Notably:
* Don't require the model in the new sidebar via dependency
injection, but connect the event handlers later. This is
relevant because we currently create the new sidebar in the
wrong spot. Removing the hard dependency allows us to split
the code and utilize initialize() and getSetupProcess()
correctly. This will be done in a following patch.
* The change event now includes the new position. This makes
it very easy to add this missing feature to the new sidebar.
Also:
* Stop triggering change events when nothing changed. These
events are expensive. They bubble all the way up to the
TransclusionModel, and to all linked
onTransclusionModelChange() handlers.
* Update event documentation to make this more visible.
Bug: T274544
Change-Id: Iafe29f18a6fed14d9c3124c9756aa840886afbbc
Notably:
* Include parameter aliases, labels and descriptions in the
search.
* Don't use a possibly outdated search index, but live data.
* Clear filter when a new checkbox is added.
Bug: T272481
Change-Id: Ie90a803af6178a8bb6de370a0f8e079800d9f8a2
In detail:
* Allow clicks on all elements in the new sidebar. This should
focus the corresponding element on the right.
* Make all elements in the new sidebar tabbable.
* Fix MWTransclusionOutlineTemplateWidget.createCheckbox() to
not need a temporary param object any more.
* Rewrite more code in MWTransclusionOutlineTemplateWidget to
be shorter and easier to read.
* Fix MWTemplateModel.addParameter() to not do way to much
stuff when a parameter already exists.
* Update code documentation.
* Use more specific, less ambiguous variable and method names.
Bug: T274544
Change-Id: Iaf6f7d1b0f7bf0e9b03eb86d01f3eceadece6fe4
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
Separation of concerns:
* The template model knows which parameters are currently used,
but doesn't know what's documented.
* The spec knows what's documented, but doesn't know what's
currently used.
Change-Id: I97cac00d6775a17a07059d0e8a7a116adc6080b3
For example, checking if a parameter is required works just fine
for unknown parameters. They are never required. Since I16708b0
we don't need to guard the spec related methods any more.
Change-Id: Id90e4cb810dc9faca3b26f122a534f276ee31709
Before the method fetchRequestAlways() was doing two entirely
different things. Note how the two function arguments are
split now. Each method uses only one of them.
Change-Id: I592a1f29fd9c677a0ff18115cccda36950172001
These methods are special in so far that they create *minimal*
wikitext where optional whitespace is not preserved. I tried
to rename the methods to reflect this, but could not find a
caller. What's used instead are the .serialize() methods.
Bug: T284895
Change-Id: Iedaa5b7efa9675151cc0553854d8aef3f9a46cbb
A lot of the checks are redundant. The first check still is
redundant because the later two cover everything as well. But
I left it for performance reasons.
Additionally:
* There was no test for the method.
* This patch also updates a few pieces of documentation in the
same class.
Change-Id: I10f2944a844cc070bdc08dec6719929b383e34fa
If a known parameter is present using one of it's aliases, then
only the aliased name should be shown to the user. This patch,
therefore, resolves the issue of the same parameter being added
to the sidebar twice.
When adding a parameter that is aliased, it will receive the same
position as the non-aliased parameter it is replacing.
Bug: T274545
Change-Id: If4e58c941fd0f0e690d3603935f5a5d3f9938163
This also removes a few lines of text that don't explain
anything that would not be obvious from the code or @return
tag anyway.
Change-Id: I2f8f02dd61c50d9990d72c0e8ea79d679c9b11f2
This fixes a minor issue in the spec class. In the first step,
parameters from the template are added to the list of known
parameters. Later, aliases are resolved. The original behavior
was that such a parameter moved to the end of the list. This
is rather unexpected.
This dosn't have much of an impact. The pretty much only place
where the parameter order from the spec can be seen is in the
parameter search widget. Still I believe it's worth fixing.
Bug: T285483
Change-Id: I455818451811e92bba3e9320c2d41e1db8d563f2
I don't want this code to crash when the TemplateData API
returns an unexpected result.
Bug: T285483
Change-Id: I237cbfbb85892a53a08d9e7e34cf4974775d627a
This is just not necessary. It removes a level of indirection
that possibly makes it harder to understand the code. It makes
it easier to possibly get rid of unused methods.
Change-Id: Iaf8b213a5e1ae64a24b5bcdf2a0b200d5d3cbf46
This doesn't "extend". It was never used like this. What it
actually does is to link between a (cached) TemplateData blob
and the spec class that want's to use it.
Is this the best possible name?
* fillFromTemplateData( … )?
* propagateTemplateDocumentation( … )?
* readDocumentationFrom( templateData )?
* …?
Do we want to rename the "spec" class as well?
* MWTemplateDocumentation?
* MWTemplateMetadata?
* MWTemplateDataAccessor?
* …?
Bug: T285483
Change-Id: I6c52ef42d411c2f47fc0080768d36ebda4dd2a55
Just store the JSON blob from the TemplateData API as is.
This comes with a bunch of nice consequences:
* Less code.
* Less class properties that don't do anything but copy what's
in the TemplateData blob.
* Easier to understand what's going on. The `this.templateData`
property is now a reference to the *actual* TemplateData
documentation.
* No need to cache the documentedParamOrder. Just do it when
needed.
This also removes an unused feature from the `extend()` method
that didn't made sense anyway. Before it was possible to merge
conflicting documentations. But this is not only unused, it's
impossible to have multiple documentations for the same
template.
The method acts as a straight setter now. The next patch will
rename it accordingly.
Bug: T285483
Change-Id: I3ffc202577e9a20fc7491234601ccd981113f866
Instead of faking entries in this.params, let's use a separate
tiny data structure to keep track of parameters we have seen so
far, and in which order.
This finally allows to easily distinguish between documented and
undocumented parameters.
Bug: T285483
Change-Id: Idf62b0661178a3bbef7e817edf016dbd572d415b
The so called "spec" class keeps track of parameters that have
been used before, no matter if documented via TemplateData or
not. Removed parameters are still "known" (i.e. have been seen
before).
This feature allows to easily find previously used parameters
names when an undocumented parameter was removed and the user
tries to add it again.
Bug: T285483
Change-Id: Ia1555eea87cd99e7a3f386f4279ec5a80fb98a79
These tags don't do much, if anything. But they provide a hint
in which scope a method might be used.
Bug: T284895
Change-Id: I0b4bdd416ee89d26961c4ded4d8bbace8c57da76
In I04b8a14fbec7be5a1c4defabf92e94f694c1e638 we sepearted params from
aliases. There we missed that re-filling the parameters from the
template could re-add the aliases.
Bug: T285483
Bug: T285843
Change-Id: I1928b443a5f708bc8c57efa5ad0a86b5915b159c
While the term "canonical" is not wrong, I find it still
somewhat ambiguous.
1. "Canonical" could mean different things. E.g. is the order
of parameters as they appear in the article's wikitext the
"canonical" one? It's possible to argue like this, esp. if a
template doesn't have TemplateData documentation. In this case
the only order known is the one from the wikitext.
2. "Canonical" sounds like the parameters must be reordered.
But this should never happen. Not having dirty diffs is more
important than having the parameters in a specific order.
Bug: T285483
Change-Id: I23658d37fea50b727667677ac6a49066673b2135
This property is a reference to a static variable with the
same name, initialized at the very top of the file. All
instances of the class use the same cache. They all use the
shared specCache directly, not the reference.
Depends-On: I0084410b7eab29048451ad67c18d6c2180c4f1b1
Change-Id: I9fd79ce3abd533dbb48a210e596802ea9e692855
These comments don't add any knowledge. The text is either
duplicated, or the method signatur says it already. Having
to read these comments just to realize that they don't give
any additional information is not helpful, even error-prone.
Change-Id: I014028b1e9311b831a22c37859b2130aed2e9539
Wait, what's going on here? This patch looks like it changes the
behavior of this code. But it doesn't. Here is what happened
before:
* Let's say a template contains 2 parameters, A and B.
* We don't know yet if these names are aliases.
* getParameterNames() returns [ "A", "B" ].
* extend() is called. The TemplateData documentation contains
the parameters "B" and "C". "C" does have an alias "A".
* extend() can't find "C" and adds it to the end, as if it's a
new parameter.
* extend() also iterates the aliases. For each alias it creates
a reference to the specification object. In this case a
reference from "A" to "C" is created.
* But "A" already exists. The position of "A" doesn't change,
but the specification now says it's an alias.
* getParameterNames() skips aliases. It skips "A" and instead
returns the new "C" from the end of the list.
This was the behavior before. It's unchanged, proven by the tests.
Change-Id: I04b8a14fbec7be5a1c4defabf92e94f694c1e638
The idea is to not actually store all these default values, but
fall back to the default only when needed.
Some more details:
* The only remaining property is ….name. The only reason to
have this property is to distinguish between aliases and
primary parameter names. This will be reworked in a later
patch.
* The description falls back to null because this is the
documented fallback, not undefined.
* The default value falls back to "", same as the auto-value.
Why not null you might ask. This is intentional. Both the
auto- and default value are effectively wikitext snippets,
while the example is a label in the VE UI.
Bug: T285483
Change-Id: I1be3cca18f9ad6fc1c16362b24633f7613f02539
This is done for two reasons:
1. It fixes the behavior of two methods in rare edge-case
situations. They aren't documented to return undefined.
2. It reduces the amount of stuff this class stores when it's
nothing but a default value anyway. Note this patch does this
for the template-level properties only. Another patch will do
the same for the parameter-level properties.
Bug: T285483
Change-Id: If2e4d56da1fa52e32dc94191f36d7dc6a1487829
This reflects much better how this method is meant to behave.
Note I will continue to remove documentation that doesn't
explain anything in addition to what the code already says.
Bug: T285483
Change-Id: I81fa8a5d9d0752f3aeac4015c9a27b50e054d4df
This patch also marks 2 methods as @private that are not and
should not be used outside of this class.
Bug: T285483
Change-Id: I8a8ffc4868a369b5c47068beb0e83f023872543d
This reverts the revert commit d47b95eb4a.
When no `paramOrder` is given, known parameters should appear in the
order returned from the TemplateData API.
Previously, when TemplateData was present but no paramOrder
specified, then the parameters would appear in alphabetical order as
"unknown" parameters. Now they will appear in the order listed in
TemplateData. This is similar to the fully-specified behavior when
paramOrder is present.
This will only affect the Visual Editor template dialog, and has no
effect on serialization.
Bug: T274545
Change-Id: If8315781572af688ea1c1b14b3694b828f076b4a
This makes the code more readable and easier to reason about.
The ESLint rule responsible for this code style was removed
just recently.
Notes:
* I focus on classes that are relevant for what the WMDE team
does right now.
* I merge multiple `var` keywords only when the variables are
strongly connected.
* Caching the length in a for loop makes the code hard to
read, but not really faster when it's a trivial property
access anyway.
Bug: T284895
Change-Id: I621fed61d894a83dc95f58129bbe679d82b0f5f5
When no `paramOrder` is given, known parameters should appear in the
order returned from the TemplateData API.
Previously, when TemplateData was present but no paramOrder
specified, then the parameters would appear in alphabetical order as
"unknown" parameters. Now they will appear in the order listed in
TemplateData. This is similar to the fully-specified behavior when
paramOrder is present.
This will only affect the Visual Editor template dialog.
Bug: T274545
Change-Id: I32538de07641c288081042a41fe39eedfed7d939
Note that the tests expose a bug, getAllParametersOrdered fails to
list an unused parameter. Fixed in I32538de07641c.
Also, a minor fix to avoid an impossible template spec: paramOrder
must include all parameters.
Bug: T274545
Change-Id: Icfa7a765773d04ef05a76ecc09467305e311f6cb
There are at least 3 different methods that are all named
getWikitext, not counting subclasses. They behave rather
different, most notably in terms of whitespace preservation.
Bug: T284895
Change-Id: I8b47f5bd21675a431ba2bc2d4a8cb0c55dd50f76
Most notably:
* Introduce variable names that explain much better what's
going on.
* Reduce nesting.
Bug: T284895
Change-Id: I793677d8107abb6354f9e19d79c4879a41c4bd93
Splits out a useful intermediate calculation from getOrderedParameterNames,
exposing the full list of parameters including those that are not
present in the transclusion.
This will be used to build the sidebar checkbox list.
Bug: T274545
Change-Id: I1c6a9ea8a5e9a163751fee87f974f63c72fd1f61
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
This class represents a raw wikitext snippet. There is also no
base class that would require us to follow a generic
getValue/setValue naming scheme.
Change-Id: I0891a2f6c0ae0121429a47c39221e99b9653e8e3
There are 2 methods with the same name, but they are very
different. This makes it much easier to understand the
difference, I hope.
Change-Id: Ie1f049b2b14e1fe23f078e281ee797da29dfe3db
The variable `html` had the value of undefined and was treated as a string.
This would then be displayed on the editing surface.
Change-Id: I4682ea121aa37f06cac41dde618af847586ae01e
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
The idea is to possibly rename some of these classes, based on
these descriptions. But this should be done in later, separate
patches.
Change-Id: I7f9e5b2382711b434d6dd618489fa3ed8b7a46b4
Returns true if there is no meaningful user input yet.
Will be used in the next patch.
Bug: T272355
Change-Id: I4f88ce31662bbc46755f78d574c46b907581d438
Previous, reverted attempt: da9b6fffbd.
This attempt also includes 6037fefbe0,
and fixes minor conflicts with other changes.
* In normal images, parse relative 'href' attributes instead of
expanding them to absolute. This resolves Parsoid generating
|link= options for copy-pasted images (T193253).
Keep them in the underscore-form to avoid causing dirty diffs like
T237040 again. Unlike in the previous attempt, we don't need to be
super-careful about the 'resource' attribute, thanks to the Parsoid
changes in T108504.
* In gallery images stuff, prefix the 'resource' attribute with './',
same as normal images do. This causes no functional changes, but it
makes updating tests easier, and the consistency is probably good.
* Update test examples to also prefix 'resource' and relative 'href'
attributes with './', like the real Parsoid does.
Bug: T193253
Change-Id: I91131728a87c9406bf069d46d3c94c9a8905a003
Similar to Ic79aba4d4364227c3ecf7fb5411e90532b531f44
This only works if the gallery goes unedited. Probably something needs
to be done in ve.ce.MWGalleryImageNode if we care to be complete.
However, as noted in T214648, the DOM diff'er doesn't traverse into
gallery content and notice these element names. So, it's purely
academic to be doing this anyways.
Bug: T266143
Change-Id: I37799076852fa6f062c9d85bcebb15998fb44a80
This parameter name was deprecated and replaced in 1.31. See also
Ie5fe2097cda45968bb080643d3afcac0b2868a6c
Change-Id: Ie9d6c70d3dfe3954504d3d698c122dceede7603d
Parsoid sometimes emits malformed links (with no 'rel') when a
misnested <figure-inline> tag is moved around. Converting them to
internal links, and adding the 'rel' attribute, makes the element no
longer match in selser, and causes dirty diffs. Alienate them instead.
Bug: T64473
Bug: T150196
Bug: T267282
Change-Id: Ic7b48eb2e61585445a1fb98dc2b516d3b6da3cc4
Make ve.dm.MWTemplateModel#serialize ignore empty parameters if they were not
present in the transclusion before the edit. This avoids dirty diffs where an
user edits a template transclusion via VisualEditor, and the editor adds all
available template parameters to the edit wikitext, even if they were not
changed during the edit.
This logic was ported from the old Wikia-WMF VisualEditor project.[1]
Additionally, add tests for ve.dm.MWTemplateModel serialization.
---
[1] https://github.com/Wikia/app/pull/6450/commits/858eaa9
Bug: T101075
Change-Id: I35f8812724658904d30034db4e4684193a661c1e
Like transclusions, extensions can contain nodes of other types. This
hasn't been an issue because the legacy parser doesn't generate content
with these annotations. But, moving forward, as Parsoid becoming
responsible for more extensions. This is the case for the new ImageMap
implementation.
Matches I95767e466803f0744b6626204a0a3a1514fff174
Change-Id: I6ff81a01207e2734090c626b177e5f4d10bb6d61
I'm not 100% sure, but this looks like a copy-paste mistake to
me. Something like this (a subclass modifying the base class)
is not done anywhere else.
Change-Id: I24677c2deb721b68d1b534f1569c925b386d4d3d
The mw:Placeholder attribute semantically means, "don't touch this,"
but french spacing should be freely editable. It's just a funny way
to write a plain wikitext space.
Bug: T254502
Depends-On: Ia164dd1318d45924aa965919e7939c6f817f5d0d
Change-Id: I56e0f0c6526649ea041e023698a48936176dec4b
We don't share much functionality, and we have to do a bunch
of hacks to disable functionality we don't want.
Change-Id: I9861123d8f1cbab1923f1aa5be713c2dadaed53d
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
<includeonly /> is not valid wikitext markup, which is distracting.
Let's use <includeonly>…</includeonly> instead.
Bug: T250937
Change-Id: I9eac8b265eca16118d390da2628e0da7ca409ed8
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
The difference is that metaitems are not visible on the editing
surface, and their exact position is not preserved when the paragraph
containing them is edited.
This behavior is desirable for e.g. categories, but not for
<noinclude> and related tags, which are intentionally placed in
specific places in the text.
Note that we don't really have any editing interface for these nodes
yet. But you can see them (and they come with descriptions and links
to documentation pages), and delete or copy-paste them.
Bug: T250937
Change-Id: I104e7abbd650567df0e59813653c46a66d955d58
Bring in ve.dm.MWInternalLinkAnnotation.static.getTargetDataFromHref
and ve.resolveUrl, so that the file has no dependencies on VE.
Change-Id: I03bc455d5484a6c51f3fa2397c64936b829fe7e3
By removing this line, we fall back to the default behavior,
which is to copy the attribute from the original DOM element.
The gallery is supposed to have a class indicating the type (packed,
traditional, etc.). However, Parsoid doesn't care about that and
instead reads the type from 'data-mw'. Instead, changing the attribute
is causing dirty diffs.
Bug: T214649
Change-Id: I96b5a21777046b1caf07a3b1def9fad81bb15939
ve.dm.MWImageNode:
* Define sensible scalable properties for audio files. They are now
scalable to any width but have a fixed height. (Ideally they would
have no concept of height, but that would require many more changes.)
This prevents them from resetting to 0x0 when resized.
ve.ce.MWBlockImageNode:
* Remove override for #isResizable, audio files can be resized now.
* Move #updateMediaType to MWImageNode mixin so it applies to
MWInlineImageNode as well.
ve.ce.MWImageNode:
* Add #updateMediaType from MWBlockImageNode.
* Hide the real image 'src' using CSS rather than changing the
attribute. It seems the previous solution depended on the order in
which methods are called, because it stopped working when I moved
the code here. (This depends on VE/VE change If5b1b5b5d.)
audioPlayer.svg:
* Make the file nicely resizeable. The dimensions of the "play" icon
and time are fixed, the bar adjusts to the width of the container.
Bug: T206022
Change-Id: Ia0f38ca11e0d55a5b725fd9aeb6c79ec1345376d
The previous attempt to fix this didn't preserve any attributes
but removing data-parsoid can result in a loss of wikitext formatting.
This reverts commit bdfd4b6d8f.
Bug: T207325
Change-Id: I2a38e651d17262889eddb149c72c9e08b4e56ed0
When a template does not have user-provided TemplateData documentation,
the TemplateData API falls back to extracting possible parameters from the raw wikitext
to generate an API response with a list of potential parameters. However, it also
sets the "notemplatedata" field in the response, causing the VisualEditor to think
the response contains no useful information and ignore it. This appears to have been
an unintended side-effect of I97a1bfc9f9ead082a673a91b9d2053630a90309c.
This patch ensures that the VisualEditor will correctly consider such responses from
TemplateData by modifying ve.dm.MWTransclusionModel to check if the response contains
a parameter map. Some unit tests were added for the class to verify this behavior.
Bug: T243868
Change-Id: I72005880d9301a53224473900efe2917379e8708
It's supposed to be non-editable but deletable text, like mw:Entity.
We decided to handle them this way in 2015 but never implemented it
(T94509). Currently, accidentally editing text inside of
mw:DisplaySpace node causes the changes to be lost when saving.
Bug: T241906
Change-Id: I78a0cc7a75061a7eefb8b677898b5756326615d6
* In normal images, parse relative 'href' attributes instead of
expanding them to absolute, and parse 'resource' to keep it
identical to 'href' if they refer to the same page (including
same percent-encoding and space/underscore). This resolves Parsoid
generating |link= options for copy-pasted images (T193253).
* In gallery images stuff, prefix the 'resource' attribute with './',
same as normal images do. This causes no functional changes, but it
makes updating tests easier, and the consistency is probably good.
* Update test examples to also prefix 'resource' and relative 'href'
attributes with './', like the real Parsoid does.
Bug: T193253
Change-Id: If2d7f080d9d693568054f8311c1e1b15ca27ea5c
* Handle mw:MediaLinks pointing to to non-existent files, which come
with typeof="mw:Error" (similar to image nodes).
* Fix regression from c66f8e0547, which
caused all mw:MediaLinks to be treated as plain external links again.
* Add test cases.
Bug: T232754
Change-Id: I9ae5bcfc4e24e8c0d22ef77d6a4d03f817fc9768
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
As detailed in T95850#5078990, Parsoid incorrectly converts HTML to
wikitext when a link `href` refers to an interwiki page and contains
percent-encoded colon characters ':'. VisualEditor doesn't know
anything about interwiki pages (it treats those links as normal local
links, and expects MediaWiki and Parsoid to handle them specially),
so we can't easily special-case them. But since leaving the colon
unencoded is valid for all links anyway, we can just do that.
Bug: T103635
Change-Id: I87d7e6952983a72d90ea739b0bc8488d9f6a9be3
The likely case for this is: copying from within VE in one wiki, and pasting
into VE in another wiki. This change will notice this happening, and fall back
to treat it as an external link. (For the wiki-internal links, this will turn
them into interwiki links rather than raw external links.)
Bug: T223322
Change-Id: Ie0157fc3aee6e5fd9973a2889be7ebd287bc90a5
Parsoid does not use relative links anywhere anymore (T72743). There
is no reason for us to support this. And previous code allowed
"hrefPrefix" to be empty '' sometimes, which is scary, as it could
lead to XSS vulnerabilities if titles starting with 'JavaScript:' are
not handled correctly elsewhere.
Bug: T206357
Depends-On: I8728f63084902c76d1c61193be4367939b069f1a
Change-Id: I99be18877aae2b505cf261bd7cdef6cf0d7a8670
We could generate incorrect links to pages whose title contains a
colon ':' and therefore looks like a fully-qualified URL.
Bug: T206231
Bug: T206357
Change-Id: Ie34694d903a6d97589cc46417f70659559494619
This reverts commit be628a5b7e.
87b20f9b Revert "[BREAKING CHANGE] Do not cache document model data in DM selections"
Change-Id: I47bbf757a4ad227346d3734f6e50d928a2de1409
New changes:
ccb4de82c [BREAKING CHANGE] Do not cache document model data in DM selections
Bug: T208228
Change-Id: I564399ad864751d1690077b45a06e098b5509a93
I am surprised this was disabled. I investigated this after reviewing
some code by a new contributor which I was certain should have failed
the lint check, but passed.
Change-Id: I5b3c837b8ca3292f6e268b3922443bd9587eadbe
Split on regexp for whitespace instead of a single space. Avoids multiple-
spaces causing `'foo bar'` to become `['foo', '', 'bar']`.
See also: I1f467f51017e2deae30905163bf5e6b07048cecf
Change-Id: Id7a887a20fac99715b79045f01e861b4efe9f2c7
I ran Closure Compiler over the codebase just to see what would happen,
and it printed some useful warnings.
Change-Id: I56d40b11e6d1dd7ce68a5e59da511f66e928647f
