* Bucket and sample on server by using the
`WikimediaEvents.WebABTestArticleIdFactory` service from
WikimediaEvents (soft dependency)
* Add linkHijack.js so that users bucketed in one group have the
possibility of remaining in that group if they click a link to another
page.
Bug: T302046
Depends-On: Ie6627de98effb3d37a3bedda5023d08af319837f
Change-Id: Iff231a976c473217b0fa4da1aa9a8d1c2a1a19f2
Headings can also appear in templates inside divs and subtitles.
These do not
get rendered in the table of contents and should not be tracked.
This also excludes headings from the legacy table of contents which
may be in the article during the A/B test
See English Wikipedia examples [[Portal:Biography]] and [[Main page]]
Change-Id: I4ca8933a0e7736157f80e5e68077b153e5bfc81d
- Remove isStickyHeaderAllowed() from stickyHeader.js, move to main.js
- Rename variables in stickyHeader.js to be consistent
Bug: T301429
Change-Id: Ib445a19cbfab52a008b749ea63cef178d6288e6a
- Leverage scrollObserver to use for TOC scroll events.
- Add new hooks to target legacy TOC intersection.
- See corresponding changes (not strict dependency) in related
WME patch 773628 for capturing scroll events from Vector TOC:
https://gerrit.wikimedia.org/r/c/mediawiki/extensions/WikimediaEvents/+/773628
- Dependency on https://github.com/wikimedia/typescript-types/pull/30
which updates mwHookInstance interface with fire property to prevent
TS error on fire method of mw.hook in scrollObserver.
Bug: T303297
Change-Id: I5c2dd5f3a25ffcb0ed03b76ae28e65eb18ad8d33
- Remove some @ts-ignore statements.
- Update invocation of mw.util.debounce to have correct order of
parameters.
Bug: T303297
Change-Id: I4bb795fbd4d026c21c66e9a3d161afff4d7ef09f
* Eliminates AB.js dependency on sticky header
* Code coverage has been raised to 100%
* Instead of importing ABTestConfig, these props are now passed into the
function along with a token.
* WikimediaEvents hook is now fired when experiment is initialized. The
experiment should not be initialized if it is not enabled.
* Removes several methods (e.g. initAB, getEnabledExperiment) due to the
preceeding changes.
* Adds `isInSample` and `isInTreatmentBucket` methods so that the client
has less work.
Treatment buckets now follow a naming convention so that the client can
do less work querying if the subject is part of the treatment:
* Treatment buckets should have the case-insensitive `treatment`
substring somewhere in their name (e.g. 'treatment',
'stickyHeaderTreatment', 'sticky-header-treatment' )
Bug: T302046
Change-Id: I4febec42b4c471b2f2ef02be2e334bd6d2c31eec
This mostly reverts commit f5ad6fe78a but
keeps the SectionObserverProps typedef.
Instead, we will use @module introduced in
Ib68de8cd97cc1111a5a33e100e688d6832fc7e6e.
Change-Id: I7aff49a3d922889cc99bc4313a6cb416410a7a0d
Server render the table of contents in a collapsed state when the total
number of headings is equal or greater than the value of
`$wgVectorTableOfContentsCollapseAtCount`. Otherwise, the table of
contents will be server rendered in its "expanded" state.
In addition:
* Revise table of contents tests to call one `assertion` per element so
that it is easier to see the exact element that may fail an assertion.
* Revise table of contents tests to call a mount function that can merge
props to allow for a more flexible set of tests.
* Revise table of contents tests by wrapping a `describe` around tests
that expect the same prop state.
* Adds typedef for table of sections props
Bug: T300973
Depends-On: Ifaee451e1903f2accd0ada2f2ed6dfa3f83037b6
Change-Id: I382200bc603b6abf757a91f14a8a55a6581969bd
From what I could tell, most instances of `toHTMLElement` or @ts-ignore could be fixed by using typeguards or using Element over HTMLElement.
Element is a looser typing than HTMLElement, but given `querySelector` returns `Element|null` and the fact that we were already inconsistently using both Element and HTMLElement in this file, I feel it was a worthwhile tradeoff
Change-Id: I3512a98fa67c13a4383b9497e8588960259b5b68
Collapses sub-sections in the new table of contents by default
(except for non-js and reduced-motion users) and expands the
sections when the top-level section link has been clicked.
Refactors the `activateSection` TableOfContents methods into separate
`activateSection` and `deactivateSection` functions.
Adds `expandSection` and `collapseSection` methods.
Adds triangle icon as a visual expand/collapsed indicator
next to all ToC section headings and are hidden via CSS based on
whether or not the section contains subsections.
Adds test for tableOfContents.
Bug: T299361
Change-Id: I36b3ae7f9f633877683bc17a9444c970d7fa7293
Move jsdoc comment closer to the methods they are describing. This also
enables better typehint support.
I36b3ae7f9f633877683bc17a9444c970d7fa7293 will handle revising tableOfContents.js.
Change-Id: Ifcac7cfd88cd3f1c0405611c880a0d101d2aed3b
Given our use of constants for tracking classes this eslint rule
is more an annoyance than helpful.
Change-Id: I37570e3e851997d058f2d93777990dddb3d04089
Rather than test for fetch, limit the code to ES6 browsers.
Depends-On: I96a03796628a74ace93579d45a582711400c09c1
Change-Id: I4ca10182491118e61e155f99c713d4cb1b4fc7f0
We want the link that the user has clicked inside the TOC to be "active"
(e.g. bolded) regardless of whether the browser's scroll position
corresponds to that section. Therefore, we need to temporarily ignore
section observer until the browser has finished scrolling to the section
(if needed).
However, because the scroll event happens asyncronously after the user
clicks on a link and may not even happen at all (e.g. the user has
scrolled all the way to the bottom and clicks a section that is already
in the viewport), determining when we should resume section observer is
a bit tricky.
Because a scroll event may not even be triggered after clicking the
link, we instead allow the browser to perform a maximum number of
repaints before resuming sectionObserver. Per T297614#7687656, Firefox
wasn't consistently activating the table of contents section that the
user clicked even after waiting 2 frames. After further investigation,
it sometimes waits up to 3 frames before painting the new scroll
position so we have that as the limit.
Bug: T297614
Change-Id: If3632529f58c15348a7200258f4f5999ea0dadc4
This commits sets up the Table of Contents to bold the active section
when the section is scrolled.
Unfortunately, because our content does not have actual sections but
instead has a flat list of headings and paragraphs, we can't use
IntersectionObserver in the conventional way as it is optimized to find
intersections of elements that are *within* the viewport and the
callback will not reliably fire during certain scenarios (e.g. with fast
scrolling or when the headings are not currently within the viewport).
Furthermore, iterating through a list of elements and calling
`getBoundingClientRect()` can be expensive and can also cause
significant forced synchronous layouts that block the main thread.
The best compromise in terms of performance and function that I've found
is to use a combination of a throttled scroll event listener and
IntersectionObserver's ability to asyncronously find the
boundingClientRect of all elements off the main thread when `.observe`
is called which is the approach this patch takes. Although this is an
unorthodox way to use IntersectionObserver, performance profiles
recorded while holding the "down" arrow and scrolling for 10 seconds
with a 6x CPU throttle are comparable between master and this patch:
master: https://phabricator.wikimedia.org/F34930737
this patch: https://phabricator.wikimedia.org/F34930738
Bug: T297614
Change-Id: I4077d86a1786cc1f4a7d85b20b7cf402960940e7
Waiting for one animation frame seems to make the sticky header
re-appear consistently.
Bug: T299114
Change-Id: Ie1230bf861f12e4e18a6adb0f6779c199d6954a1
I haven't found any code responsible for making the scroll position
jump. It looks like Safari is doing this on its own. Looking at the
focus event in detail [1], it looks like there is an `preventScroll`
option you can pass to .focus() which might help in this situation, but
unfortunately, Safari doesn't seem to support this. Therfore, a hack
like this may be necessary.
[1] https://developer.mozilla.org/en-US/docs/Web/API/HTMLElement/focus
Bug: T297636
Change-Id: I90651293b7dd0f7f2970ba06255a12617b43661f
Move A/B test code to AB.js
Consolidate the show/hide code spread across scrollObserver
and stickyHeader by adding a show and hide function.
This is needed to fix T296680
Change-Id: Ia2e0c50278df0dfc1600610f281be20f4cc755c2
- Permits logging for scroll events without sticky header.
- Update function name to be more precise.
Bug: T292586
Change-Id: I441b4bf81bc4a36a03f0f1c215d86b01dce2911d
- Pull IntersectionObserver into new file to share observer with different callbacks:
- Wrap show/hide functionality of sticky header in conditionals based on user test group or by default.
- Fire hooks for scroll event tracking in WME.
- Add new js for A/B test functions and variables:
- Fire hook to send data for A/B test initialization.
- Update main js to include scrollObserver, A/B test init functionality.
- Add A/B test config.
- Update ResourceLoader package dependencies for sticky header.
- Though not a strict dependency, see I42e3e7c2084c1e88363d5d1662630ed23a28c4d2 in WME repo which uses these hooks to log scroll events.
- This patch includes changes from I56f40e706f8706fde1c0891a0561dd32c5e02bfc which were consolidated here for simplicity and ease of review - related to T292587 which calls for logging an init event for bucketing of users during A/B testing.
Bug: T292586
Change-Id: If6446e1e84cea3649905808c4f0e9f6862255fa3
stickyHeader.js, a file in the "skins.vector.es6" module, clones the
user menu. Because of this, it must initialize before dropdownMenu.js, a
file in the "skins.vector.js" module, in order for dropdownMenu.js to
bind the correct checkboxHack event listeners to the user menu in the
sticky header.
Therefore, change the es6 module to export its main method. The
skins.vector.js module can then use mw.loader.using to ensure the
skins.vector.es6 module initialization happens first in browsers that
support es6. Browsers that don't support es6 will continue to initialize
the skins.vector.js module.
Bug: T291096
Change-Id: I1bb6f2da9703ed2679eacfdb42b9818efe614ab9
Can be disabled via &vectorstickyheaderedit=0 or configuration
change.
This will allow us to fine tune the edit features without blocking
deploying the existing feature.
Bug: T294383
Change-Id: Ic282ea4f2ff0108eeaa154c8a77e4e5fd30daeae
Current expected behaviour: the editor experience will
load and the user will be thrown to the top of the page.
Bug: T293158
Change-Id: I3585616c2244a6b91ef5f160beb1cf51af3599aa
- Can now use const/let
- No need for feature detection for things like fetch and closest
as we can assume they exist if ES6 support is available
Change-Id: I85b01add13fd74e1514119498815403e42a09af0
This will allow us to write ES6 code for the new features which
is limited to those browsers.
For browsers that do not support ES6, the code will not execute
because of the "es6" flag. Doing this will help us avoid issues
like T293402
Change-Id: Iffb7098cb22395e33b87352fb4f08516f6f25e6f
