Previously we only looked at error info/code from the root of
the API response, not the ones from the root of the response
that action=visualeditoredit forwards from action=edit.
This changes the message for e.g. AbuseFilter from:
> Error: Invalid error code
to:
> Error: Hit AbuseFilter: [name of triggered filter]
Also changed default error for other reasons (e.g. hooks of other
extensions that we don't yet have specific support for) to "Unknown error"
instead of "Invalid error code".
Bug: 50472
Change-Id: I3b32eddafd8fff83f323606f9922ac60b5d3b58e
* Rephrased visualeditor-savedialog-error-badtoken to emphasise
that it is the old session that become invalid, not the one
the user started browsing with since in a different window.
* If the session changed, the user will be asked whether they
agree to save with this new session instead.
* We explictly update mw.config so that future save attempts
in the same window compare against the correct environment.
Without this there are two problems when saving and then
making a second edit in the same window and saving that:
- It will bring up the same question again (user A -> user B),
which is annoying.
- If the user logged back in again (new session, but for
user A again) it would silently try with that new token
without asking, thus saving as user A when the user still
thinks it switched to user B. It switching back automatically
is not obvious since we asked them from A->B, so we should
also ask the other way around.
This can be reproduced by opending ve-edit logged-in, then
logging out in a new window, save, confirm anon, save,
open edit again, log back in in a new window, save open edit
in the old window, confirm new logged-in, save.
Bug: 50424
Change-Id: Id055eca1886f85aeaf615f645de29898afc0373c
An empty document is one which contains no 'real' data, so
we should check for meta-only documents when deciding whether
to add in a wrapper paragraph.
Bug: 50289
Change-Id: Ib3ebf0717aa0c6c51fd1d0b14e95de50b2842647
Without making the code much more complex (and possibly create
performance issues) the warning will fire on pages which already
contain escaped wikitext (when that text is edited). I think this
should be a small enough minority that it won't be an major annoyance.
Bug: 49820
Change-Id: I0f67ec04b890f4add9247be6126bdc086b6ae72f
Adjacent annotations should not be merged if they both
originate from Parsoid. This is a hack because this logic
should be in Parsoid, not VE.
Bug: 49873
Change-Id: If1e23e3039178300d72b1c0c585931417bb603b5
Objective:
* Make the majority of link inspector, which is generic to any annotation,
usable for other annotation inspectors
This was merged earlier (f7107fa20d) but broke master, so it was
reverted (092fa74dee). This commit also incorporates 5dcf5d1c49.
Change-Id: Ib9190dee66ce064d69962f9c4c5b3a710be8ad07
Problem:
Adding or moving templates or content resulted in the incorrect item
being selected.
Diagnosis:
Although recently we solved a few issues by making addPart async, it
caused some other issues where callers of addPart were assuming
otherwise
Solution:
Return a promise from addPart which is resolved after the part is
actually added and setup callers to use the promise when needed
Changes:
ve.ui.MWTransclusionDialog.js
* Use promises to auto-select new or moved parts
ve.dm.MWTranclusionModel.js
* Make addPart return a promise
* Resolve promise when queue is processed
* Automatically remove existing items before adding them in different
locations at the time of processing the queue we don't yield between
removing and adding
Change-Id: Ie43c5b89ca4ed34d5f797714e19c9a7e1824cdec
Problem:
Parsoid has a property called "i" which we don't use, but they need for
round-tripping purposes. Since we were generating a structure from
Parsoid data and then generating data from the structure without
preserving properties we didn't use, it was getting lost.
Solution:
Abstract creating a template from data vs. creating it from name. Make
only templates have an origin argument in their constructors, so and
set it within a set of static constructors that create a template for
either data or a template name. Store the original data in the former
case, and use it as a base when serializing.
Changes:
ve.ui.MWTranslcusionDialog.js
* Remove no-longer-needed mw global declaration
* Move most of the addTemplate function to a static constructor in the
template model class
ve.dm.MWTransclusionPartModel.js,
ve.dm.MWTransclusionContentModel.js,
ve.dm.MWTemplatePlaceholder
* Remove unused origin argument/property/getter
* Add serialize method (if needed)
ve.dm.MWTranclusionModel.js
* Move template/parameter generation from data into static constructor
of template model
* Move serialization to part classes
ve.dm.MWTemplateModel.js
* Add mw global declaration
* Stop passing origin to parent constructor, store it locally instead
* Add original data property/setter for preserving unused properties
when round tripping
* Add static constructors for generating a template from data or by name
* Add serialize method
Bug: 51150
Change-Id: Ide596a0ca0ae8f93ffce6e79b7234a1db7e0586c
Previously all it did was surface api response error.info,
which surfaced underneath the edit summary form as mere:
"Error: Invalid token".
Bug: 50424
Change-Id: I60169b42701ae3b88e54626c4ff7050549e6ef55
This allows us to make the token no longer a requirement
for non-save actions while still using the built-in system
for token verification.
Update documentation for "oldid" since it is not required even
outside (paction=save). In fact, if we require it VE loading
fails because it doesn't pass oldid unless it has to when
restoring a specific version.
Bug: 50424
Change-Id: I7b1b50a43648b1cc40a984340846efdb0ba2ecc9
Objectives:
* Automatically add required parameters to templates that users create
using the GUI, without touching existing templates loaded from data
* Cleanup some confusing terminology and APIs
Changes:
ve.ui.MWParameterSearchWidget.js
* Remove special logic for skipping aliases, which are no longer included
in the list of names given by getParameterNames
ve.ui.MWTransclusionDialog.js
* Add origin arguments to constructors of transclusion parts
* Re-use onAddParameter method during initial construction of parameter
pages
* Add required template parameters for user created template parts
ve.dm.MWTransclusionPartModel.js
* Add origin argument/property/getter for tracking where a part came from
ve.dm.MWTransclusionContentModel.js,
ve.dm.MWTransclusionPlaceholderModel.js,
ve.dm.MWTemplateModel.js
* Add origin argument pass through
ve.dm.MWTranclusionModel.js
* Add origin arguments to constructors of transclusion parts
ve.dm.MWTemplateSpecModel.js
* Rename origin to name - was a bad name to start with and will be even
more confusing with the new part origin property
* Add isParameterAlias method
* Make getParameterNames only return primary names, excluding aliases
ve.dm.MWTemplateModel.js
* Update use of parameter origin, now called name
Bug: 50747
Change-Id: Ib444f0f5a8168cd59ea52a6000ba5e42ccdc2a24
Also in this commit is a fix to the regex for brackets in general
as previously it was grouping multiple pairs together which was
making the stack count incorrect.
Bug: 50801
Change-Id: I2cd825eaa7242e37a6317286cb8f9a4a78e693e2
Some of the ones that were moved to ve-mw disappeared from index.php
(the MW-only test runner) and/or VisualEditor.hooks.php (the runner for
Special:JavaScriptTest that's also used by Jenkins).
Add the missing tests, and rename the SurfaceFragment test. We can't
add the ViewPageTarget test to the index.php file in ve-mw because
it only works when running inside an actual MediaWiki instance.
Change-Id: Iec34f2029f5f9c34855b4d79ef70db5751461a96
This is harmful because we don't know whether <span> is acceptable
in context. They're not allowed in <tbody> for instance, and so if
we replace a <tr> with a <span> that breaks things.
Instead, use the nodeName of the first original DOM element.
Change-Id: I95ea9f13985767123f692706c6cc71fefd74d517
Objectives:
* Trim leading or trailing whitespace that parsoid may have left on
parameter names
* Preserve the original name for round-tripping cleanliness
* Ignore leading or trailing whitespace when entering new parameter names
in the parameter search widget
* Consider aliases when listing suggested parameters
Changes:
ve.ui.MWParameterSearchWidget.js
* Use hasParameter method instead of using indexOf - uses map lookup
internally, which is much faster, and also take aliases into account
* Trim query input value to prevent leading or trailing whitespace from
being considered when filtering known or creating unknown parameters
* Take aliases into account when showing filtered results
ve.dm.MWTransclusionModel.js
* Use original name when round-tripping
ve.dm.MWTemplateParameterModel.js
* Store original name for round tripping, and trim the original name for
other uses
* Add getOriginalName method
ve.dm.MWTemplateModel.js
* Add hasParameter method, which currently just does a map lookup, but can
do other processing in the future
ve.dm.MWTemplateSpecModel.js
* Add isParameterKnown method
Bug: 50715
Bug: 50717
Change-Id: I36a5e93ca8938ac3401a6e274647597704700468
The presence check used to be against the VE global, but we
recently made that impossible without providing an alternative.
Though by accident, mw.libs.ve has become the new way to check
for presence.
Change-Id: If85695525777a71dde467675052d2ede4e52c9b7
* Hooks for activation, deactivation, save dialog state change,
and tab setup.
* Class for save button (needed to point to it in a clean way).
Also done for cancel button to be consistent, though GuidedTour
doesn't currently use this.
Change-Id: I4a0e0631d513fb09c3408f2f36a0de0bd51e1a37
Objective:
* Cut off really long parameter names with ellipsis
Changes:
ve.ui.Widget.css
* Use ellipsis when option labels get too long
ve-mw/ve.ui.Widget.css
* Limit the size of a parameter result widget label - this is a hack, but
since there are strange bugs with using ellipsis together with % based
width, it's the best we can do for now
Bug: 50800
Change-Id: Idbfa336d2bba376b0d078fdd0ece65926d610d81
Follows-up Ic0c6d190c9b78 which introduced a full linmod copy in a
scope that is also accessible by other closures (namely the
callbacks to jQuery.Deferred #done and #always, and setTimeout).
Though in theory engines may be able to garbage collect it,
I doubt it. Though browsers probably destruct the setTimeout
callback I know at least the closures pass to jQuery.Deferred
are not released by jQuery, so an engine would have to
understand these other functions well enough to know it doesn't
access the `data` variable.
Let's release explicitly to be safe.
Change-Id: I11000edcad4690dcce53b6e9d1a45bf2ab82fbcb
Objective:
* Prevent input while the inspector is animating open
Changes:
ve.ui.LinkInspector.js
* Disable and then re-enable the surface while the inspector is opening
ve.ce.DocumentNode.js
* Remove opacity changes on disable/enable
ve.init.mw.ViewPageTarget.js
* Change the opacity of the document when save dialog is open
Bug: 51075
Change-Id: Ic7910a666b33b41b57b035a15cf1f8c9264e7111
Objectives:
* Make option widget icons correctly placed in RTL
* Make embedded context toolbar correctly placed in RTL
Approach:
Use separate elements for icons within option widgets which have more
flexibility when rendering in either LTR or RTL when compared to CSS
background position. The simpler approach, using CSS background
position offsets, isn't cross-browser compatible at this time.
Changes:
ve.ui.OutlineItemWidget.js
* Remove custom icon implementation, using parent class implementation
instead
ve.ui.OptionWidget.js
* Add icon option, which adds an icon element only if an icon was
specified
ve.ui.MenuItemWidget.js, ve-mw/ve.ui.MWParameterResultWidget.js
* Add override for icon config option
* Document icon config as private
ve.ui.Context.js
* Add detection and special handling for positioning embedded context
toolbar when rendering in RTL
ve.ui.Widget.css
* Add styles for option widget's new icon option
* Add styles to make indentation still work for outline item widgets
* Adjust styles for menu item widget as per changes in option widget
ve-mw/ve.ui.Widget.css
* Adjust styles for parameter result widget as per changes in option
widget
Change-Id: Ibfa4b613e0fd7902f8a2c78b5717de402c5f82b8
We previously manually loaded CSS into these frames, which is flawed
because it completely bypasses ResourceLoader (so CSSJanus didn't flip
them, necessitating a bunch of hacks for RTL), and doesn't pull in
MediaWiki styles (so templates inside references don't render correctly).
Instead, this commit copies all styles from the main document into each
frame's document, inlining what it can.
Loading all styles in dialogs and inspectors caused some problems,
initially. We didn't namespace our styles for dialogs vs. inspectors
at all; the only reason inspector styles weren't being applied to dialogs
and vice versa was because we controlled which files were being loaded
in which context. This commit namespaces the inspector and dialog styles
where needed so they don't conflict and try to override each other.
Tested in Vector and Monobook, but not in Apex and not in RTL.
ve.init.mw.ViewPageTarget*.css:
* Namespace styles that are only intended for the main document
* Undo Monobook's font-size: x-small; in frames
*Dialog.js:
* Remove addLocalStylesheet() calls, we don't need those any more
** ve.ui.MWDialog seems to be unneeded now, we may want to remove it
*.css:
* Remove @noflip-ped RTL rules where they were just flipped versions of
their LTR counterparts
ve.ui.Dialog.css, ve.ui.Inspector.css:
* Namespace styles with .ve-ui-dialog-content / .ve-ui-inspector-content
ve.ui.Frame.css:
* Move the margin:0 and padding:0 here (were in the frame <body>'s style
attribute) and add background:none to prevent frames from getting
the skin's background (grey in Vector, a book in Monobook)
ve.ui.Dialog.js, ve.ui.Inspector.js:
* Add ve-ui-dialog-content / ve-ui-inspector-content class to the
frame's content <div> so we can restrict styles to only apply in
dialogs / inspectors
ve.ui.Frame.js:
* Replace infrastructure for @import-ing stylesheets with transplantation
* Remove code polling to see when the stylesheets were loaded
** We can't do this in the new approach AFAIK, since all styles in the
frame are either inlined or inaccessible due to the same-origin policy
** We also shouldn't need it because the browser should have cached the
styles when it loaded the main document
* Apply ve-ui-frame-body class to the frame's <body> so we can style it
** Move inline padding:0;margin:0; into ve.ui.Frame.css
** Move the ve-ltr/ve-rtl class up to the <body>
ve.ui.Window.js:
* Remove infrastructure registering stylesheet URLs to load
Change-Id: I4a37115301811ad860f4578344a04873ea8c2b69
Whoops, merged too soon. The dropdown part of it doesn't
actually work right.
This reverts commit 3c51ebad93.
Change-Id: Ieafbf18ca8a43b07e33a787772abbc77aef63e4c
Conversion is apparently pretty slow on large articles, so put it inside
the setTimeout(). We still need to copy the data array synchronously
though.
Change-Id: Ic0c6d190c9b782f8c643d00d335f0e004d860bcf
Firefox 13 and 14 were the cause of links magically turning into [[./Foo]].
Once the immediate rush of deployment is over, we'll want to investigate as
to why and hopefully find a way to unblacklist FF.
Bug: 50720
Change-Id: I7a9dddb693091fa1e44b4325e77b9e4f55e5c193
The handler for the Edit tab already is in this conditional,
for edit section we were making the assumption that they only
ever appear on a view page, but that's wrong. They're also shown
on a diff against the latest revision of the page.
Bug: 50925
Change-Id: I802e548cbcdc03cfca66129466668854604bc3e7
* Only pass the oldid to the API from #load if we restoring from
oldid in the url. Otherwise load the latest version.
* Setting 'restoring' from mw.Target instead of mw.ViewPageTarget
so that we don't rely on mw.ViewPageTarget in mw.Target#load.
* Fix the API to not require 'oldid' to be passed.
* Fix the API to actually return the 'newrevid' property. It
was doing a no-op on a $result that is never used due to the
same variable being overwritten with the result of parseWikitext.
* Moved updating of wgCurRevisionId to mw.ViewPageTarget as it
belongs there (possible future inline editors probably act
on a different page than the main one). Also made it only
update if it isn't undefined, so that a null edit doesn't
result in wgCurRevisionId being unset.
Bug: 49943
Bug: 50441
Change-Id: I221e5038f95eadf6d87013e80f12394f0376a293
