75ff121b29 for the last time.
Change-Id: I9d81aff6a24ec28850563e00206e21c4a6593d2e
Tapping the toolbar save button while the save dialog is open triggers a save
because of the accesskey. It shouldn't save on a double-tap, because that's
easy to accidentally do / trigger on a slow device.
Bug: T230816
Depends-On: I4c3afce9d57c9bca737272b40b9a4862b5794bac
Change-Id: I1925b1b97de6a811f73196b616ec09a2c30c336f
Depends on MediaWiki change I48d4bb3f, which adds the 'arrayParams'
option to handle explicitly indexed array parameters like
`&preloadparams[0]=a&preloadparams[1]=b`. Previously we only handled
implicit indexes like `&preloadparams[]=a&preloadparams[]=b`.
Bug: T231382
Depends-On: I48d4bb3fdf0ea7f5eb133c59bf63651ba356fc42
Change-Id: I8c899bce1b19fa286bd385f89e102a4b87db4db3
New changes:
77076f828 LinkAnnotationInspector: add a "label" field on mobile
Local changes:
* Updates for mobile link label editing
Bug: T229431
Change-Id: Ib0489f6f59b228ebc4a20f7a0a515be938a8f6d3
When saving fails for a reason we don't handle explicitly, the error
message will have HTML formatting and will respect any on-wiki
overridden messages, rather than being plain text generic message.
Extensions providing custom SaveErrorHandlers may need to be updated.
The only one in Gerrit that requires a fix is TitleBlacklist:
Ibeae79c95557a7af699716c9d921f34c310bee6d.
* Remove handling for errors returned in .visualeditoredit.edit.info
rather than .errors (.error in old format). AFAIK this is only used
by some extensions, it is probably incorrect to do (T229539) and all
extensions I know of that do this (AbuseFilter, SpamBlacklist,
ConfirmEdit) have custom SaveErrorHandlers.
* Remove custom error messages for 'readonly' (identical to API
response) and for 'hookaborted' (very unhelpful and there is a
chance that the API response is better, if the extension causing
this error generates any error message).
* Add a silly shim for MobileFrontend integration, because we allow it
to handle error responses, and it expects them in the old format.
This is probably subtly wrong in many ways, but MobileFrontend code
only uses this for logging, so it shouldn't explode. In the future
we will hopefully change it to use errorformat=html (T228897#5366960).
Bug: T229532
Change-Id: I3b9c4fefc0869ef7999c21cef754434febd852ec
A @method annotation is only necessary when the docblock is not
directly followed by a function declaration (in which case JSDuck
assumes it documents a property), e.g. when defining an abstract
function or referencing a function from another library.
I verified that JSDuck generates exactly the same output before and
after this change (docs/data-<hash>.js files are identical).
Change-Id: I7edf51a8560ab9978b42800ab1026f0b5555c3bf
The .oo-ui-popupToolGroup-handle styles only need to apply to the
editTools toolbar, where they override OOUI styles so that our
"stretched" toolbar works; they don't need to apply to the pageTools
toolbar, which contains just the editor switcher, and where they make
it look different from the MF wikitext editor toolbar for no reason.
Change-Id: I21315b34be0a7c3938f84ada720ff5754d953879
We were incorrectly always adding action=edit to the URL in that case.
The condition was always passing because `this.section` is `null` when
unset, but `this.currentUri.query.section` is `undefined` when unset.
This is a similar fix/bug to a68cc38b22.
Bug: T209163
Change-Id: Ic80ac377b763aea53678c4209ba6b3a6ba2996c9
New changes:
854a1fa2c Distinguish active link styling
Local changes:
* Pull through active link styling
Bug: T228220
Change-Id: I925f88d32a514a749b96f501a211003bc4c924f0
Since we're inside the Target instance, `ve.init.target` refers to
this object. Some of the code I'm changing even uses `this` instead
of `ve.init.target` on the next or previous line.
Most of the mistakes are a result of mass search-and-replace changes
(478b0bcb, e1a887cc), or moving the code from other classes (d294006d).
But I can't explain the "ve.init.target.getSurface().getDom()" line,
it would be good to figure out why it was this way before we change it.
Change-Id: I0d7c6a48369242d4c99620fcd775ab537420d84a
New changes:
a06204317 Fix TableNode unit test getOffsetFromCoords failure on Firefox
dfe0eb025 Refactor mobile context logic into ve.ui.MobileContext
Local changes:
* Pull through for edit cards refactor
Bug: T227532
Bug: T228767
Change-Id: I6c043e039fbef62a56f475b0dc365e171ab7bf59
New changes:
1a7460058 Remove ve.newMobileContext feature flag
Local changes:
* Remove ve.newMobileContext feature flag
Change-Id: Ia8def997b7cba4623866080752b06068d2118cc3
We also show this dialog on the old wikitext editor, where
ve.init.target is not set, because the relevant code is not loaded.
Follow-up to 478b0bcbb9.
Similar to e88cd81f94, which fixed the
same issue in a different file.
I checked all uses of ve.init.target in files under ve-mw/init/ and I
think this was the only remaining mistake.
Bug: T228684
Change-Id: I15551870cdb01d570e24ba9668e67330b8072e01
Our overlay could be higher than the viewport, allowing the viewport
to scroll again.
Bug: T212159
Change-Id: I1e97d1963b214fc7673c33ae6c14ab7b0b80f31d
Depends lightly on a patch to WikiEditor, which will hook up the logging there
for the case where the switch is happening from WikiEditor to VisualEditor on
the same pageload.
Bug: T221191
Change-Id: Ibafec77b2eabd3b3b3767472b7b5a40e3312bf18
Way back in 2c24efae in 2015 this got disconnected, which resulted in
saveComplete logging not having the expected updated revision_id.
Bug: T226847
Change-Id: I6dc14bb4b2dacedbe27493a97fa25c9b0542818c
If the translations of save/publish button messages like
'visualeditor-savedialog-label-publish-short-start' contain spaces
(e.g. in Bengali 'bn'), the button on mobile would wrap over two
or more lines, due to weird styles we have for the mobile toolbar.
Change-Id: Ieb439ae489ab7110b81382ffdcf0d3d3ad2f84ac
mw.Uri requires undefined rather than null to unset a parameter;
null instead generates a parameter with no value (and no equals sign).
Our own code in ve.init.mw.DesktopArticleTarget.init.js parseSection()
can't parse that and causes an exception.
Change-Id: I783ea6b91c115b79bbd9deac6669bea0661139af
The EventLogging extension no longer uses these internal modules.
They were phased out as part of last year's "lightweight EventLogging"
project (detailed at T187207). Migration notes at T205744.
VisualEditor has migrated already, mostly. It still depended
on the existence of these module names for some condition guards.
* The subscriber for 'mwTimingHandler' was guarded by 'schema.EditAttemptStep',
but did not emit events of that schema. What 'mwTimingHandler' really
needs is the '*SamplingRate' variable for its call to 'inSample()'.
This previously worked because the variable and the schema are both
provided by the WikimediaEvents extension.
* The subscriber for 'activityHandler' had a separate schema guard. This
might suggest an intent for the code to silently degrade if WikimediaEvents
were to be changed to no longer supply the second schema, or for the code
to work for third-parties without WikimediaEvents if they register only
the schema. However, this subscriber too calls 'inSample()' and needs those
variables.
I've assumed for now that it is okay for these to all be guarded together.
Even if the schemas were to be removed and we were to forget updating this
code, the new EventLogging client degrades gracefully from this (no errors).
Bug: T221281
Change-Id: I260c25752c3becfe6e499813197fbf7a3dba88c3
* Remove animation for toolbar sliding into place. It now happens on
the fake toolbar in MobileFrontend shown before the real toolbar
loads, and our toolbar just transparently replaces it.
Bug: T217784
Depends-On: If21aa0ea619ec2500ce5fca6fe81eb27f26bb047
Change-Id: Ib6ff7594e1982d1b46e9ca89d6b9722d025e8207
Abandon warnings are already handled by the code in MobileFrontend's
EditorOverlayBase. Using window.history.back() causes that code to run.
Having a duplicate way to trigger them only results in inconsistencies
because our dialogs animate in a different way.
Bug: T222315
Change-Id: I19c5616a6aeecf0ac63f37a564ef44f11df010b0
* Change the query in ve.init.mw.ArticleTargetLoader#requestParsoidData
so that in non-RESTBase mode with wikitext it still returns the
metadata required to initialize the editor, using the backend API
code added in I1b35b28e428a1f86d2e34d90ddbe73361ce14818. This fixes
the exception from T222312.
* Introduce new configuration option $wgVisualEditorAllowLossySwitching
to control this feature. It is enabled by default, fixing T214542.
We allow it to be disabled because switching in non-RESTBase mode may
cause "dirty diffs" (non-semantic changes to the wikitext), which are
undesirable on wikis where users carefully review all changes.
Bug: T214542
Bug: T222312
Change-Id: I58879cba5612002c70c24731306214d2577c2c52
Section=new behaves more like a form than a full
document editor, so allow focus to be fully moved
to the title input without leaving a deactivated
cursor behind.
Change-Id: I7e3835da925b27f5df79dcbdd4550445795cdc51
In general action=edit could be bound to a wikitext-specific
edit link, but in the case of redlinks we can use the
preferred editor instead.
Bug: T223793
Change-Id: Ib0851e9e2ce441ae93311153801e2c3de0a2063d
There are various circumstances where the wgIsProbablyEditable check
gives incorrect results (hence the 'probably'):
* User is blocked (T111217)
* Page is protected from creation (T173763)
* Page is transcluded on a cascade-protected page (T217217)
Bug: T111217
Bug: T173763
Bug: T217217
Change-Id: I7df8909c31f29d2e7521bef8612c27cb61146a4d
If a new namespace was added to $wgVisualEditorAvailableNamespaces,
but VE was loaded on a page with old cached HTML, the 'Edit' tab's
text would incorrectly be 'Edit source'.
If $wgVisualEditorTabMessages['edit'] or ...['create'] was changed
from non-null to null, but VE was loaded on a page with old cached
HTML, the tab would still use the old text.
Change-Id: I2d5c7b922ba480eb90fa0a6da7a1901f062c96df
Prior to 80bfbfc54b this worked by
accident, and with a number of bugs depending on your settings (see
T219457). It turns out that Wikipedia users have invented various
workflows that depended on this bug (mostly involving sandbox pages in
namespaces where VE is not enabled). Restore it as a supported
feature, and in a way that avoids the problems it previously caused.
Bug: T221892
Change-Id: I62714b6f2905efd1d1b34c7a13b9917cb6c609fc
Use an API 'parse' call with a sectionid to trigger
a full document expansion using replaceSectionAtRev.
Then send this off the RESTBase to convert to HTML
and statsh.
Ensure the etag is passed back to the API response.
Bug: T117716
Bug: T223023
Change-Id: I1b35b28e428a1f86d2e34d90ddbe73361ce14818
RESTBase is changing the way it is storing HTML/data-parsoid renders. In
order to still support VE (and other editors) with quick access to HTML
and matching data-parsoid (needed later for transforming the modified
HTML to wikitext), VE now needs to let RESTBase know it intends to
perform a transformation call later on. In that case, RESTBase will make
sure to keep the matching data-aprsoid around for long enough.
Bug: T222639
Change-Id: I02672e29bd0f331794fd77d9e56f9cc6822d9b9e
This message was upstreamed to core and later renamed.
Use the upstreamed dialog itself when switching sections.
Bug: T222525
Change-Id: Ibd2d75ec503e92b5ddec2105f762b0c9f0dc96fb
Set this.loading to dataPromise before the done
handler is attached as that can run synchronously
and try to set this.loading back to false.
Change-Id: I4e5208b7bac9968951f30555dc56e054c38a3935
If the editor was opened via a section edit link, but visual section
editing was disabled (the editor contained the entire article),
generating visual diffs would fail.
Follow-up to ec6cbeaf5e.
Bug: T222489
Change-Id: I74d589bb9f7ee09312ef816148c581440beb65e5
* Remove 'discardChanges' from switchToWikitext. This was
intended to discard changes even when the document was
modified, but it is no longer used as we always keep
changes if we can.
* Remove 'leaveVE' param, it was only used once and has
been replaced with a direct call to switchToFallbackWikitextEditor.
* Don't reset 'section' if there are no changes.
Bug: T221981
Change-Id: Ia39345da44d203ba67ae331917c8d5ece7d42ef7
I think mostly the incompatible changes were made in commit
I5a7d73b20617cb3c6d6379084ac4bea23ec3bc74, but I didn't try
to track them down.
Also fix an issue where hrefs in section edit links generated
by this code were wrong.
Bug: T208102
Change-Id: Ibf6564bc0dcb7fcb420739a897b54346a01add02
When this was introduced in 7b2cacbe57
(2013), the confimation dialog was a generic confirm() popup. Now that
it is a OOUI dialog, the dialog overlay serves the same function.
Change-Id: I9812ab55c7a8179524865d93a6d269e388d4c4ab
This is a clean up after collecting the necessary data related to
blocks and how often users see the block notices
See: https://phabricator.wikimedia.org/T189724
Bug: T214214
Change-Id: I532a0cd95009109ba25caa8dd31badd5c1900da7
If a page can't be edited in the requested mode (e.g. veaction=edit on
talk pages), the parameter will now do nothing (the editor won't
load), instead of trying to load the editor in another available mode.
Bug: T219457
Change-Id: I2cd78ea13ba13ff622f5e4b7db033f82dfa7875e
In case we get confused and try to load the wrong kind of content for
a given mode, it will now fail with an error message rather than e.g.
display a source editor full of HTML.
Bug: T219457
Change-Id: Ic23c2bbda2931da8ca015b4ea3673dbce4e2d994
New changes:
f039957f3 [BREAKING CHANGE] Use keyed objects for importRules blacklists
Local changes: Use extendObject to set importRules
This allows us to inherit the ruleset from the parent
so we don't have to worry about keeping it up to date.
(For example alienTableCell from upstream was missing
in MW).
Media/Gallery dialogs: Add missing mwTable types.
Change-Id: I366a091ff4def66cc25200b3d1b2c23ba6b716f7
Depends-On: I8ff7e8242c8db235a0f9e11e2e52f90d62d368a0
This code is supposed to check if we have scrolled past the end of the
page, and if we did, scroll back up to a more reasonable position.
However, it seems to incorrectly trigger in other cases, and scroll
the page back to the top, because getBoundingClientRect() occasionally
still returns bogus values (even though we try to avoid that). This
happens inconsistently when focussing the surface or closing a dialog.
We can't correctly display the toolbar while scrolled past the end of
the page. Removing this code causes it to go into an endless animation
loop in that case. But that may be preferable…
Bug: T219200
Change-Id: I152f2b39351ffd3c9799eea33cce95e05d2f9ab9
The EditAttemptStep schema doesn't have some of the fields we attempt to
submit, because we're generalizing and it only wants specific ones. Be more
selective about what we submit, so validation messages in the console will be
quiet.
Bug: T218163
Change-Id: Iaf9f47b61c138bb0106b2dfc4be2caec02a10d1d
We need to set 'padding-top' matching the height of the toolbar
somewhere, so that the toolbar doesn't cover the top of the editing
surface. Apparently moving it from the current place (the top-level
wrapper for the MF overlay) to the document node (the node with the
'contenteditable' attribute) allows Safari to properly scroll the
cursor into view when focussing, rather than scrolling it offscreen.
Bug: T219066
Change-Id: Iee1e03bce24c2f149a0aa0f393a37b9db43eaca6
The edit might have changed the section title, which will change the
section-hash, which will make the redirect break. The only way to avoid this
is to use the HTML provided to saveComplete to check for the new hash.
Bug: T213120
Change-Id: I5adfdb44a8304ed4f30def74400e4512e9e8c0ae
It appears that I did a Ctrl+X in one file, but forgot to Ctrl+V in
the other. And no one noticed that 100 lines of code went missing.
Follow-up to 73561f7aba.
Bug: T218946
Bug: T219041
Bug: T219043
Change-Id: Ib1fd85d121083239397698ff1a30a7908deca25f
ve.init.Target#isToolbarOverSurface has been removed in VE/VE in 2016:
8b1208cb976278bd44025e6d2c86a3ea6ed8c177. Nothing calls this method.
Change-Id: I9640978b45e568412db4b1c5aa80631a68d847b2
On iOS Safari, when the keyboard is open, the editor toolbar could
previously be scrolled out of view, due to how the keyboard affects
the viewport (or rather how it doesn't).
Detect when this happens and bring it back in, with a similar slide-in
animation as when the editor loads. Technical restrictions prevent us
from really keeping it in view at all times, and I think this is the
best we can do (and it looks almost intentional).
Bug: T218414
Change-Id: I5eed360d4644815bc9829fbc6b0ffd79b205d10b
It will be easier for us to maintain this way. The code I'm moving had
a lot of comments saying that it should be moved here.
See MobileFrontend change Ibe192360bdecab86519de1781f66f90a3441c551.
Bug: T218946
Change-Id: I908e035ec245a9b190f05e64c35dbb29936434de
It was here because our old hacks prevented the viewport from being
scrolled, so the keyboard would always cover the last few lines of
the surface. But it is no longer necessary after we ditched the iOS
scrolling hacks in MobileFrontend.
Bug: T217769
Change-Id: Iaf3f86c0fc43f75d11a43462721f44d62abc6eb3
This is no longer necessary and doesn't work after we ditched the iOS
scrolling hacks in MobileFrontend. And the default implementation works!
Bug: T218429
Change-Id: I5fba78a3877901dac5afda46d3004c07cad383d0
This enables mobile section editing if the user is logged
in and has an odd user ID. Otherwise it is disabled.
Bug: T218851
Change-Id: I2c22d7636ae11d2db7780ae5adb3abe9df532b7c
Passing section=undefined resulted in no <section> tags being
unwrapped, which broke the historical diff. Ensure 'null' is
used instead for whole documents.
Bug: T217752
Change-Id: Iec33e6ab83bfbd011df9dc05f4daccc26b1df8b5
Catlinks is hidden via a class when there are no non-hidden categories on the
page. We thus need to toggle that class depending on the categories
added/removed from the page.
Bug: T213528
Change-Id: I4067c5721c28041542b9ef2dbc796fbc41b1afe8
Generating the templatesUsed list is relative slow, and is only
used in an obscure part of the editor, so only generate it when
needed.
Bug: T209078
Change-Id: I1cecdad65b80c4c9b1746e752ea4b41bc0fc0037
This code, added in 703b2c2ed0 (2015),
is no longer necessary.
// Disconnect the tool factory listeners so the toolbar
// doesn't start showing new tools as they load, too
// much flickering
this.getToolbar().getToolFactory().off( 'register' );
Introduction of targetLoader (d371014e5d)
resulted in all tools already being loaded before a Target is
constructed, so this is definitely not needed.
// Disable all the tools
this.getToolbar().updateToolState();
The tools are already disabled because we set the surface to read-only
above, so this does nothing.
Change-Id: Idb162b60891cd1b961e29d2b9f62b74908f17957
Previously, we tried to keep the list of category members and the file
thumbnail, history and metadata visible while the editor was open.
I am removing it because:
* It is not very useful, as you can't interact with them (e.g. links
are unclickable).
* It is inconsistent with the wikitext editor (except for non-existent
category pages, and I'm proposing to change that behavior in T139191).
* It causes issues when other code doesn't expect the special setup
for those pages (T194068).
This introduces a minor change to the handling of normal pages: after
the save, instead of replacing all contents of #mw-content-text with
the new page content, we only replace the .mw-parser-output child.
Normally the effect is the same (it's the only child), but this could
theoretically affect interactions with other extensions or gadgets.
Bug: T194068
Change-Id: I26cc82d3e0f0d64e3f18a80d232005fc7ab3b374
Red-link metadata was added to Parsoid 18 months ago.
On long pages evaluating this information is very slow
(hundreds of milliseconds) and completely redundant.
Bug: T64803
Bug: T209078
Change-Id: I5b6c6da588301ed59fb21e1ce930f5b72db48e67
