Eli Zaretskii <eliz@gnu.org> writes:

>> From: phillip.lord@russet.org.uk (Phillip Lord)
>> Cc: Eli Zaretskii <eliz@gnu.org>,  23871@debbugs.gnu.org
>> Date: Fri, 01 Jul 2016 23:21:01 +0100
>> 
>> (let ((p (start-process "bc" (current-buffer) "bc")))
>>   (process-send-string p "2^10\n")
>>   (goto-char (point-max)))
>> 
>> The strange thing here is that it's not specifically an undo problem.
>> Point doesn't move after eval-defun.
>
> The problem, AFAIU, is in point movements _during_ eval-defun: they
> seem to not be recorded in buffer-undo-list.  The undo list I get
> after C-M-x is this:
>
>   (nil (117 . 122) (t 22391 27551 0 0))
>

So, I don't think this is a regression caused by my patch at all. It is
an bug that has been there since I altered undo.c last year.

The problem was caused because of undo only records point after a
boundary (or the first element). I'd changed things during slightly when
I update undo.c so that the timestamp list got added before checking
whether I was at a boundary, hence blocking addition of the point
restoration information.

This is why I found the problem so erratic to replicate; it only occurs
immediately the first change to a buffer.

I believe the following patch addresses the issue.

Phil