Kitty can’t fix it — Old questions inaccessible for logged-in users, giving “unexpected error”

For the past few days, any attempt to access the question Collecting proofs that finite multiplicative subgroups of fields are cyclic results in an internal server error:

It seems the poor kitten needs help from a human.

The front page indicates some activity on this question meanwhile, hence it cannot be broken for everyone. Moreover, I checked that I can access the question when I log out, but after I log back in, it’s broken again. This happens consistently across multiple devices (albeit on the same network).

The associated timeline and revisions pages work fine.

The same problem affects many more questions. As observed by Sam Hopkins, they all seem to be $$\ge10$$ years old. Martin Sleziak found out an important distinguishing feature that all affected question have some deleted answers. Moreover, evidence suggests that the bug only happens for users with $$\ge10$$k reputation (possibly because they can see deleted answers).

It appears from quick sampling that about 2/3 of all old questions with deleted answers are affected by this bug, which means it’s likely several thousands of questions.

As for the cut-off date, the last question affected by the bug that I found is Computing Ext in Exterior algebra (related to Koszul duality) from June 24, 2013, which is the day before the integration of MathOverflow into the Stack Exchange network. Thus, it seems likely that the bug only affects questions from before the migration.

Based on further experiments, it appears a more precise criterion is that the bug affects questions that have a deleted answer that was deleted by June 24, 2013, except when the answer was deleted in a special way (e.g., here the answer was converted to a comment; here it was deleted by spam flags).

A small sample of questions exhibiting the problem as reported in the comments:

• I'm getting the same error (with chrome on mac)
– Tim Campion Mod
Feb 17 at 12:57
• mathoverflow.net/questions/17313/… has been giving me Kitty for some time now. Feb 17 at 21:26
• Another example: mathoverflow.net/questions/26821/is-thompsons-group-f-amenable Feb 18 at 2:37
• Another one : mathoverflow.net/questions/81204/…
– Tim Campion Mod
Feb 18 at 4:21
• And another: mathoverflow.net/questions/14376/… Feb 18 at 13:51
• mathoverflow.net/questions/37211/… Feb 18 at 21:52
• Feb 18 at 22:08
• An obvious pattern: it looks like the problem appears only on old (10+ year old) questions, though certainly not every old question has this problem. For example mathoverflow.net/questions/21003/… is broken but mathoverflow.net/questions/54430/… is ok. Feb 18 at 23:31
• @TimCampion Could you or another moderator add a status-review tag so that the problem is actually noticed by staff? Feb 21 at 8:30
• @EmilJeřábek Thanks for the nudge. Added now.
– Tim Campion Mod
Feb 21 at 14:43
• There are many, many more instances. If you go to one of the questions already in the list and look for questions with nearby post IDs then you will probably find several. For example, the first question currently in the list is number 8972, and 8974 and 8976 are also examples (8973 and 8975 are not questions). However, it doesn't seem to be a contiguous block of IDs that are affected: 8983 seems fine. Probably we're getting to the point where adding newly found instances to the list is not going to help much. Feb 22 at 9:28
• P.S. I’ve checked OEIS so that you don’t need to. Feb 22 at 14:23
• @MartinSleziak You are onto something there. I just checked an old post that I remembered to have a deleted answer by myself, and sure enough, it's broken: mathoverflow.net/questions/67015/… Feb 28 at 15:03
• Hey all, I spoke with the dev looking into this and she gave me a status update - we’ve identified the problem and are looking into how to fix it. We'll have more info soon. Thanks for your patience!
– Catija StaffMod
Mar 1 at 20:04
• @Catija Thanks for the update. Mar 1 at 21:02

Thank you all for reporting and excellent documentation of this issue! We have applied a fix and all the affected pages should now load as expected for logged in users as well as anonymous users.

As you all observed, the issue was indeed linked to older pre-migration post history deletion data which was formatted differently than it is today. Some code was released recently which did not handle the old history format, resulting in the error. This impacted both questions and answers but only in some conditions as you identified, i.e. the page would load if the post was flagged as spam/offensive.

The fix that was applied reformatted the old data so that we no longer need to be concerned with special handling, hopefully preventing headaches like this in the future.

• Many thanks for this! Mar 3 at 17:41
• Thank you for fixing it! Mar 3 at 17:57
• Thank you for this! Who would have thought the migration would still make itself felt this much later?!
– David Roberts Mod
Mar 6 at 9:48

This is not a resolution of the bug, but at least I have found a way around it: If I log out of my account the affected pages all become accessible. If I then log back in I get the "kitty" error, but logging out makes the page accessible again.

• Yes, this is noted in the question. Feb 20 at 11:20
• ah, apologies, I had not read carefully, this answer then does not add anything, I will delete it. Feb 20 at 11:24
• Note: Instead of logging out, you can open an incognito tab on whatever browsers support this. Feb 27 at 15:30
• This is an okay temporary fix, but I really hope someone (at StackExchange support, I guess?) is aware of the bug and working on a real solution, because this is a big loss in functionality and has been going on a while now. Mar 1 at 2:13