Refactor the page issues A/B test logging implementation to a distinct
new file that only has the responsibility of tracking.
T191528 is referenced in this commit as I was having difficulty
answering the feedback and bugs reported in the current implementation
without working through and restructuring the flow as I understood it.
This refactor is merely a byproduct artifact of that effort to focus on
the parsing and presentation responsibilities.
Bug: T191528
Change-Id: If547a0a67fbc9a532f834fe374abf668309e73df
Uses `mw.trackSubscribe` to create an intermediary data handler
named `wikimedia.PageIssuesAB` which extends event-logging data
before passing it to the eventLogging through `wikimedia.event.PageIssues`.
Event hooks are placed where appropriate and the `CleanupOverlay`
class is extended to capture events from within the page issues
modal.
Additional changes:
* Merge two identical on click event handlers for
.edit-page, .edit-link elements
* change pageIssueParser.maxSeverity to accept an array of severity levels
instead of an array of pageIssues objects
Depends-On: Ic84e4a3286220407863167e0f57cef1b13a72964
Bug: T191532
Change-Id: I67fb6e448f6ecc97c89c1187e491ee05f7a312ef
These are currently only being run in the daily build
but not on commits meaning this sneaked through our automated
QA into production causing
https://phabricator.wikimedia.org/T200867
Let's protection against this happening again.
Bug: T200867
Change-Id: I2ad6fab8fafa2125be45c5052add9605a9d8121b
The hook that enables the Reading depth test should send an
additional paramter that specifies which test bucket the hook
being is calling from.
Bug: T191532
Change-Id: Ifd9f43220c476ece8a0c0cee46b62b58a717c616
- Fix a bug where the all issues endpoint would incorrectly collect
issues from all sections.
- Update the page issue iconography. This increases the size of the
delivered code and images by 1743 B minified uncompressed according to
mw.inspect() (from 16.4 KiB to 18.1 KiB).
- Add support for identifying page issue severity based on template CSS
classes.
- For multiple issues templates, show the highest priority icon.
Bug: T191528
Change-Id: Ie0a4c83ec7cfb856ec581d058797109746e3cb99
Provides a class that initiates AB-test bucketing and registers
as a MF module. Activates the reading depth test for users who are bucketed
in either buckets "A" or "B".
Does not add event-logging or visual style changes for page issues AB test.
Bug: T193584
Change-Id: If8504a35059c6d1b056cef063a595b1c2ffd351a
Changes:
* Default skin is not configured so noticed on test runs it's using
Default skin for desktop mode. Given this has no footer and way to
switch to mobile this could be problematic in future.
* Remove empty README
* Drop suggested language integration test - setting up interwiki links currently
happens via InterwikiLoadPrefix hook however this seems to be unreliable. The beta
cluster is a much more reliable place to test language links so let's rely on that
instead.
* Drop unnecessary heading check from "Successful edit" scenario. It's unnecessary
and flakey (presumably given the title doesn't change)
* Rename test scenario and remove a duplicate scenario
Depends-On: I888b3c546f77fa350853a7bf9bfbfbeb8ed6de67
Change-Id: I45792a95df7fd4c3299accbffadfa447baefe0ce
String concatenation should use . operator, not +, otherwise it will
report some Warning
Bug: T195645
Change-Id: Id27c981622e5ed87519324193abd2249aa1df7b6
Replace all test assertions for calledOnce / Twice with callCount.
assert.ok( calledOnce / Twice ) only lets the dev know that a test
fails. assert.strictEqual( callCount, EXPECTATION ) starts the debugging
process when it fails since it provides the difference in the failure
output. strictEqual() was deliberately used since it's a saner default
and the codebase already favors === equivalency checks.
find tests -name \*.js|
xargs -rd\\n sed -ri '
s%ok\(([^,]+)calledOnce%strictEqual(\1callCount, 1%g;
s%ok\(([^,]+)calledTwice%strictEqual(\1callCount, 2%g;
'
Change-Id: I5c4c595c4520cecfad46d652f639a63a1c2c00b4
Instead of requiring a full IContextSource object and only using the
Title, only ask for the Title in the first place.
Change-Id: I33034193140ca53919f29f847a03caf26250ce54
We only need to opt into beta to test the beta indicator displays.
There is no need to do this anywhere else in our browser tests.
For the categories test we will enable the feature flag in test mode
and not run it against the beta cluster.
Bug: T174018
Change-Id: I83b5f24236cef6ddd6fc1882bdfff3618a8bf599
* Drop special.feature - it checks a search box is present on
multiple pages. This dates back to when special pages and normal
pages did not use the same code path.
Bug: T174018
Change-Id: Icf59ff95135af91650f8cdeef79bbaed534165f0
More hackery!
This adds some tests and ensures that our own icons are mapped to the
existing template icons.
Bug: T187916
Change-Id: I49073f22995c6730369235d6039939915ba2079c
Additional change
* Add it to the list of smoke tests
Depends-On: I731608412eb2ade95abb79ed8240cfec0f06fb98
Bug: T189194
Change-Id: If8a20202365d92766c52091d96633a74d72e480e
Important note: Make sure to distinguish unseen from unread
One way to reproduce minerva and non-minerva notification inconsistencies:
- Have all your alerts and notices seen. This is displayed with grayed out
number on vector skin or no number at all, if you have (marked as) read.
- Generate new alert or notice (one is enough) in your preferred way.
- You can check minerva and non-minerva at this step. Both should be in sync.
But don't perform any additional action.
- Open the notification popup in some non-minerva skin (I have tried with
vector and monobook), marking it as seen.
- Check the notification icon in minerva. At this point, you should see
notification displayed as unseen.
The reason bug appeared in the first place is that alert/notice timestamps
were mixed up when seen time is obtained. We get seen time from
EchoSeenTime class, where we get smaller of the two timestamps,
using PHP method `min()`. See I27109ee6a248. Then, we get last unread
notification timestamp (which can be either alert or notice), and compare
that to seen time. That leads to the situation when you have only one of
alerts or notices with unread items, smaller timestamp is used for seen,
and most recent for unread, at which point we compare timestamps for
two separate things.
Previous behavior of getting seen timestamps (using max instead of min) would
probably solve the problem, but some other inconsistencies might arrise.
This should prevent any weird and unpredictable behavior to happen.
Bug: T183076
Change-Id: I20bbd6c590086b1c3eccf82983aad59eb3144a7a
Changes:
- moved DownloadButtton checks & initialization to separate function
- introduced supportedNamespaces variable for better readability
- reorganized huge if(){} statement to set of smaller if's with
nice comments why this configuration is not supported
- introduced getAndroidVersion and getChromeVersion helper functions
- added check to not allow Android < 5 or Chrome < 41
- added unit tests
Bug: T182059
Change-Id: Ib5064459ee56aed68179389f37b4bc3b5c2c4492
When the print button is clicked, load all images from the page
before calling window.print
Add a timeout to make sure the user doesnt wait too long.
Change-Id: Ie922d239f9c5b5757237dc10b673fb500ff203ad
Depends-on: Id7f21606be3db22fe8dfde2db675f9905547cfea
Bug: T180058
Fontchanger code now runs on all skins under the `mobile` target.
All the code will now live in MobileFrontend meaning developers
can operate inside one code base.
Depends-On: I857cfe2d9be9fe49c04c860bc234384c787239b2
Change-Id: I2759455cb6d7ddf13798e94452cb74baf502bafe
The talk page JavaScript progressively enhances an
existing button in the page.
Remove the frontend logic and rely entirely on whether
the button is in the page or not.
Additional change:
* The browser tests incorrectly suggest a user needs
5 edits to be able to use the talk feature. This is not
true. They just need to be logged in. Update that logic.
Bug: T167728
Change-Id: Iacedea30bdd0775b3d785db5b143abafd7a18b39
Changes:
* Minerva now maintains a MinervaUI - a simplified version of
MobileUI that provides iconClass and buttonClass helpers.
* Minerva now maintains its own ResourceLoaderParserMessageModule
Remaining issues:
* Main menu links to '#'
* Unknown dependency errors are thrown due to the missing
JS libraries e.g. mobile.watchstar
thus JS based UI components are unusable e.g. search autocomplete,
and edit button
* Language button navigates to a missing special page without
MobileFrontend (see T104660)
Bug: T169569
Change-Id: I89e2e15faabab73b0cba91afc2f2c5e785edef29
On the beta cluster, the tests are run using "Selenium user" account.
There is a test which assumes the user's page is blank which is now
failing as a result of this edit:
https://en.m.wikipedia.beta.wmflabs.org/w/index.php?title=User:Selenium_user&oldid=371028
Test is now disabled there but will still run on commits via the
@integration tag.
Additional change:
* Drop the additional @login tag
Change-Id: Ie116caed32883c58edfa2ab3dfa47bbbefdd6f31
These flake too much. We have coverage in @firefox or @integration
so let's remove these test runs from the beta cluster job.
All tests will run somewhere.
Bug: T170890
Change-Id: I269991a22ab5a2b54aed8ad453013bb9be648502
If mediawiki.notification has loaded that should be enough to assume
the toast is ready to have its text checked.
Change-Id: Ic546877eae0ea6dd59dbf88bf9267bcd1957f779
These are failing consistently due to T172835. Let's not run it in the
@chrome job and just rely on the Firefox coverage. We'll keep one test
in @chrome as it runs in integration mode (per commit).
Change-Id: Ic417148574208389b66249b2d98e009beef70fa2
It seems trying to test both the steps can cause
false positives. Relaxing these checks seems to make
our Jenkins job happy without breaking the tests themselves
Change-Id: I119111e97f23d2f0dac7cbb0e5b86c1df0562598
Changes:
* Use css rather than class for finding toast
* Correct a test typo
* Add a step to wait until the mediawiki.notification module
has been loaded
Bug: T170890
Change-Id: I86e48e00ebb83772149da7c7f20097b5436a0cf5
Copies approach for the text of the first heading should be
accounting for the fact that the toast can have an empty
message "" at any given time.
Bug: T170890
Change-Id: Iba8a503a2aea30cb46fba27f000843183e9c46f1
A toast autohides within 5 seconds and its display properties are
inherited from #mw-notification-area. This slight tweak waits for
mw-notification-area to be visible before verifying toast and its
contents
Change-Id: I89beaf9d131155e958cc9aae84a9e30ffd8e9e4f
Introduces a new generic
"I should see a toast with message ".*""
step reducing toast steps to two generic ones.
Change-Id: Ic8b91c78f6df088244f15223ee4ed658847a05b5
Changes:
* Update docs
* Update browser test artifacts
* Update comments
* Update phpunit test groups
* Update phpunit test namespace
* Update `die` when MobileFrontend not installed
* Remove the migrate script which is no longer needed
Change-Id: I83432b3f7f0bcd07ed08259972b8ff89147104b6
Sniffs that are currently failing are disabled in phpcs.xml.
Additional changes:
* Fix problem in test file
Change-Id: I53642e9d7bc1ef96e359cfe04a8f93dabbc977eb
Test scenario for getContextSpecificModules() mocks only Skin->getTitle()
behavior, but while executing isAllowedPageAction() Skin will create a
UserPageHelper with default RequestContext. As RequestContext is not mocked,
$context->getTitle() will return undefined what could lead to tests crash
Changes:
- instead of mocking SkinMinerva::getTitle() pass test context with injected
title. Other tests will work properly as MediaWikiTestCase::tearDown() always
restes RequestContext to default
Bug: T170624
Change-Id: I872fddf8d9c52a6875bb6c69a12407a8125fba4c
This is programmatic output from python3 scripts/migrate.py
This will result in a Minerva skin dependent on MobileFrontend.
Post merge we will rename message keys to have minerva- prefix
Bug: T166748
Change-Id: Iff1f7e63e796cc5d4a6d2ab0370e0c33248d2fce
This moves all browser tests from MobileFrontend to the Minerva repo
in preparation for separating the two.
Note, this means browser tests will exist in both repositories for a
period of time. This is important and necessary to ensure we do not
break anything.
See:
https://lists.wikimedia.org/pipermail/mobile-l/2017-July/010536.html
Bug: T168758
Change-Id: I84ae3ea14191f672cabcd52020e80b0a40a72ce1
