The same problem exists in Russian Wiktionary. For example: https://ru.wiktionary.org/wiki/%D0%9A%D0%B0%D1%82%D0%B5%D0%B3%D0%BE%D1%80%D0%B8%D1%8F:%D0%A6%D0%B8%D1%82%D0%B0%D1%82%D1%8B/%D0%A1%D0%BE%D0%BB%D0%BE%D0%BC%D0%B0%D1%82%D0%B8%D0%BD%D0%B0_%D0%A2%2E_%D0%AE%2E

May be related to http://phabricator.wikimedia.org/T111199 ?

Something else reminded me of this issue today and I decided to investigate. I'm sorry this went without a reply for so long. I looked at https://ru.wiktionary.org/wiki/Категория:Цитаты/Соломатина_Т._Ю. (linked by DonRumata above), since it's a nice small category, and because no articles in it have custom defaultsort.

It looks like this at the moment:

I don't speak Russian, but compared to the expected order https://en.wikipedia.org/wiki/Russian_alphabet, some headings are in the wrong order (for example, Д is near the end, it should be after Г) and some articles under each heading are in the wrong order too (for example, банан should be the first article under Б).

Looking at the data in the 'categorylinks' table for this category:

select page2.page_title, cl_from, cl_to, HEX(cl_sortkey), cl _sortkey_prefix, cl_timestamp, cl_collation, cl_type
from categorylinks, page, page AS page2
where cl_to=page.page_title and page.page_id=974359 and cl_from=page2.page_id
order by cl_sortkey

(Full results: P2732)

This is the order in which the articles should be normally sorted, they display differently on the category page because they're grouped there by first letter, and the code doesn't expect inconsistent ordering.

A few things are clear:

• There are two large groups, articles with sortkey beginning with '2' and those with sortkey beginning with '5'.
• The articles within each group are correctly ordered (as far as I can tell).
• The latest cl_timestamp date for the '2' group is 2015-01-27 (I think this is the date where the category was added to the article).

So, something happened near the end of January 2015 that changed the format of the sortkeys generated afterwards, resulting in incorrect ordering of the articles. I don't remember anything related happening then, there's nothing that looks related in deployments history nor SAL, and no one from Operations that I asked could remember anything. It also seems that no projects other than Russian-language ones were affected (and there were quite a few projects using UCA collations at the time), or at least no one else complained.

So, to fix it, we just need to run a script to re-generate all the sortkeys – files T129411 about that. The script can take a long time on large wikis, so this might not be trivial, but probably can be done.

I have no idea what's the cause of this issue, but if anyone has any leads, I'd love to know.

I have no idea what's the cause of this issue, but if anyone has any leads, I'd love to know.

Normally the cause would be if someone updated the version of php (or libicu)