It looks like the only actions that are disabled by default are block, rangeblock, and degroup. The reason blockautopromote is available on enwiki is that no one's ever explicitly disabled it with $wgAbuseFilterActions['blockautopromote'] = false;, yes? The question is, do we need "community consenus" to get that line added, given that AFAIK no one has used that option in years on enwiki?

Marking this as low priority for now. @suffusion_of_yellow - could you expand a bit on why you think that the red circle is not a good enough indicator? I agree that we are most definitely not a social media site, but regardless I think it's an effective way to notify.

In T240889#5752169, @Xaosflux wrote:

@suffusion_of_yellow regarding "range talk", lets assume for a moment that it did exist - how would you expect notification/clearing of such a notification to function?

In T240889#5751876, @Aklapper wrote:

I'm curious how high the rate of IP users is that are not ignoring "their" talk page. Especially if they don't have a static IP. (Anybody having any numbers?)

@ovasileva: Why was this one marked low priority? I understand that people might disagree with me about T240976, and I'll respond to your question there later, but this one's a big deal. We literally have no way whatsoever to initiate a discussion with logged-out mobile users. Worse, we think we're talking to them.

@Xaosflux: Thanks! I consider the IP issue a high priority problem, the logged-in issue less so. I had assumed that the goal had been to deliver the banner to all users, in one form or another (as on desktop), and a simple bug was preventing the display. Now that I know that part of this is intentional, I will split the task.

@Ammarpad: I only see the red circle when I'm logged in. I see no indication of any kind that there is a message, when logged out. Do you?

In T240889#5745757, @Aklapper wrote:

If this ticket is about the lack of a "You have new messages" banner in the mobile interface for logged out users, then I do not see any bug here. Not sure as the steps are confusing and don't explicitly list an expected and an actual outcome, and if that was in mobile or desktop or in that one window or that other window.

In T240889#5745743, @Aklapper wrote:

@suffusion_of_yellow: Is one of these browser windows in private mode, or not?

Certainly seems resolved on enwiki. I re-enabled the filter in the task, and there were no hits in about a day. Thanks!

It's a good default to try to find and test vulnerabilities like this locally, but again, sometimes that's just incredibly inconvenient or even impossible and so discreet testing somewhere like the testwikis becomes the only viable option to fixing these vulnerabilities.

How do I find out if an XSS impacts production? I don't want to save anything like that on testwiki, even if I delete it one minute later.
Anyhoo, this page works for me even with $wgFragmentMode = [ 'html5', 'legacy' ];:

User-agent Mozilla/5.0 (X11; Linux x86_64; rv:71.0) Gecko/20100101 Firefox/71.0
MediaWiki 1.35.0-alpha (fdfa0e9)
MobileFrontend 2.1.0 (a812ef8)

Haven't tried to set up the 2017 editor locally. That means installing VE first, yes?

Could varnish be doing something with the line endings? I'm in the US, so we are probably connecting to different datacenters.

User agent: Mozilla/5.0 (X11; Linux x86_64; rv:71.0) Gecko/20100101 Firefox/71.0

I just realized that almost all of those edits were saved, and _lines variables are just unavailable.

Reproduced on testwiki, again with the 2017 editor: See https://test.wikipedia.org/wiki/Special:AbuseLog/55835. The some content triggered this on enwiki; see https://en.wikipedia.org/wiki/Special:AbuseLog/25537680.

Ok, that explains 25321331. I also don't see tboverride in user_rights, so it looks like GorillaWarfare didn't grant Huggle editprotected rights. This can probably be closed as invalid, but 'll leave that to Daimona.

@JJMC89: That makes sense, in theory. But is it really impossible to edit EC-protected pages from Huggle? If not, EC should still be part of the rights.

Original discussion here (PermaLink).

@Urbanecm: That was quick! Your commit message says users can "view any version of any filter", but the problem isn't that severe. Only old public versions are visible, which is only a problem when filter managers tick the wrong box. At least, I get an error for:

This change could actually reduce the privacy of registered users. If "privileged user" means CheckUser, this will require massively increasing the number of CheckUsers needed to deal with routine day-to-day vandalism. A compromised (or rogue) CheckUser account is a disaster, so the privacy of registered users depends on keeping the number of such accounts as low as possible.

@Catrope: Now, in multiple browsers, document.body.scrollWidth > 10000 in MonoBook, producing a huge amount of empty space to the right of the content. Seemed to start with this fix.

@matmarex: Also tried in my sandbox, and it seems to work! And based on the latest data (created by piping variable dumps from an existing abuse filter to a throwaway Perl script), the problem seems to have stopped for everyone else.

Based on this extract from one of our section-blanking filters, it looks like this problem started on around 9 May, and is happening about 40 times per day on enwiki.

In T213763#4881554, @Bawolff wrote:

In principle, you should only get the session failure error if your token doesn't match your cookie, or if the request was missing some fields. In the event you have cookies disabled you should get a different error message. (I have not tested this myself - perhaps this code is broken and giving the wrong error message)

A very quick fix would be to restore the status quo, and temporarily remove the abusefilter-private right from all CheckUsers.

In T110329#4677511, @Deskana wrote:

In T110329#4676140, @Xaosflux wrote:

en wiki (https://en.wikipedia.org/w/index.php?title=Wikipedia:Edit_filter_noticeboard&oldid=864548831#VisualEditor) has created an edit filter to stop the rampant misuse of indexing in user space, that upon review we think is from this tool.

Do you have evidence of that, or is it a guess?
It would be relatively straightforward to figure that out, by seeing how many of the edits that add the magic word have the "Visual edit" tag on them.