Use
$this->getServiceContainer()
instead of
MediaWikiServices::getInstance()
in maintenance and tests.
Change-Id: I845e8f083b19da093166d6551683f2013eec7bb1
- Remove set of $wgDiff, not used in core and seems not needed
- Avoid array_keys on foreach
- Save settings directly by UserOptionsManager for the fresh user
- Combine calls to overrideConfigValue into one
- Referer core config via MainConfigNames
- Remove unneeded reset after overrideConfigValues, get reset there
- Reset specific service and not all
Change-Id: Ia00efae85e17dfac1768b12b35f44eb834f879ec
In EventMapperTest, suppress notifications created by calling
getExistingTestPage. So far, this test only worked because the existing
test page is created in addCoreDBData (and not inside the test), but
this will no longer be the case after core change
I308617427309815062d54c14f3438cab31b08a73. Clear the PageSaveComplete
hook handlers to prevent that.
DiscussionParser has a static cache of revision data that can become
stale when data is deleted between tests (because revision IDs can be
reassigned to different pages, similar to T344124). This cache seems
needed, and converting the class to a service seems hard, so add the
page title to the cache key to try and avoid collisions. This can still
break if two tests are using the same page, which is hopefully quite
unlikely.
Change-Id: Ic4cbd8ff424e1260544ff9754e0c89dd4bb2f733
This code has been moved to Minerva and is no longer serving any
purpose. The handlers are replaced with an already standardized
mw.hook for Minerva to subscribe and respond to.
Bug: T342907
Depends-On: I55c18cf723a32f80b93a01dd0687e005162c4e93
Change-Id: I2f923e509d24524a2375ffbe6b3ef336487574bb
Singletons are bad, amongst other reasons, because they're never reset
in tests. They can therefore occasionally cause test failures if the
cached data stored in one of these singletons becomes stale.
As noted on the task, ideally these two classes shouldn't exist at all,
and core should be responsible for caching the information it deems
expensive to compute.
As a temporary (TM) workaround, make both classes actual services, so
that the setUp/tearDown logic in MediaWikiIntegrationTestCase can
properly reset them between tests.
Dependencies are intentionally not being injected, precisely because
these classes should just be deleted, not improved.
Bug: T344124
Change-Id: I58b8d9610f9447468235b94d25732528ab6acce6
The test was assuming that the ID 1 belongs to a user other than the one
being created in the test. This is not necessarily the case, so make it
use an ID that is guaranteed to be different.
Also rename the test fail to avoid a PHPUnit deprecation of test class
name not matching file name.
Bug: T342428
Change-Id: Ied1869e380b023c1a726a5be610e1de1b68c78c0
The test doesn't need the database, but it's currently trying to use it.
Mock all services used by EchoDbFactory to avoid that (the Factory
cannot be mocked directly because it's created with a static method and
not injected).
Change-Id: I359489ed3f8de6e868a590d5c7c02f81f0a2049d
Use mocks when possible, and add the test to the Database group
otherwise.
Also rename 2 test files to avoid a PHPUnit deprecation of file name not
matching class name.
Change-Id: I16b2392daf01f63511c82b4ed0e67a38e35ddbe1
This will essentially all be deleted after I55c18cf723a32f80b93a01dd0687e005162c4e93, but thats fine\
Bug: T343053
Change-Id: Ifb382af388cdc24dc1ecef105ec89c9129194c19
When calling editPage() with a string it converts it to a WikiPage each
time it is called. When edit a page only once that is okay,
but when adding more revision to the same page it is better to pass a
WikiPage instead to allow reuse of internal states of WikiPage to speed
up the tests
Change-Id: Iac4ef3a3838ba22cd98b944e3ac7f596d540b405
* Remove redundant file closure.
* Remove needless QUnit.newMwEnvironment() calls as the tested
methods don't alter mw.config or mw.messages. Slow, as per
https://gerrit.wikimedia.org/r/c/mediawiki/core/+/912936/
Change-Id: I61bf03325d5067568d057ff2af6bf65f279a8a84
Test ServiceWiring.php using tests copied from CentralAuth. Because
phpunit does not support marking a file as covered, the ServiceWiring
file is ignored for code coverage as the tests should partly cover
the file.
Change-Id: I3ea5ae553d17e17680fcac4d92746fc7ce282eaa
- Don't manipulate $wgHooks directly, use setTemporaryHook.
- Avoid premature access to MediaWikiServices from data providers
- use overrideSettings instead of setMwGlobals
Needed-By: I0ebba9a29f81b0d86ad8fd84d478fb244f9e9c15
Change-Id: I52b9251793a5d559658fcf1e0d20f5a497e5fc43
Notably: any() is the default anyway. It doesn't really make the
tests more specific or better readable when we repeat it all the
time.
Change-Id: I56d201bfce454587b00015b7208f313dd8ed9624
fixes:
ContainmentSetTest::testCachedListInnerListIsOnlyCalledOnce
strtr(): Passing null to parameter #1 ($string) of type string is deprecated
Change-Id: I9bb916383eda1176a2d56c06770a60c28431dfda
Replaced WikiPage::newFromID with newFromTitle,
because a Title object exists and could be reused
Bug: T297688
Change-Id: Ide30f259477ed8e0b48df31f5a23cafeb38d7316
* Declare variables inline, as per the current code conventions.
* Convert ad-hoc cases objects into native QUnit.test.each().
This makes for shorter and cleaner code, as well as for more
detailed test reporting, and removes the need to manually construct
assertion messages based on test case prefix string etc.
* Start adopting ES5 Array.forEach in a few places where otherwise
ESLint would complain about variable name clashes.
* Future proof the test module names, by stripping the global variable
namespace that some classes still use, matching packageFiles convention
as used for NotificationBadge.js and its tests already, by specifying
only the bundle name and the exported class name. Note that the
QUnit UI does fuzzy matching so filtering works the same either
way, e.g. "echodmfilter" will match "ext.echo.dm - FilterModel".
Change-Id: I49858dd2c95d0869f2cd15693f05c38312a9f710
The default timestamp of 00000000000000 cannot represent as timestamp,
because it gets a negative timestamp -00011130000000
This is needed for proper cross-RDBMS support
This reapply a change from I46206e0b3a687dff3168a81cf0020e669133e876,
reverted with I1c8c409b7820512b3e31246a7f3d8c1cf4db209c.
Bug: T244898
Change-Id: I109b783de0a8d60ccb161b280ce5fa09e145017b
== Problem 1 ==
As of I09c27a084100b223, tests/qunit/index.js or equiv was used to
load test files asynchronously from a using() callback. This was
untracked by RL or QUnit, and thus sometimes ended up finishing after
the test runner was already done executing all tests. In CI this
means the tests are sometimes never loaded and the browser (or Node)
process already killed before they even have a chance to arrive.
Prior to QUnit 2.17, this was no way of detecting this. As of
QUnit 2.17 (core upgraded last week) when running tests manually
the following helpful warnings appear in the console:
> [warning] Unexpected test after runEnd.
> [warning] This is unstable and will fail in QUnit 3.0.
> test @ qunit.js
> tests/qunit/model/test_mw.echo.dm.SeenTimeModel.js
> require
There were about 1072 instances of this warning, all from Echo.
Fix this problem by removing the async callbacks and specifying the
two modules as normal dependencies instead.
== Problem 2 ==
Class NotificationBadge was being loaded in a strange way out of
bound. This was a violation of module boundaries and should not be
needed other than for a temporary hack or other tech debt. More
generally when a test uses `packageFiles` this is a likely sign of
tech debt or misunderstandings.
Instead, depend on `ext.echo.mobile` and export/import the class
as normal.
After this, the test module can use `scripts` instead.
== Problem 3 ==
The `ext.echo.mobile` uses a Mustache template which the test
was also duplicating a reference to. This is no longer needed now.
Due to the `qunit/index.js` file carefully splitting the operations
between template assignment and file loading, I wondered whether
it was meaning to replace or mock it with something else, but it
simply refers to the same file and only does this because it wasn't
using the module directly. This is now resolved.
If you do need to mock in the future, this can simply be done
by assigning `NotificationBadge.prototype.template` from a
beforeEach() callback in the test suite, or by supporting it
property as a constructor option in NotificationBadge.js and
assigning `this.template` there, which is supported by the
mobile `View` class already it seems and would follow DI patterns
more effectively.
== Problem 4 ==
Most of the Echo tests were ignored sometimes and executed other
times.
The test for `ext.echo.mobile` in particular though was never
executed in CI specifically because:
> Undefined module: 'mobile.startup'
This became a hard error with this patch, which is fixed by
the CI config change with Ie9dabe3269c56fa76db8e51.
Bug: T299780
Change-Id: Ie4a87f3b8085fd6ae53ec586c1782cc266d5288a
Moved the factory deeper into the code right before the one usage it
still needed a full user object
Narrow done method arguments from User to UserIdentity
and use the identity directly
Change-Id: Ic118f23ef504c7fda892480df61ea68c10915f78
The section titles are wikitext extracted from inside heading markup
like `== … ==`, so start-of-line markup like `*` should not be parsed.
Bug: T299572
Change-Id: Ie3995b943e5fe20ad86041d6be755f14f32eb01e
WebdriverIO has dropped support of sync mode due to breaking changes in Chromium, hence all tests of Echo have been changed to async.
Bug: T293073
Change-Id: I8327d33c99e495b109d97df7b525181dfc41c18d
Replace User::getOption() with UserOptionsLookup::getOption() since this method will be hard-deprecated.
Bug: T296083
Change-Id: I0ecdc63b0344bc4c24196cc5edb3d02b6a7ed615
It's an attribute, so it is definitely the manager's
responsibility. Unit test case included.
(EchoDataOutputFormatter really calls for becoming a service.)
Change-Id: If2658dd8c107246158cd93cbb233d8af62fd4424
Previously, ApiEchoNotifications returned only events which were
enabled for the `web` notifier type. With this change, the notifier
type or types to consider can be specified by argument to the new
`notifiertypes` parameter.
This change is required so that the apps can request notification
content from the API for push notifications.
Bug: T287909
Change-Id: I2d1155e113f2defb0f02416a7a659c3ee162d3a6
At the moment we support a maximum of 10 tokens per user for subscribing
to push notifications, stored as a basic list that runs out when the
limit is reached. There may, however, be some edge cases where an app
registers a token and then forgets to unregister it (and repeats this 10
times), after which time it will be unable to register any new token.
This changes the token list to behave more like a circular buffer, by
simply deleting the oldest token before inserting the new one. This way
an app could register a new token even in the rare case of forgetting
the previous ten.
Change-Id: I387de63460882e4e56d1aa6db1f78d73a0495208
* Fix off-by-one counting
* Fix asserting properties on the wrong event (first one instead of
last one), luckily they are identical
Change-Id: Ib1a02476554b29b2a1d0bfdfa89859d71d7691b1
User::setOption() is deprecated and should be replaced with UserOptionsManager::setOption()
Bug: T277818
Change-Id: I001301fb95635c421a0bbb921fd909c5312dc896
::doUserEditContent() is available since 1.36 as a replacement
for ::doEditContent(), which has been deprecated. Extension
already requires MediaWiki 1.37+, so the method is always
available.
Bug: T255507
Change-Id: Iee5de356dbccd453a3083e0a58859b4cd83a946b
