@nray what do you think about using replaceState to rewrite URLs created via link hijack? If those are shared anywhere a non cached version of page would be served

Is the idea to use replaceState when the page loads to replace /wiki/Tree?tableofcontents=1 with /wiki/Tree ? I suppose that's doable, but didn't the SRE team approve the current strategy as is?

@nray I don't think I asked :-) @BBlack we're putting together an A/B test for anons and currently we'd be serving anons with URLs like https://en.m.wikipedia.org/wiki/Himalayas?tableofcontents=1 for a short period of time (duration of A/B test - about 2 weeks). While this is the case, there's a possibility these links get distributed around the web and those links outlive the A/B test. Is that a problem? We can rewrite the URL in JavaScript to minimize this if so.

@Jdlrobson I think expired or invalid query parameters usually can potentially break things with a bad URI but it is up to the server that receives it. Perhaps a rewrite condition can be created somewhere but that is outside of my domain.

Change 779995 had a related patch set uploaded (by Jdrewniak; author: Jdrewniak):

[mediawiki/skins/Vector@master] Sticky header edit button A/B test bucketing

https://gerrit.wikimedia.org/r/779995

Just to note the approach taken here. The patch above depends on enabling the edit button for everybody by default, then hiding it for all users who are not in the treatment bucket.

So for the wiki we'll be running this AB test on, we'll have to add the following config to enable the sticky header edit button for all logged in users:

$wgVectorStickyHeaderEdit = [
    "logged_in" => true,
    "logged_out" => false
];

and the AB test buckets will look like this:

$wgVectorWebABTestEnrollment = [
    'name' => 'vector.sticky_header_edit',
    'enabled' => true,
    'buckets' => [
        'unsampled' => [
            'samplingRate' => 0
        ],
        'stickyHeaderNoEditButtonTreatment' => [
            'samplingRate' => 0
        ],
        'stickyHeaderEditButtonTreatment' => [
            'samplingRate' => 1
        ]
    ]
];

(sampling rate will differ) and then the sticky header edit button will be disabled via javascript for users in the unsampled and stickyHeaderNoEditButtonTreatment buckets.

Change 779995 merged by jenkins-bot:

[mediawiki/skins/Vector@master] Sticky header edit button A/B test bucketing

https://gerrit.wikimedia.org/r/779995

This appears to have caused T309370 so moving to needs more more.

Change 800697 had a related patch set uploaded (by Jdlrobson; author: Jdlrobson):

[mediawiki/skins/Vector@master] Revert "Sticky header edit button A/B test bucketing"

https://gerrit.wikimedia.org/r/800697

Change 800697 merged by jenkins-bot:

[mediawiki/skins/Vector@master] Revert "Sticky header edit button A/B test bucketing"

https://gerrit.wikimedia.org/r/800697

Change 801458 had a related patch set uploaded (by Jdrewniak; author: Jdrewniak):

[mediawiki/skins/Vector@master] Sticky header edit button A/B test bucketing (updated)

https://gerrit.wikimedia.org/r/801458

Change 801823 had a related patch set uploaded (by Jdlrobson; author: Jdlrobson):

[mediawiki/skins/Vector@master] Adds tests for stickyHeader A/B test logic

https://gerrit.wikimedia.org/r/801823

Change 801458 merged by jenkins-bot:

[mediawiki/skins/Vector@master] Sticky header edit button A/B test bucketing (updated)

https://gerrit.wikimedia.org/r/801458

Change 801823 merged by jenkins-bot:

[mediawiki/skins/Vector@master] Adds tests for stickyHeader A/B test logic

https://gerrit.wikimedia.org/r/801823

Test Result - Beta

Status: ✅ PASS
Environment: beta
OS: macOS Monterey
Browser: Chrome
Device: MBP
Emulated Device:NA

Test Artifact(s):

QA Steps

✅ AC1: Per discussion in standup, since the TOC experiment is completed the criteria for passing this is validating test bucketing for the edit icon in the sticky header.

I wasn't able to get a 50% bucketing rate, but with only 10 users tested the sample size is small. I was able to validate that two buckets exist, but the rate is not verifiable with a small sample size.