New changes:
dc0cad28c tests: Factor out ve.dm.TestRebaseClient/Server into their own files
98061108b Convert RebaseServer tests to data provider
3e746a51b Update OOjs to v2.0.0
16231a812 [BREAKING CHANGE] Drop support for Internet Explorer 9
64daeb4b5 Localisation updates from https://translatewiki.net.
3310e876c Update DOMPurify to 0.8.5
579033a83 Follow-up 16231a81: Drop IE9 support in createDocumentFromHtmlUsingIframe
Bug: T162277
Change-Id: If17c6a3fac0920db29f80069f344ffd6be46c3c9
The bare minimum of elements that must be present on the page for
VisualEditor to start and mostly work is:
* '#content' to attach our interface (toolbar etc.)
* '#mw-content-text' to replace with the editing surface
* '#ca-edit' to start up the editor
When those elements are missing (and we expect them to be present,
e.g. we're not on a special page), we print a console warning.
For example, VisualEditor now launches on Timeless and CologneBlue.
It does not launch and prints a warning on Modern and Erudite.
(It is actually possible to edit and save pages on CologneBlue,
although various pieces of the interface look funny.)
This does not change which skins we claim to support: these are still
currently only Vector, MonoBook and Apex. But it allows third-party
skin developers to more easily test the editor and fiddle with their
CSS to make it look right.
Bug: T161373
Change-Id: I97c786f3c8c795c238b10bef332a129ea26f86c2
The element has both the class and the id set to 'catlinks'. This is
just for consistency, I have no preference on which to use.
Change-Id: Ic96b0f21fe790d0aa03bfd6366ba246e8493a038
Redirect pages have two extra things not present on normal pages:
* Redirect subtitle: the "Redirect page" shown under the page heading.
* Redirect page content header: the "↳ Foo" at the beginning of content.
Our handling was pretty messy and had some bugs (T161614).
Notable behavior changes:
* Update 'wgIsRedirect' in mw.config after saving the page.
* Use #mw-content-text rather than .mw-jump for inserting the fake .redirectMsg.
.mw-jump is not guaranteed to exist on the page (it's a skin feature).
* Never replace the real .redirectMsg existing on the page, except by the
new HTML after a page is saved. Our fake is separate now.
Bug: T161614
Change-Id: I96a5e45a71bf10bf6a2b501dc0cf81e6c37060ec
To reproduce: start editing a page, turn it into a redirect,
cancel editing and go back to view mode. The "Redirect page"
subtitle under the page title should disappear.
Note that #redirectsub is correctly spelled all lowercase,
unlike #contentSub.
This reverts commit 0e1bc7309b
and fixes the code instead.
Change-Id: Ibacd73122dfec63268a77794bc77c4b88876d3ee
As it happens, #redirectsub is correctly spelled all lowercase,
unlike #contentSub. This line does nothing. We remove #redirectsub
correctly elsewhere; I don't think this was ever meant to be here.
Follows-up 96421b283c.
Change-Id: I3e4c6eb2ff94f363b488477b3ddd248e571e723a
Running setNotices( [] ) makes MWNoticesTool run this.destroy() to avoid taking
up extra space in the toolbar when not needed.
Also, make setNotices() bail earlier.
Change-Id: I492fd59a1e15e8f296bf7f03d9b0035c40dafa59
This avoids an OOui deprecation warning for trying to toggle a popup which
isn't connected to anything.
Bug: T160161
Change-Id: I9ad97785f0de4ecf81d189985459689644e25923
Distinguish between the initial summary state, and a value stored from the
summary when switching modes. This lets us avoid overwriting a stored value
when section-editing, or assuming a stored value after a switch was never
edited.
Bug: T159686
Change-Id: Ie7640538140a14bbafd539b3a45928f5c55cf804
Switching from visual to source reruns setupUnloadHandlers, which caused an
infinite loop where it thought it was its own fallback pre-existing handler.
Bug: T153346
Change-Id: I8df55a5395ede02fc34ec47a94f2c780dc08595f
Since December, the handler now expects to be passed the mode we're switching
to. Since we're filtering for just the edit link and not edit source, we can
hardcode 'visual' here.
Bug: T159374
Change-Id: I160045dd59ff1b4c2d180dbbe37280a1839886df
Pencil is the icon we use for 'edit' elsewhere in the UI.
Use the eye icon for the VE tool to specify 'visual'.
Bug: T116417
Change-Id: I12b6bab2a52758685abde04579b274a32d651174
This will make it easier for third parties to load the required
modules for an article-like target.
Also cleanup mobileArticleTarget to better match desktopArticleTarget.
Change-Id: Idb75e2bd99d8cf2c298e101cf92e8af4f85f5ad5
These are now computed upstream based on position of toolbar.
The menu tool now needs its indicator explicitly hidden.
Change-Id: Id2280f70553e1e45f3a42af45684c5808797925c
As for the Translate extension this will read a global set by the
TwoColConflict extension to disable the VisualEditor and its hooks
on the modified edit conflict page there.
Bug: T156251
Change-Id: I164d1546c9d98d8715650e13c63a01d06c76315d
When switching from the old wikitext editor to VE, we stored "editing..." as
the original page title. Then restored that when saving / switching to "read".
Instead, detect if we're on top of the classic editor, and use the pagename
message to make a plausibly-correct title.
Bug: T126077
Change-Id: Ib0289de71c3ae947ca0a3e45fe1e620378689c35
This wasn't such a big deal in VE because we load the full
page anyway, but in NWE it is more annoying if you reload the
page and are dumped in a different edit area.
Change-Id: If49569cbe1a2759d8cbbd7ee42dd1b88da39a946
Provide a common function to parsing, and always pass
section to the target, instead of having the target
re-compute it itself.
Change-Id: If9fc24ffa51507cd969fa1b4dfc1519a0b50a572
This has to go to the API, because it's skinnable. Also, separates out a few
of the initial page setup functions so they can be re-applied to the new
content inserted.
Bug: T151651
Change-Id: I8d5a6504edc2e0486a0b4f016d8ee6d9a6228de9
Create a new MWSaveDialogAction, which can be hooked up to triggers. Switch
the existing save dialog accesskey to just use the trigger system.
(Maintaining all the accesskey logic, so we don't change the shortcut on
people.)
Bug: T149914
Change-Id: I9b4ef8504148703556f802b266a517dd5098c06b
New changes:
f58ddea DiffElement: Use document slices with full internal lists
83800c0 DebugBar: Remove hard-coded i18n
b4f89e1 Update OOjs UI to v0.18.1
40c7bf6 Factor out active node functionality from SectionNode for captions
2d967be Move 'mode' property to surface, rename target property to 'defaultMode'
5d8c214 wrapAllNodes in sourcefragment
dd3d9e5 Refactor ve.dm.Transaction
9d61aca Use canonical ve.dm.TransactionBuilder.static.newFrom* methods
df4f72a Make table caption node an active node
b79f72d Core source mode
7533ac4 Localisation updates from https://translatewiki.net.
ae30d71 SourceSurfaceFragment: Check range is not collapsed
Local changes:
Get edit mode from surface where possible
Depends-On: Iec758c1892d518ad4bc2c0d1aaf6ca00fa354323
Change-Id: Ifaf6a26078b2731b374aaad2cb40c08928de9c84
This is set to 'true' in mw.Target#setupSurface(), but not initialised or
used there in that class.
It was previous initialised in an indirect subclass (DesktopArticleTarget).
Move this to the parent class instead for clarity.
Change-Id: Ie3e12d254aa4b689fdce64620e110164311bc60a
Reject the activating deferred making handleLoadFailure
redundant.
Fix logic for switching back to OWE aware of NWE.
Change-Id: I328fc944bb6b9152752742fe35c56b95e3255b88
Otherwise if the target was already loaded the mode will
not be changed, which can cause the wrong editor to load
in multi edit tab mode.
Change-Id: I64071dfdfc8511cce51bda20b9c62d7fc126c528
We should avoid using alert()s as much as possible due to their unhealthy
interaction patterns with any other open tabs or user tasks.
Change-Id: Ib6a217c988322ad17bc7e649c3281eb053b54bbc
Gives us extra functionality like checking for
isUnmodifiedLeftClick which makes CTRL+click on SET
not change the current page, as expected.
Change-Id: Icb37d7383374ee63798443659a2bcb2f1545c8c5
Just use the normal section parameter, haven't been able to figure out why VE
would need to be special about that.
Bug: T149958
Change-Id: I9338d0f1498fbb579fa8c340d6e7274c6d48155b
This prevents the discard warning from being shown if the user
switches back to NWE then back to VE, as by this point they
have left section editing.
Change-Id: I734684c54a8900aab3332ed8421cb42f7f845738
New changes:
e4c8003 Bring in target 'mode' property from MW
Local changes:
Use upstream Target#setMode functionality
Depends-On: I9d501cb77c714fbd299b5816d302b0bdde7833cd
Change-Id: I2fcda6ca7d82d880101d9ba2a027d4ef066aa238
NWE can return data in 'visualeditoredit' instead of
'visualeditor'. We may want to change this later.
Also restore removed hack code that popuplates missing
VE API data when requesting wikitext.
Change-Id: I9a782a85444d37a3e21c071db7e175ee9b0b1c4f
Use history.replaceState if available to get rid of the venotify query
parameter to avoid polluting URLs.
Ideally this would just use a cookie like the core post-edit
notification.
Change-Id: I6ce142012053e2fc18dd44fd0f7b82914acea076
Some older browsers may fail with >1024 arguments. This
does not have a noticeable effect on performance.
Change-Id: I8c85f43425997735bbbe08d27374b79502e02ecf
Get rid of the 504 Gateway Timeout handling while we're here, in preparation
for bug T135748 where that kind of thing will be made normal.
Change-Id: Ia4abb1a68ff7ab7ff365c7348bf4d2ad35c43a27
Converting a document we know contains just plain text paragraphs
is slower than it needs to be. The speeds up the conversion of
[[Barack Obama]] from ~2000ms to <100ms.
Change-Id: I2081858d835286b48db0627696a9eb5294317c2e
transformPage() makes significant changes to the DOM, so measuring the
scroll position synchronously right after that results in a slow forced reflow.
Instead, measure the scroll position before reorganizing half the DOM.
Change-Id: I3baee5c11ca228696ddbfb30789745f5f0faa20a
Instead of getting wikitext from the HTML, which is probably
slower and leads to issues with the whitespace stripping hack,
override getDocToSave and get text from the linmod directly.
Bug: T144621
Change-Id: I6b6cf16ee8f3720398ba8f5c85e7715c2e68329f
So we ensure we send the actual revision ID, not potentially undefined, to the
method. If we send undefined, it'll give us an etag for revision number 0...
And that will cause issues when we later attempt (in a future editor switch) to
pass an etag for revision 0 along with the correct revision.
Bug: T143920
Change-Id: I0c27956abdd26ae4f5f1b91010a5eab8ee0354ab
Perhaps in the future we could enhance this so that clicking a different edit
link queues that up for after the current activation finishes / aborts the
current activation cleanly.
Also, error-checking on setting an invalid progress state when loading.
Bug: T143160
Change-Id: I6c78cb19953df6fb564c06bac807e6897b63df19
To align with the linked patch in MediaWiki core. Taking advantage of
the opportunity to use core's messages for these, and remove some dead
wood old messages that were never used like "restore" items in mobile.
Bug: T139033
Depends-On: Ie81b5edd275963a965cd44d0fd325cae9ee2f1a6
Change-Id: Ie00e94cc77cb750a7e8d1104366bb3dad65af8a4
Pretty much fix everything on the insert menu... RB doesn't like it when you
try to specify both title and revision ID, but don't provide an ETag.
Change-Id: Ib25d023309d984ed8848f67b349b23231f27a957
If you viewed a page with an ?oldid= query parameter set to the ID
of the current revision, some parts of VE would believe we were
in oldid mode (because there's an oldid present), but others
wouldn't (because the revid we're editing equals the newest revid).
This caused bugs when opening the editor a second time after saving
(which is normally impossible to do after an oldid-mode edit, because
we navigate to a new page after an oldid save, but we don't do that
in this case).
Ensure that:
* The internal state of DesktopArticleTarget is updated correctly
after saving in this case
* The ?oldid= parameter is removed from the URL after saving
* DesktopArticleTarget.init doesn't preload the article HTML
on a second/subsequent editor load: this causes issues because
it caches the oldid, and generally speaking the Target's internal
state is not considered
Bug: T141330
Change-Id: I74034328797c59f7249f1f6f4f53a92ee1c26334
When our remember-last preference changes work, alongside changing the
label of the edit tab, also change the label of the edit section links.
I seriously have no idea WTF I was thinking back in December when I
put some of this code into #activatePageTarget.
Bug: T137424
Change-Id: I581c0acf0a3ad11ad3bcf4e2e46242907ca9166f
If you deactivate without saving (e.g. using the back button), the page
contents aren't replaced and so adding the page click handlers is pointless
and harmful because they will run twice.
This was my fault in Iad1713e1
Bug: T135387
Change-Id: I3c83009bfc0f42125cdcde7f7a51de9bc2f72abf
So if the user supplies a summary in VE, WTE with the "Prompt me when entering
a blank edit summary" preference won't moan.
Bug: T135979
Change-Id: I0e2d2b6f8fb03bb56d600f1118daf82fb3715b66
In case we were switching to source.
Normally MWWikitextSwitchConfirmDialog#getActionProcess would do this for us,
but this closeWindow call doesn't trigger that, we can't get it by overriding
MWWikitextSwitchConfirmDialog#close either.
Bug: T134333
Change-Id: I66a12ff6d13601250b9d470e1be54fe38a1ef06c
* Ensure ve-loading isn't removed early
* Quickly detach then reattach the loading bar during
activation so it doesn't get moved and marked as original content
Change-Id: I263c9627348953a11966f8bcc435d0d89b0b6084
T52720 (link corruption in FF13/14) appears to be fixed.
T52780 (editor not loading in FF11/12) appears to be fixed.
FF11 has a new issue where templates are normalised even
when not touched (T136607)
Change-Id: I34358e3d90b8186e6b89c04c038ab79c908fc81b
Define $editableContent on target construction, and mark
all non-ancestor nodes between that at the target container
as uneditable (50% opacity, no pointer events).
Bug: T58289
Change-Id: I7fe51104bd5aa1bd53ffc604e5f02752c7553578
The SupportCheck method asserts that the method is available so use it
instead of the confusingly named jQuery version.
Change-Id: I2105384cc4f6f2ec1cdb24a7bf3b0f4cce7672d3
You can still open VE with prefer-wt mode by going to veaction=edit, and
T116406 proposes making it accessible with a keyboard shortcut as well.
Change-Id: Ifc25b5147a96a200ac3a7de465d5cdf67e2e255b
transitionend events seem like a neater solution, but we should
migrate all of our code at once, and provide a polyfill for
browsers which don't support it (IE9).
Change-Id: If6ae030856f8e69cc8bb26f21bfed30d5d22775c
On the client: getLastEditor in ve.init.mw.DesktopArticleTarget.init.js
On the server: VisualEditorHooks::getUserEditor in VisualEditorHooks.php
Change-Id: I9cc0f367aee2dda43cffc6918bfb042ac8ae3bb2
Place things like the page title and redirect link inside a new
container $originalContent, which is appended to the surface
when ready. Replace margins with padding in various places.
This will allow us to surface-height-matching sidebar for dialogs.
Change-Id: I60d80fb303bdaf93e9d121f62d534ee3a3056e59
veswitched and wteswitched together cannot result in any sane behaviour.
To reproduce:
* Open VE (while having multi-tab pref?), switch into WTE
* Make a change
* Switch into VE
Change-Id: I90e19169e3fab60ab876c8e4d349801309db262f
wgTitle is basically just $title->getText()
wgPageName is $title->getPrefixedDBkey()
So we need to take wgPageName and run getPrefixedText (difference is underscores
get replaced by spaces)
Change-Id: Ib99f0d20f1ba99338f80bbbf39cffd544887c3fa
Regression from 863a2c2974.
Sometimes when clicking "Switch to source mode" in the toolbar this
error is thrown and the interface freezes because setEditorPreference()
only returns a promise if the value is different.
> Uncaught TypeError: Cannot read property 'done' of undefined
> - ve.init.mw.DesktopArticleTarget.switchToWikitextEditor
Change-Id: Ie59f06a8ed9af9c4fb8199b013c06992d0c4f700
New changes:
0371bbe [BREAKING CHANGE] Refine VisualEditorSupportCheck call and document pattern
Local changes: Use new VisualEditorSupportCheck pattern
Change-Id: I7dc0c360b54a93397180b18d88d72532e439da5c
Warn the user about an empty edit summary when the "edit section"
link is used to trigger VE.
Re-does Ic7b456ca a different way.
Bug: T114857
Change-Id: I319c9c5bed47140a81eb409d490c9f82b89a49fe
New changes:
c02ea46 Clarify documentation of directionality methods
43c9b6f Move initialisation browser support checking from downstream
c89be66 Localisation updates from https://translatewiki.net.
5fb7907 Follow-up c89be66: Fix build for added language 'nan'
fc74fab dm.ElementLinearData.compareElements: Add a few missed test cases
Local changes:
Use initialisation browser support checking from upstream
Depends-On: I27a8e7b4376647f01cee648de987dc3ca6a060d5
Change-Id: I3fc79422ee5e487cf5fae3929fe492f21cd2840a
hideLoading always gets called after activating, not just failing
Follows-up: I13058ae1
Bug: T127184
Change-Id: I47d1892feacbfcf832aa32f689f173601b76c4ba
* Try to hide loading bar on failure
* Don't set wgAction back to 'view' if we're dropping the user back to the
wikitext editor
Bug: T125580
Change-Id: I13058ae131a1dda3b172e78d9b143d70831c47f1
* grunt-contrib-jshint: 0.11.3 -> 0.12.0
This version of the linter cleans up a number of options. 'es5' is now assumed to be
true by default, and replaced as 'esversion'; 'latedef' is extended to also cover
functions' defintions, disabled for us for now; 'futurehostile' is introduced to aid
future-incompatible names for ES6+ support.
I adjusted ve.init.mw.DesktopArticleTarget.init.js based on this, as this file is
meant to be loaded by all JS-capable clients to determine if it is (amongst other
things) capable of ES5, so now we assert ES3 compatibility via jshint.
* grunt-jscs: 2.5.0 -> 2.7.0
Minor version bump that doesn't affect us.
* grunt-jsonlint: 1.0.6 -> 1.0.7
Trivial version bump that doesn't affect us.
Change-Id: Ieb7b6748b0cecf275cfc284fc66617189372461f
