This appears to be a caching issue, yes

Change 76552 had a related patch set uploaded by Catrope:
Explicitly disable VE for anons on dewiki

https://gerrit.wikimedia.org/r/76552

Change 76552 merged by jenkins-bot:
Explicitly disable VE for anons on dewiki

https://gerrit.wikimedia.org/r/76552

(In reply to comment #0)

My suspicion is that logged-in users are hitting certain pages, the pages are
being cached, and then anons are receiving these cached pages (with the VE
init
JS loaded).

We don't mix user preference cache between logged-in users and logged-out users, this is definitely not what's happening.

What's happening is that a while ago the 'visualeditor-enable' preference was enabled by default on de.wikipedia.org. Though it was disabled for anonymous users by other means, that preference was there and as such is cached inside the anonymous user cache that visited pages while VE was enabled by default on de.wikipedia.org.

The 'other' means to disable for anons have been removed now that the preference is disabled again on de.wikipedia (it is now an opt-in for logged-in users). However this other means is not cached inside the page but globally from the startup module.

So people visiting the pages generated while VE was enabled by default are now getting VE since there is no 'disableForAnons' flag in place.

Roan's putting this in place now.

(In reply to comment #4)

Roan's putting this in place now.

Done. This bug should stop happening in the next ~10 minutes.

For the record, if my patch set from Gerrit changeset 76199 and Gerrit changeset 76468 had been merged and deployed, this bug never would have happened.

I'm fairly annoyed that I was told that VisualEditor "doesn't work the way you think it does" this morning and my patch sets were roundly and soundly ignored (in favor of Gerrit changeset 76516), only to have the exact changes implemented a few hours later after this bug appeared.

(In reply to comment #6)

I'm fairly annoyed that I was told that VisualEditor "doesn't work the way
you
think it does" this morning and my patch sets were roundly and soundly
ignored
(in favor of Gerrit change #76516), only to have the exact changes
implemented a few hours later after this bug appeared.

Your patchsets weren't ignored, we worked off them. At the time we honestly believed that this was unnecessary, and the code that uses them is written in a way that leads the reader to believe that it is unnecessary: if wmgVisualEditorDefault is false, wmgVisualEditorDisableForAnons is unused.

However, caching threw a wrench in all of this. Two wrongs made a right, if you will.

I'm sorry if we were dismissive earlier. When I said "VE doesn't work the way you think it does", what I meant to say was "despite the confusing way these variables are named, they mean something slightly different, such that one of them is unused if the other one is set a certain way".

(In reply to comment #7)

I'm sorry if we were dismissive earlier. When I said "VE doesn't work the way
you think it does", what I meant to say was "despite the confusing way these
variables are named, they mean something slightly different, such that one of
them is unused if the other one is set a certain way".

I wasn't annoyed with you (and you weren't the person quoted in comment 6). You're wonderful and I adore you.