Since VisualEditor cannot be sure about how skins have marked up the
page, the code that marks content as uneditable (not clickable) can
be tightened further to make sure the edit and view tabs are not
marked.
This fixes the issue where the tabs are part of the content area
Bug: T162503
Bug: T310197
Change-Id: I4e7388c2e3b681b766068639ab9325bb6289562d
As this is a surface command, rather than a documentTrigger, pressing
escape will not close the editor if done elsewhere on the page, e.g.
in the site search bar, or an unrelated modal.
Move the logic to ArticleTarget as this could theoretically work
on the mobile site with a keyboard.
Allows us to remove some hacks around the ToolbarDialog that are no
longer necessary.
The command can also be used by the MWBackTool (which should be renamed)
and allows us to remove some custom logic from it.
Bug: T310694
Bug: T310695
Bug: T306763
Change-Id: I91ee6916a91d80d9872b3b44dad7eca55ad1acc4
Depends-On: I29f6af4cc7c71a63a6d1bafc53d16b9abd1b60ec
The EditAttemptStep instrument is a candidate for migration to the
Metrics Platform [0]. The first step of the migration is to log events
both using the Event Platform (i.e. via mw.eventLog.submit()) and using
the Metrics Platform Client (i.e. via mw.eventLog.dispatch()).
The Metrics Platform Client can mix in additional information -
so-called context attributes [1] - based on the stream configuration.
The majority of the default values mixed into each event via the
mw.eventLog.Schema defaults mechanism are already known to the Metrics
Platform Client.
Note well that the Metrics Platform Client will not log an event without
one or more streams being configured to receive that event. Therefore,
this change is a NOP.
An example stream configuration is given in [2].
[0] https://wikitech.wikimedia.org/wiki/Metrics_Platform
[1] https://gerrit.wikimedia.org/g/mediawiki/libs/metrics-platform/+/aed6738b845/js/src/StreamConfig.d.ts#31
[2] https://phabricator.wikimedia.org/T309013#7953227
Bug: T309013
Change-Id: I7627f116cf32ceb3455a33f4f7bb55208ba92671
This only happens in AMC. Currently it is also not possible
to update categories in VE, but this may become possible if
we enable 2017WTE on mobile.
Change-Id: Ifeb6cc18910ce2fca634bc3e2245aac7e5c37e52
Needed after Ia18f31a299338f94e69f1882e6e477f3a22ae905 in VE core.
Bug: T307849
Depends-On: Ia18f31a299338f94e69f1882e6e477f3a22ae905
Change-Id: I87f3ac0974702ecaf7f5459604371de06f4a5756
Now done in VE core since Ie86217ba5651df8c427464e460ed836903834a3c.
Bug: T308200
Depends-On: Ie86217ba5651df8c427464e460ed836903834a3c
Change-Id: I3f65b0bc100895f7bd2262c3cbb5231fb055758b
This is currently only handled in DesktopArticleTarget in teardown,
which happens after we've wasted time trying to update the page.
Also ensure we always reload on non-view pages on MobileArticleTarget
for other types of teardown (e.g. quitting the editor).
Change-Id: I7fb352fcacc8727bb113115e98af38a3940a8f9c
Requires switchToFallbackWikitextEditor to return a promise.
We can now pass dataPromise to the progress bar shown when
switching editors, so it hides if the switch fails.
Also fix logic for when a failed load is not retried.
Bug: T306763
Change-Id: I752ca505e7957b392202d44455b1e21b6e50fa63
Also invert logic and rename variables to make tests in tryTeardown
and switchToWikitextSection identical.
Bug: T294463
Change-Id: Icb86a676c3f315e4a3d229f46eb383dcd7959973
The Apex skin (not to be confused with the Apex OOUI theme)
is abandonware. Keeping files around that pretend to support it
is just maintenance debt.
Change-Id: I176e4e25307ab8ca4166a6e9b55e3bec4f058d05
Same random finds while working on something else. I carefully
checked and made sure these methods are actually called without the
optional parameter.
Change-Id: Iab36fd130258322985b5d6e7f8e1f7b4ee235ba2
These are only needed when we need to access a specific `this` from
within another `function () {}` context. This is not the case in the
situations here.
This is split from Ibf25d7e to make it smaller and easier to argue
about.
Change-Id: Ide1476de91fc343aa992ad92a1321d3a38b06dd0
This ensures the full loading sequence is shown when the
user opens VE using history navigation.
Bug: T301843
Change-Id: Ia7a641c8bd5a036f23c9da94bc539d8cf66c5021
Previously, a toolbar with the extra tool could be shown for a moment
while the editor was loading.
Bug: T301626
Change-Id: I7f5d01ef6e64ef85141a38a236fe2a1d1ada00e8
When the editor is loading, and the browser window is narrow enough
that we should use the narrow toolbar, the wide toolbar would appear
for a moment. It was particularly jarring when it caused it to wrap
over two lines.
Bug: T301626
Change-Id: I551a66edab40235c4a26751eed163b5b30ef698f
Make sure it works on pages which don't trigger DT, such as full-page
wikitext editing.
Bug: T301497
Change-Id: Id3d46eef319f278ee71fbbedbddd16d5ffd540d8
* Pass 'noPrompt' to #tryTeardown
* #loadFail gets triggered by the data promise being aborted,
so if 'activating' has been cleared at this point, don't
prompt about there being load error.
* Retry/cancel dialog assumed you were coming from a wikitext
editor. Rewrite the logic to handle teardown the target if
the user is coming from read mode.
Change-Id: If9e5d1fbf00f8359f4d06fe7fc3064cc7e349bc9
Previous attempt in 005a8d24ef,
reverted in 3c1d167b33.
The deduplicateStyles() function lacked a check for fosterable
positions, which caused T299767. This is now fixed.
Also added tests.
Bug: T287675
Bug: T299251
Change-Id: I0d22be9b66d26d09373cee63dd6ce52c1659e62d
We introduced these methods in Idc0cccbe but didn't update this
logic in MW.
This prevents focus being put inside an image caption if the section
starts with a block image.
Change-Id: I4cc1a116c1dcc8e1c642359ea768b6f5265a061b
Parsing it in the RL module caused the module cache key to depend
on the parse, which is slow and makes ResourceLoader sad. The usual
approach for solving this (I206bb05d28) can't be used, because of
how EditPage generates this message.
Bonus #1:
Generate the message for the correct page title. MediaWiki allows
customizing it per-namespace or even per-title, which we haven't
supported before.
Bonus #2:
Pass the context for message localisation (depends on I5f7c77970d).
EditPage::getCopyrightWarning() was parsing messages without the
interface flag, causing some needless processing elsewhere.
Depends-On: I5f7c77970d0525c0ff394f8bd72c69dcb5d00623
Bug: T298822
Change-Id: Iaa626f0e6379a5a370f9c465cea8528bb5bde7f7
* Find the first section below the top of the viewport
(usually visible) and measure its offset.
* After loading the editor, ensure this heading is still
at the same position on the page.
Bug: T296910
Change-Id: I9a05ea74ba3c19a4a91ddc1bc0afe311851c53e6
* Use this.getElementWindow
* Use this.surface.padding to avoid race condition
when the toolbar hasn't fully rendered yet.
Change-Id: I055b1d9458d73e435ede6096941a3e72c9c1ce74
In case the save is triggered without the save dialog (for example, when a null edit is made during suggested edits task in GrowthExperiments)
Bug: T298552
Change-Id: Id49b967cfa52d33848e9c911086000fa4501fa7f
Prevents accidentally treating plain text or user input
as HTML, which could be an XSS vulnerability.
Change-Id: Id4af48447a0907962a57340cb60aca08df9cc505
* Make ve.ce.MWBlockImageNode autofocus=false, remove
unused transition property
* Remove ignoreChildren from ve.dm.MWBlockImageNode
based on new definition
* Remove tests which assert that deleting in a list next
to a block image always de-indents. If this is desired
behaviour it should be fixed without reference to
ignoreChildren.
Bug: T295905
Depends-On: Idc0cccbe73d1b49d07b60c14a192a40f47d64608
Change-Id: Ib79a070f5d36dbe7742fa0760f8cdf55fe3046ed
Load errors are already handled in the MobileFrontend part of the
mobile visual editor, by this code:
66c55573e5/src/mobile.init/editor.js (L375-L387)
// Wait for the data to load before we show the editor overlay
overlay.getLoadingPromise().then( function () {
...
}, function ( error, apiResponse ) {
// Could not load the editor.
(1) overlayManager.router.back();
if ( error.show ) {
// Probably a blockMessageDrawer returned because the user is blocked.
document.body.appendChild( error.$el[ 0 ] );
error.show();
} else if ( apiResponse ) {
(2) mw.notify( editorOptions.api.getErrorMessage( apiResponse ) );
} else {
mw.notify( mw.msg( 'mobile-frontend-editor-error-loading' ) );
}
} );
Compared to our code:
ve.init.mw.MobileArticleTarget.prototype.loadFail = function ( code, errorDetails ) {
...
(1) this.overlay.onExitClick( $.Event() );
(2) mw.notify( this.extractErrorMessages( errorDetails ) );
};
The lines marked with (1) and (2) do basically the same thing. And
the function parameters "error, apiResponse" and "code, errorDetails"
are actually the same objects, just with confusingly different names.
This causes the popup with error message to appears twice (although it
isn't too obvious, since the two popups appear in the same place, so
only one is visible), and also causes bogus data to be sent in event
logging (T237063).
Bug: T237063
Change-Id: I7fe7a944707fe585251ce9e16bbb78ccd123a7ed
I tried to review all of them. Some of the changes I did:
* Make sure the `config` parameter is not marked as optional
when it is not.
* Make sure default values are mentioned.
* List individual `@cfg` options when it makes sense.
Note I don't list all options a class could accept (e.g. via all
its parent classes and mixins). That's too much. Instead I checked
how a class is actually used and list only these options.
Even then I don't list everything, e.g. unspecific options
like "classes" that can be used pretty much everywhere.
Change-Id: Idf4fbe1dc3608ace277df9e385f2f140df3a2f50
This just repeats what the base class did. It looks like this
was forgotten when the base class was introduced in I097311e.
Bug: T75822
Change-Id: Ie5f5d358f24be7f9168214ea80713b0f7f295f47
If editor loading was aborted while the surface was already being set up,
this code would cause the following exception:
Uncaught TypeError: Cannot read property 'tools' of null
Introduced in c02c529537 (2017) because
our coding conventions at the time demanded that all `var` statements
must appear at the beginning of a function.
Bug: T287487
Change-Id: Id657d6f1e1189c17ede25362f145bb7b10f441db
Reasoning:
* format=json must be the default. Nothing else makes sense in
the context of this code. This should not be a surprise.
* formatversion=2 is only a default when the custom
getContentApi() is used, but not when mw.Api is used. One
might argue that it's safer to always specify formatversion=2.
However, this is not done in other places in this codebase.
It should never be done or always.
* I find it confusing when the action=… is missing. Let's not
rely on this default.
Change-Id: I6ca29f76bffc0849103c5bcff4aaf28fcaaa4c52
Cloning the #catlinks node loses all native event handlers registered
on it. jQuery's `.clone( true )` method can only copy jQuery event
handlers, so it does not help. As a result, when the node is
reattached to the page after cancelling visual editing, HotCat's
interface is still visible, but not functional.
To fix this problem, keep a reference to the original node rather than
a clone.
https://developer.mozilla.org/en-US/docs/Web/API/Node/cloneNode#noteshttps://api.jquery.com/clone/
Change-Id: I2c0b3d1a919b67053a17dd11fd2b7dc7556267ef
These don't add any knowledge but make the code harder to read
and maintain, and are an additional source of errors.
Change-Id: Ied57741a3f985e355adfddb4e75378d5c497faa9
The class .ve-init-mw-target-surface is used on the same element
as .ve-ui-surface. This element contains surface overlays
.ve-ui-overlay, which can contain other .ve-ui-surface elements
(inside inspectors), which would then erroneously have the target
surface styles applied.
Bug: T284312
Change-Id: I8d20a830dc48f6a098b0f9e9a7c7c1656de0fe56
Just reading the method signature gives the exact same
information in these cases. In other words, this code is
able to explain itself.
Change-Id: I04d031f2b24c3b0d21fede2c19c64b54d30b5b0c
* Create getSurfaceClasses method.
* Pass surfaceClasses to target widgets.
This ensures that the 'content' class is passed to mobile
target widgets, and the 'mw-body-content' class is added
in a less hacky way.
Change-Id: Ibce6d1a1d0fda63cca354761f1b91f808858e95b
VisualEditor recreates the mw-body-content element. The element
with mw-parser-output already exists as a child. All skins now
consistently follow this pattern.
To limit the impact to the editor, we use ArticleTarget and add the class
to the surface, which corresponds to the mw-body-content element of a skin.
This avoids unrelated regressions in experiences such as DiscussionTools.
Bug: T283014
Change-Id: I4833d1ca9fda4fc0bd433760e47fe7010f00db05
Ib957eac2d checked to see if actionTools.notices existed before
destroying it, but assumed it always existed if editNotices was
set. This patch adds a check before attempting to show editNotices.
The error occurs because Ibc7fa48df unregisters the 'notices' tool
(along with many others) for AddLinkArticleTarget.js in
GrowthExperiments. I92a3162ef in GrowthExperiments will empty out
any notices to work around this problem.
Bug: T281960
Change-Id: Idacd365efa82ecd5c0074ead035eda0cb9444b1f
The regex that removes the wikilinks to create the initial edit summary
should have the global flag since it was taken from preg_replace which
replaces globally implicitly.
Bug: T276722
Change-Id: I21e3cdfe752657ad37d9a6bd473a7e7dbb6e4cd6
This is needed to reconstruct population estimates from a sample.
Depends-On: Ie5cf24e84a2ed041bf7c4f0b891387c45667467b
Bug: T273454
Change-Id: I3a40e74f8ccb80aa6ed7d3313a5394aa31baf572
If someone is enrolled the DT a/b test, we want to know about their
editing here as well.
Bug: T273096
Change-Id: I235f4ccbcbfbf95c6aa0df327a9a5a7d5ddb1038
We need the whole DM doc to show reference diffs
correctly. We can filter down to the active section
after the conversion like we do with the editor.
Bug: T272813
Change-Id: I2081dd520ff414caadaed2efda955d600953c957
New changes:
c17816c5f Diff sidebar: Make font size slightly smaller
f8439f4cc Deep-freeze linear data
a8919f78e Deep-freeze linear data added by transactions
Local changes:
Fixes for deep-frozen linear model
Bug: T119236
Change-Id: Iae4362c8dab0f2bd335e24498f3e0522b8b1d4fc
This was used when we used to pass API errors to showMessage, but
is now unused by the two remaining users (missing edit summary, and
"press ctrl+enter to submit").
Change-Id: I8a6b4db78d4e451cf3ec85fcdfd8293328aaaa3c
* Remove custom internal events in ArticleTarget for every error type.
The indirection was just making it harder to figure out what data
goes where.
* Centralize the actual logging in ArticleTarget, instead of doing it
in a dozen methods.
* Directly use the error code from the API for 'save_failure_message'.
Previously we'd lose the original error code and generate a new one
in the event indirection stuff, except for 'responseUnknown'.
* Update 'save_failure_type' map. Remove unused error codes, update
the ones that changed, and sort in the order in which the types are
listed on the schema page.
Bug: T272162
Change-Id: Ied602c456f4b0e7e9bb135e3200bec5ce65641ba
Switching from visual editor to old wikitext editor results in a
page reload and loss of the mw.libs.ve.disableWelcomeDialog()
flag. Use an URL parameter to preserve the flag.
Bug: T235812
Change-Id: I3968e5b7ae536d45fd764a8b7c3ea1f6d616033f
originalDmDocPromise is dervied from target.doc, so if that changes
ensure the promise is cleared.
Change-Id: I51219e06109b0ccf1a17c920131b764862be85e1
If a teardown has started, there should be a teardownPromise,
otherwise return a rejected promise.
Bug: T268358
Change-Id: Ia5cbd6b409a38f97243234ea7c87d24f71bdf3d6
In the new-page case, wgRevisionId will be 0 so it'll try looking for
parentRevId and the cast-to-int on an undefined will get us NaN. That
fails validation, so we should give one last fallback to 0.
It's _possible_ that we could instead make this an explicit check for
using wgRevisionId if it's anything not-undefined, but I'm not certain
about whether there are cases that wouldn't cover.
Bug: T237063
Change-Id: I8a38c0f3b8f8b2b596f5d0933e1a9e7f1326d7be
MediaWiki doesn't have such a limitation. This might have been copied
from the edit summary code, which used to be limited to 255 bytes.
Change-Id: I4afe9b1cde0663c47c0c2502b6e32116b912208b
the former is no longer present
note: i assume the style is still needed, I have
not tested
Bug: T255718
Change-Id: I4e7851362fc0d64097aeff5ec1535f8fe7480682
New changes:
c53286843 Make NodeWindow a standalone mixin
a787fbef1 Update OOUI to v0.40.4
0d74538ee Refactor dialog logging out of ui.Command into ui.WindowAction
ea55ede8d [BREAKING CHANGE] Improve structure of special character definitions
Local changes:
Fix output of fetchSpecialCharList to match new format
Bug: T264146
Bug: T264690
Change-Id: I2a28bb9c3e54cb5f9308ab361dee99bc801b467a
A language code is better than an empty string.
The string would be used in the LanguageContext to
describe an annotation like <span lang="en-fakeVariant">.
Change-Id: Ibc18c238d6216927447ca26f336e9d973c6b93eb
EditAttemptStep requires an integer, and just getting the value of an
input is always a string.
Bug: T261664
Change-Id: I57e76857086474365124b5b016902211b0e63166
This patch starts using watchlist related values from ApiEditPage
results instead of updating the "watch link" based on whether the
checkbox was selected or not at the time of saving the article.
This change does not depend on T261030 and can be merged without it
but T261030 needs to be fixed or temporarily watched items will not
display the right tooltip when hovering the "watch link" or star icon.
Bug: T260434
Change-Id: I2c844223620d7d28f36a0cd8ae3dee4b0c8ae5bf
If we call getBoundingClientRect() while the 'transform' animation is
still ongoing, it's going to return values reflecting the transform
- that is, the rect will be partially offscreen - which will trigger
our code that runs the animation again.
I don't know why this wasn't a problem on iOS 13.3 and earlier. Either
the timing was slightly different and the 'transform' animation was
able to finish earlier, or getBoundingClientRect() was buggy and
returned wrong values that conveniently worked right for us.
Bug: T259321
Change-Id: I6be0ddaeb6df54295fb14c45ba15fee41d61e33f
Rebuild the category list from the data we got from the API. This makes
it work regardless of whether we came here from activating on an
existing page, switching from source mode, or loading via an edit URL.
Because Parsoid doesn't include redlink status on the <link>s it
represents the categories as, we do have to manually update those
styles.
Bug: T251398
Change-Id: Iaaef3223816269bba1371fbb07956119a120c1ca
Change $originalContent to $editableContent, as counter-intuitively
the editing surface is attached inside $originalContent, while
$editableContent contains only the hidden existing page content.
It was not a problem previously because of the order.
Bug: T254518
Change-Id: I68f7ddb45f09dff475b183ccfeaf98e67b83066e
It's supposed to be a string or null. The parent class sets
`this.section = null` for us by default.
Bug: T257355
Change-Id: I375a3f6724235f22888bb7f0cb77a7081153768e
Translate itself will show an warning dialog when user tries to edit
a translatable page in visual mode. Editing in the source mode is
allowed.
Bug: T192052
Depends-On: I03ea64353b2d5b28d684f90936cab75ea4fd40d2
Change-Id: Ie4f3a6003c27773c6d0c5086c876fe424befe25a
There used to be similar code here, but it was removed in 2015 in
85b745666f because it wasn't correct in
some cases. Restore it and instead separately check for those cases.
Bug: T257124
Change-Id: Id4a17266111d064805aa0ca865182bb2f4b019eb
This reverts commit e5c1ef651b.
This change caused an error where templates failed to be inserted
into wikitext in the 2017 wikitext editor.
Bug: T255785
Change-Id: Ie57c49e68e594be22af2b1b479840f29e46131db
mw.Target doesn't know about revid and etag, so move that logic
to ArticleTarget, where the param can still just be a boolean.
Change-Id: Idf4632cd28554aaf5bbf5f2b44ded047c0c4b182
This handles the minor edit and watch checkboxes, as well as any added
by extensions (e.g. FlaggedRevs).
Switching in the other direction already works fine, that is
implemented in ArticleTarget#getSaveFields.
It doesn't seem ideal to put this code into the constructor of
ve.init.mw.DesktopArticleTarget, but that's where we already have
similar logic for the edit summary. I filed T253696 about this.
Bug: T250388
Change-Id: Ia6a9c0465ed215e8f74b9fff4590593383e9a1e6
Previously we didn't deactivate them, so if you opened VE, then
switched to NWE, then exited the editor, all three tabs would appear
active.
Change-Id: I904d6daf2896ceadf004f5e57a88c2359f33fd44
Use hasContent to
1. Catch cases where the document is empty, e.g. <p></p>
2. Avoid having to use the converter
Change-Id: Ib1bb36824ca871e535bef38cef8137fdfb81b53e
Move shouldShowWelcomeDialog() and stopShowingWelcomeDialog() from
DesktopArticleTarget to DesktopArticleTarget.init, and use them to
deduplicate code in init that manages the wikitext welcome dialog.
Look for both the vehidebetadialog and hidewelcomedialog URL params.
The code in DesktopArticleTarget used vehidebetadialog, but the code in
init for the wikitext welcome dialog used hidewelcomedialog.
Bug: T249954
Change-Id: I19f1a2da36bc65addb52811c3d3c73c1259bc8f5
The base class doesn't use it, it only defines it, and
DesktopArticleTarget is the only subclass that uses it.
MobileArticleTarget calls it, but also overrides it to be a no-op.
Change-Id: Ib3feea94844f4e1ed71dccece7657450516cac89
Factor out the logic for whether the WelcomeDialog should be shown into
its own method, and write it in a less confusing way. Do the same thing
with the logic for setting the preference/storage/cookie for hiding the
WelcomeDialog.
This makes maybeShowWelcomeDialog() much simpler, and removes duplicated
code in DesktopArticleTarget.
There is one minor change in behavior: if the WelcomeDialog is
suppressed using the URL parameter, that no longer causes the preference
to be updated as if the dialog had been shown.
Change-Id: I1d4f912c5f6bd7a2bbad2b209b97c3ec1f250a07
In MediaWiki, section numbers may be prefixed with 'T-' if they refer
to sections on a transcluded page, so they are not really numbers.
Change e2cb9ce93e caused us to treat them as strings most of the time,
but it looks like there are several places where we treated them as
integer numbers, which I missed when making that patch.
The first two changes in ArticleTarget#restoreEditSection fix T248795
and T248968/T249112, respectively. The other changes are cleanup.
Bug: T248795
Bug: T248968
Bug: T249112
Change-Id: I8373a7ab515595769ce6f3051a182c922415b643
Bring in ve.dm.MWInternalLinkAnnotation.static.getTargetDataFromHref
and ve.resolveUrl, so that the file has no dependencies on VE.
Change-Id: I03bc455d5484a6c51f3fa2397c64936b829fe7e3
This code path is triggered when saving a new section in NWE causes an
edit conflict, which should be impossible, but apparently does happen.
Add some TODOs for parameters being added to the API calls in weird
places, instead of the dedicated methods.
Bug: T248364
Change-Id: I0686671e86e35f9ba503d0dd84e9074dde72dc10
New changes:
cf72879d2 Show sum/average when selecting multiple numeric cells
Local changes:
Implement number parsing/formatting for table summing
Bug: T247877
Change-Id: I52af622dc8cfe7e77fd7ce88428be092d5b092a2
New changes:
8a3e25b98 build: Updating acorn to 7.1.1
032a2520d Localisation updates from https://translatewiki.net.
ff82c0966 [BREAKING CHANGE] Remove ve.init.target lookup from TargetWidget
Local changes:
* Pass toolbarGroups explicitly form Target to TargetWidget
Change-Id: I4ff6f432412ab958f2858879b2b857278866350a
This should be a no-op as the inherited switchToWikitext
implementation will always call switchToFallbackWikitext
if 'source' is not passed as a supported mode (which it
isn't currently in MobileFrontend).
Change-Id: I213e7d54d158127b5c42bc05ff9ea2dececc42fe
The code for setting 'watchlist' in the EditAPI request
was completely broken as it always evaluated to 'unwatch'.
Instead pass through 'watchlist' directly from the client
where it must be set to 'watch' or 'unwatch'.
Bug: T245579
Change-Id: Ia5a2bb76ef35a685b39bcc0c4727796acd0f510d
When using TwoColConflict with VisualEditor, the autosave buffer wouldn't
be cleared after saving the successful merge. This would cause a user to
see a "restore changes?" prompt the next time they entered the editor,
with the potential to confuse them and cause them to do extra work,
repeating the conflict resolution unnecessarily.
This change purges the autosave buffer before submitting a merge.
Note that it is not transactional, so there is a chance we're losing the
autosaved content even if the submit will fail.
Bug: T245119
Change-Id: I150023f548c5565412769d644a828176f907bc25
Things I noticed while writing I37f8e89b6d92c419d1b6569891612256342f8139,
but which felt too messy to include in that commit.
* Use promise chaining
* Update documentation
* Remove redundant code
* Split a method that now handles two different errors
* Grumble about localisation messages
Change-Id: I81e28a03af4f6c3452679ef6bbcaa89bb1235122
