The feature manager abstracts away how a feature is enabled from the
consumer of that feature. Accordingly, replace direct instantiation of
SkinVersionLookup with usage of the Vector.FeatureManager service.
Supporting changes:
- Add Vector\VectorServices, a simple wrapper around
MediaWiki\MediaWikiServices that allows us to both document and
type-hint services specific to Vector
- Add Vector\Hooks::isSkinVersionLegacy to minimise repetition
Additional changes:
- Make the MakeGlobalVariablesScript hook handler return early if the
user isn't using the Vector skin like the other hook handlers
Bug: T256100
Change-Id: I93b5ef39802323c7ac658af8fa7cc312fff68aa7
In PHP we add collapsible classes to all elements except watchstar
so that certain tabs can be collapsed under the more menu in JS.
This adds unnecessary complexity to our codebase and is not used
if JS is disabled.
To simplify this and bring Vector's PHP consistency with core this
logic is moved to JavaScript.
Bug: T259372
Change-Id: I2acbf7089198118626368ee8a37615d2de062f83
Please note I7e06a4cc226f3434c0f655212a464b8b98bcc7f4 should be
merged at the same time as this patch.
== The background ==
All extensions have been weaned of BaseTemplate hooks in
Wikimedia projects.
This change now means that Vector will no longer run
any BaseTemplate hooks. See the epic T253809 for the
implementation details.
== The change ==
BaseTemplate will now have nothing to do with the rendering of
Vector. The skin version is added to express the significance of
breaking compatibility with 3rd party extensions.
We TEMPORARILY remove SkinVector to retain git blame. SkinTemplateVector will
be renamed SkinVector in the follow up (see 2/2)
Update skin.json to use SkinTemplateVector for the skin (this will be fixed
in a follow up).
The isLegacy method is moved to SkinTemplateVector.
Changes of note:
* html-debuglog is no longer needed. SkinMustache includes this information on
the skins behalf
* html-printtail and html-headelement are now not needed in the master template
and added by SkinMustache
* Skin::getAfterPortlet does not provide the `after-portlet` wrapping element provided
by BaseTemplate::getAfterPortlet so this is added
* SkinTemplate::getFooterIcons does not support the options that BaseTemplate::getFooterIcons
does so any icons which do not have an image must be manually checked for and unset
Known changes to HTML output as a result of intentionally
delegating their output to the core SkinMustache class:
* A new line is removed between the body element and #mw-page-base
* #mw-html-debug-log now appears at the end of the body element
* #printfooter is now a child of #mw-content-text rather than sibling.
Bug: T251212
Change-Id: I4e89beb96f6401ed7e51bafdf0aac408f5a2c42f
- Creates a new user-preference called 'VectorSidebarVisible'
which stores the sidebar hidden/collapsed state for logged-in
users.
- Updates that user-preference on the client whenever the sidebar
is expanded or collapsed.
- Refactors the sidebar related javascript into a separate file.
Bug: T255727
Change-Id: Ib1ce934f3646cd8feebf0d3b15c38b5b969ec957
Adds a bundlesize test that measures the bytesize of
individual ResourceLoader modules.
This test depends on a running mediawiki instance as well as
the $MW_SERVER and $MW_SCRIPT_PATH environment variables.
ResourceLoader modules are fetched from a URL and piped
into a bundlesize command based on a custom configuration file.
The test can be run with `npm run test:size`.
Bug: T244276
Change-Id: I4f4534921ccbc6bd4f99229c8dd36b1279dde640
Begin our journey away from BaseTemplate by
moving VectorTemplate code to SkinVector. In future all
methods will live here but to lower risk, I've only targetted
the get method.
Bug: T251212
Change-Id: I58c2ff5edaacc2d5e45492c121cf0f87d08b623f
The .menu class historically only needs to apply to dropdowns.
the .vectorMenuCheckbox is inconsistent with the other classes on the
menu so we should begin its deprecation.
Bug: T253329
Change-Id: I00b4d2fd795195cd9c8add650a3b3cafdced5465
Use SkinTemplateNavigation hook instead and copy the collapsible
behavior to the menu function
The code inside getSkinData that checks VectorUseIconWatch is
redundant as it duplicates checks already inside
SkinTemplate::buildContentNavigationUrls
It is enough to simplify check whether watch or unwatch is
present in the array.
Bug: T251212
Change-Id: If6b10b0ddcbd4b21dd13a2813e60b604c3a23415
* Standardise the menu markup. This means all menus in Vector will now
be wrapped in a div and will have a heading.
* All menus now have the vector-menu class. Styles specific to personal tools
are moved to layout since these are concerned with placement.
* The ul class will always have menu class.
* emptyPortal class is generalised into vector-menu-empty for consistency
with other classes and moved from common.less into Menu.less
* Standardise hooks - BaseTemplateAfterPortlet can now be run on any
menu.
Changes to HTML:
* lang and dir attributes are moved from the h3 up to the div element
.vectorTabs, .portal(s) and #p-personal now has hidden span element inside h3
* for non portals ul.menu" is now wrapped in a div.vector-menu-content
This change does impact the following CSS selectors which will need to be updated:
I see no matches for these selectors in code search.
```
#p-variants > ul
#p-namespaces > ul
#p-personal > ul
#p-views > ul
#p-cactions > ul
```
Using global-search.toolforge.org I see one match
for p-variants, 26 for p-namespaces, 30 for p-personal,
36 for p-views and 7 for p-cactions. I see this as acceptable
breakage provided a user notice is sent out which it has been
(T252447)
Bug: T249372
Change-Id: Id59234aa6b822a24848386bdc04d8d7ed37ca145
To complete the refactor, the Portal is also refactored
as a Menu using the getMenu function.
An old code path supporting portals outputted by hooks with
strings is marked as deprecated to simplify this code in future.
array-portals-first -> data-portals-first (the value is not
an array)
Changes:
* $this->getLanguages and $this->getToolbox() always returns an array (see BaseTemplate)
but we previously supported portals made using raw HTML. Let's move away from that
behaviour and deprecate it.
* Hooks are moved into buildSidebarProps and marked as deprecated where possible
(SkinTemplateToolboxEnd). SidebarBeforeOutput can be used instead.
Bug: T249372
Change-Id: I2549af3e24e5d51c09e9a88ca50a0d9b2e154c3f
The classes were recently changed so provided this is merged before
next branch cut no need to worry about cached HTML.
Bug: T249073
Change-Id: Ib20c7a359bda858df89ebb245e682d321dd5acd0
the PersonalMenu should be generalised. In future we will use it as
the template for all menus
Bug: T249372
Change-Id: Id1c43d2e9eefef1d7aec45f0137e27f10ad935df
VectorTemplate relies on many of SkinVector's methods. To future
proof these tests we need to set the skin in the mocked object.
Bug: T251212
Change-Id: Ifd9bbc9c909626ecfe8ccd085673bc777423d560
de76ab5 added the config,
`$wgVectorDefaultSkinVersionForExistingAccounts`. Its usage in
`Hooks::onUserGetDefaultOptions()` was invoked not only for existing
accounts but anonymous users _as well._ This is a bug, due to my own
misconceptions about the hook, that went against both the config's name
and its documentation.
Unfortunately, user sessions are unavailable in
`Hooks::onUserGetDefaultOptions()` so it does not seem to be possible to
determine whether the active user is an anonymous or existing account.
This patch drops the hook and centralizes all version determination
logic in SkinVersionLookup::getVersion(). SkinVersionLookup requires a
the active User object and can make the anonymous / existing account
determination by checking login state.
The issued was identified while responding to review feedback given by
@polishdeveloper / @pmiazga in
I52d80942b4270c008d4e45050589ed9220255a50.
Bug: T251415
Change-Id: I7982b4c34283ba81d0232ee6f501c44cf0a74b98
FeatureManager::registerRequirement registering an instance of
SimpleRequirement with ::registerComplexRequirement was awkward.
Changes:
* Rename FeatureManager::registerRequirement to
::registerSimpleRequirement, which is exactly what it does!
* Rename FeatureManager::registerComplexRequirement to
::registerRequirement
Bug: T244481
Change-Id: I612af959cee4cdcd0bdcda51a81b86ed61ee2e16
Changes:
- Add the LatestSkinVersionRequirement requirement class, which lazily
evaluates the application state to get the version of the skin for the
request.
This majority of this class is taken from Stephen Niedzielski's
Vector\SkinVersionLookup class, which was introduced in d1072d0fdf.
- Register an instance of LatestSkinVersionRequirement and register the
'LatestSkin' feature that requires that requirement
- Re-introduce SkinVector::isLegacy and make it defer to the Feature
Manager
Bug: T244481
Change-Id: If6b82a514aa5afce73e571abdd8de60b16a62fa8
As described in the readme but not implemented until now, this patch
enables the skin version to be specified as a URL query parameter. This
is useful for testing both skin versions during development and on wiki,
as well as enabling sharing URLs with a specific skin (Vector) and skin
version (1 or 2).
Obtaining the actual skin version requires tying together three input
sources, WebRequest, User, and Config. It seems simple but it'd be easy
to botch. For this reason, a helper class to correctly interrogate them
and tests are provided.
Bug: T244481
Change-Id: I52d80942b4270c008d4e45050589ed9220255a50
This commit is singularly focused on adding a link to the sidebar for
Vector, logged-in users. It does the bare minimum to fulfill the
requirements of T243281.
Additionally, it will help to answer the question "Do we need to use
abstractions (other than maybe different templates) to separate Legacy
Vector from Vector" by intentionally leaving out any abstractions in
order to make it easier to compare with a follow-up patch
(Ib2ef15180df73360cc1de25b893e49d415d23e1a) which does use abstractions.
It is a good thing to question whether or not we need addtional
abstractions in VectorTemplate and if they will help us as unnecessary
abstractions can have the opposite effect and just lead to further
frustrations down the road.
Therefore, I urge you, the reviewer, to let me know your thoughts! If
abstractions are viewed as not making our lives any easier, the
follow-up patches may be completely discarded and that's totally okay
with me. :) I think it's a good think to talk about now though.
Important changes:
* The VectorTemplate constructor was changed to allow injecting the
config, templateParser, and isLegacy boolean (only the config was
allowed before this commit). According to MediaWiki's Stable Interface
Policy, "Constructor signatures are generally considered unstable unless
explicitly declared stable for calling" [3]. Given that VecorTemplate's
constructor is not marked as stable, it is justified to do this without
warning according to the policy.
* Due to the above, the 'setTemplate' method is no longer needed and was
marked as deprecated.
* VectorTemplateTest was made to adapt to the new VectorTemplate
constructor. Additionally, it now extends from
MediaWikiIntegrationTestCase which my intelliphense server can pick up.
I *think* MediaWikiTestCase is just an alias to
MediaWikiIntegrationTestCase [1] and MediaWikiTestCase file was renamed
to MediaWikiIntegrationTestCase in [2], but I'm willing to change it
back if there is pushback to this.
Open questions:
* What are VectorTemplate's responsibilities? To me, it acts right now
as a controller (because it echos the full HTML string from the
template), a model (because SkinTemplate::prepareQuickTemplate sets data
on it which it later retrieves through `$this->get()`), a presenter
(because it adds data tailored for a web-centric view), and a view
(because it renders HTML strings instead of letting the view/template be
solely responsible for that). Arguably, some business logic might be
mixed in there as well (because it checks to see if a User is logged
in/has necessary permissions to show x which my changes here add to).
This might not be a problem if we keep VectorTemplate relatively small,
but will it remain this way as we progress further in Desktop
Improvements?
* How do we write tests for VectorTemplate without exposing unnecessary
public methods? For example, if I want to test the `getSkinData()`
method to see what state will be sent to the template, how should I do
this? One option might be to use `TestingAccessWrapper` to expose these
private methods which is what
`VectorTemplateTest::testbuildViewsProps()` does. Another option is to
accept this method as public. Is there a better way? Keep in mind that
even with access to this method, there might be many things to mock.
[1] 0030cb525b/tests/common/TestsAutoLoader.php (L64)
[2] Ie717b0ecf4fcfd089d46248f14853c80b7ef4a76
[3] https://www.mediawiki.org/wiki/Stable_interface_policy
Bug: T243281
Change-Id: I0571b041bcd7f19bec9f103fa7bccdd093f6394d
Address some feedback from I7a2cdc2dfdf20d78e4548f07cf53994563b234b3:
- Miscellaneous documentation improvements.
- Add a false case test to `DynamicConfigRequirement->isMet()`.
Bug: T244481
Change-Id: Ic5637f42da755f871c5a6d545e14effd3ac8c670
We expect the vast majority of requirements and features to be defined
in services as possible. However, there are some "complex" requirements
that require additional application/HTTP request state. Unfortunately,
service wiring is done before some of that state is available.
I65702426 attempted to work around this by requiring clients of the
Feature Manager to pass that additional state on every interaction with
the system. Those complex requirements would then select the parts of
the state that they required when it was required. However
implementations of \IContextSource are God objects and their use should
be limited.
Whilst reviewing I65702426, Stephen Niedzielski mentioned that the
application state being available is a requirement. This remarkably
simple solution:
- Keeps the Requirement interface and FeatureManager API free of God
objects;
- Is true to the nature of the Feature Manager - it makes clear and
centralizes the various checks for application state being available
across the codebase; and
- Inject a Requirement implementations' dependencies at construction
time
It just so happens that the $wgFullyInitialised variable flags whether
the application state is available...
Changes:
- Add the the FeatureManager\Requirements\DynamicConfigRequirement class
and tests. The DynamicConfigRequirement lazily evaluates a single
configuration value whenever ::isMet is invoked
- Register an DynamicConfigRequirement instance, configured to evaluate
$wgFullyInitialised while constructing the Vector.FeatureManager
service
Bug: T244481
Change-Id: I7a2cdc2dfdf20d78e4548f07cf53994563b234b3
FeatureManager::registerRequirement established the interface for a
requirement: its name and whether it's met.
However, the Feature Manager also needs to handle scenarios where a
requirement needs additional context before it can be considered met.
That context may not be available when the application is booting, e.g.
checking if the user is logged in; or the logic is complicated enough
that it should be under test.
Changes:
- Add the Requirement interface and update FeatureManager to work with
implementations of it
- Maintain B/C by constructing an instance of a the SimpleRequirement
DTO
Bug: T244481
Change-Id: Id95d9e5d7125492968d0e15515224aadbc3075f8
I735fd640 bumped the required MediaWiki version to 1.31. That version
dropped support for PHP 5.x.
Wherever possible, update FeatureManager's methods to use PHP 7.0.x's
scalar and return type declarations.
Bug: T244481
Change-Id: Ib5636d0ec5ec7f0c93b5b3317a12635668b589e2
As was noted in https://phabricator.wikimedia.org/T244481#5859513, the
term "set" doesn't seem natural. Piotr Miazga (polishdeveloper, pmiazga)
and Nicholas Ray (nray) suggested a number of good replacements,
including "requirement." Serendipitously, this term is already used in
FeatureManager's documentation.
Bug: T244481
Change-Id: I559c2d4149db69235cdd4bb880697deb1a145743
Sections can be nested. Using rendering/skin as the parent
means that Vector's skin preference will always come straight
after the skin preference and before Popups.
A change in core is needed to update the selector for the element which
shows/hides the subsection as well as provide a generic message key to
replace the one inside this repository.
Note: If the "Vector" specific heading is needed, we can achieve this
with a little more work but that is a conversation for another time.
Depends-On: Idd06bcfe7935e16732a6a95c1253dbf95c8aca2e
Bug: T246162
Change-Id: I4be9764ddca186e5bfd493678afd62d446072e8f
Moving all the templateParser calls to one function
so its easier to see how the template is composed.
The diff of changes to the stories folder highlight
the internal changes which are:
* html-portals replaced with html-sidebar in main template
* new Sidebar template added which outputs to html-sidebar
* Mention of "MainMenu" replaced with better understood "Sidebar"
This is precursory work to adopt templatePartials
Change-Id: I6b2196e39087f818e774d04b2d1b9ab8cb8816a1
Add a Vector-specific user preference to Special:Preferences for
toggling skin version, either Legacy Vector or the latest Vector.
The presentation of the new preference section and the default values
for anonymous, new, and existing accounts are configurable via
$wgVectorShowSkinPreferences, $wgVectorDefaultSkinVersion (to be used by
the feature manager in T244481),
$wgVectorDefaultSkinVersionForExistingAccounts, and
$wgVectorDefaultSkinVersionForNewAccounts. These configurations default
to the fullest experience so that third-party configuration is minimal.
See skin.json for details. The configurations are each tested in
VectorHooksTest.php.
When presentation is enabled, the new preference appears as a checkbox;
enabled is Legacy mode and disable is latest. There are a number of
unfortunate details:
- Showing and hiding a checkbox is supported by OOUI. Showing and hiding
a whole section (Vector skin preferences, in this case) is not so this
additional client JavaScript functionality is added in Core (see
Iaf68b238a8ac7a4fb22b9ef5d6c5a3394ee2e377).
- Stylization as a checkbox is wanted. However, the implied storage type
for OOUI checkboxes is a boolean. This is not wanted in the event that
another skin version is added (e.g., '3' or 'alpha'). As a workaround,
the preference is converted from a boolean to a version string ('1' or
'2') on save in Hooks::onPreferencesFormPreSave() and from a version
string to a checkbox enable / disable string ('1' or '0') in
onGetPreferences(). There a number of test cases to help cover these
concerning details.
Documentation for overriding the skin version as a URL query parameter
is provided in anticipation of T244481.
Bug: T242381
Bug: T245793
Depends-On: Iaf68b238a8ac7a4fb22b9ef5d6c5a3394ee2e377
Depends-On: Ifc2863fca9cd9efd11ac30c780420e8d89e8cb22
Change-Id: I177dad88fc982170641059b6a4f53fbb38eefad6
With complex additions to Vector's codebase like the Desktop Improvement
Program upcoming, it's important that we have a shared, intuitive
language to talk about features and their requirements. Centralising
the registration of features and creating an API satisfies does exactly
this.
This change introduces a greatly-reduced version of Piotr Miazga's
(polishdeveloper, pmiazga) original proposed API and associated
scaffolding classes for feature management in Vector, which itself was
based upon his work in MobileFrontend/MinervaNeue. This is done to
establish a foundation upon which we can build the more sophisticated
parts of Piotr's proposal in a piecemeal basis, thereby minimising risk.
Distinct from Piotr's proposed API is the ability to register sets and
features that are always enabled or disabled.
Additionally:
- A Vector.FeatureManager service is registered but not used
- A list of proposed immediate next steps is included
Bug: T244481
Change-Id: Ie53c41d479eaf15559d5bb00f269774760360bde
This reverts commit 68292ad0a8.
This test seems to be running on other skins that it should
not run on - for example Minerva. I'm not sure if this
is configuration related or test related.
I will follow up and restore these personally.
Bug: T188553
Change-Id: I4aa58fd021aceab1a01c288782111faeb52723f2
