>> Patch attached.
>
> Hmm... scary, even if this is intended for the master branch.  Could you 
> at least keep the few comments which were in the original code, and 
> perhaps also add comments explaining all those tricky tests with consp, 
> equality to t and to old-bul, the significance of (car ptr) and (caar 
> ptr), etc.?  This would make the code more maintainer-friendly, as the 
> need to constantly consult the spec of the various undo-list elements 
> (and hope the spec is complete and accurate ;-) would be avoided.
>

Is the attached patch better in that respect?

>
> I'd prefer to fix this in Emacs 29 if feasible, but unless the changes 
> are much simpler and/or more clearly documented, so that it becomes 
> possible to assess their risks, I don't see how that could happen.
>

I don't see how the changes could be made simpler, sorry, but now they are 
documented.  Feel free to ask further questions if something is still 
unclear.

>
> I presume that all the undo tests we have still pass?
>

They do.

>
> If so, how about adding a test for this bug?
>

I'll see if I can do that, but it seems tricky.