There is code in MultimediaViewer that can be made use of here. Needs changes in API first before it can be used in client.

Please report this to the talk page. The template does not appear to be compatible with page previews. Thanks in advance!

I am still working on qqq.json but that should be up shortly. :)

In T380558#10349238, @Umherirrender wrote:

Happen via Html::warningBox?
It seems that 0be2bdd9c7742d2d5825d8e86f84fafbbbacc112 is at fault. It adds a codex class to a central place and does not ensure the styling is loaded. Now all usages have to be changed to load 'mediawiki.codex.messagebox.styles'.

Thanks @mmartorana for the assessment ! Happy to hear this is low risk.

No need for QA.

Ok modified the query and managed to run in on the latest snapshot

As a proxy for search within pages we can look at uri_query in the webrequest table for a given hour using https://superset.wikimedia.org/sqllab.
The queries constructed on mobile and desktop are slightly different. Mobile URLs will always begin "?fulltext=search" and desktop "?fulltext=1"
To avoid timeout issues I've limited results to a single hour. on a single day on en.wikipedia.org but I am currently getting server errors so can't get these queries to complete. @jwang / @KSarabia-WMF any chance these queries work for you?

Yep just beta testing. The fix will go out next Thursday!

I am going to need more information as I can't reproduce on Mac OS with Chrome or Firefox. Is it possible you have some browser extension running additional styles on the page in incognito mode (just want to rule that out!)?

Is this just wikispecies or does it also apply to commons? I presume this occurs because of the ULS button but I havent had time to check.

Legacy code would not be removed as part of this ticket due to caching. I added a sign off step with that in mind. I wouldn't be adverse to 2. My thinking for 3 was this is highly visible and a revert would be a pain, so factoring that in to the point score.

Since this relates to content, this is typically out of scope for web team and a difficult problem to address universally. It doesn't make sense to fix this on the platform level.

According to logging about 70k pages every 15 mins are rely on this behaviour
https://logstash.wikimedia.org/goto/1bd455888d36e19ddb5476db1257d36b

To clarify timeline, yes this would all be for Q3 (we might do stage 1 and 2 this quarter depending on bandwidth and what Olga things)

Perfect.

Tagging as medium for now. Please let me know if you change your mind and it does warrant an earlier backport.

Note: in the revert I updated the code to make it clearer that GrowthExperiments is using this, so the same mistake is not made again. Apologies, this was very avoidable, Ed's patch looked harmless but there was no urgency to merge it on my part. At least this relationship is a lot clearer now.

Upped to 5, with new feature flag requirement.

(3 assumes that in the event of any unexpected issues we stop working on this and work on a new task to unblock it)

(3 assumes that in the event of any unexpected issues we stop working on this and work on a new task to unblock it)

@bwang myself and @JScherer-WMF synced and I think it's clear what the next steps are.

@Michael is this impacting users? If so I can try to get it backported today.

That sounds great.
I think the big open question here is do we currently have access to the raw database table in the replicas in MariaDB (and if not how do we enable that or get a dump of that data to look at).

Hey there, we need some time to verify the impact of this change, so for now we've disabled dark mode for anonymous users.

Ack likely MobileFrontend's given Minerva onClickImage in stack trace:
https://gerrit.wikimedia.org/g/mediawiki/extensions/MobileFrontend/+/d2abd0d2e88d2d0d81f37fadc0d7fffd5619eb4a/src/mobile.startup/PageHTMLParser.js#189

Error also occurring as TypeError: undefined is not an object (evaluating 'welcomeDrawer.show().then')

TDLR: We'll add the experiment in RelatedArticles.