On Fri, 29 Nov 2024 09:51:57 +0200 Juri Linkov <juri@linkov.net> wrote:

>> I don't remember seeing such backtrace before this change.
>> But now occasionally an error is raised on typing 'M-/':
>>
>> Debugger entered--Lisp error: (error "Selecting deleted buffer")
>>   set-buffer(#<killed buffer>)
>>   (save-current-buffer (set-buffer buf) (if (and (or (eq (current-buffer) dabbrev--last-buffer) ...
>>   dabbrev-expand(nil)
>>   funcall-interactively(dabbrev-expand nil)
>>   command-execute(dabbrev-expand)
>>
>> because can't set the killed buffer 'buf' here:
>>
>>       (with-current-buffer buf
>>         (if (and (or (eq (current-buffer) dabbrev--last-buffer)
>> 		     (null dabbrev--last-buffer)
>>                      (buffer-live-p dabbrev--last-buffer))
>> 	         (numberp dabbrev--last-expansion-location)
>> 	         (and (> dabbrev--last-expansion-location (point))))
>> 	    (setq dabbrev--last-expansion-location
>> 		  (copy-marker dabbrev--last-expansion-location))))
>>
>> It's difficult to debug because the buffer is already killed at this point.
>> But looking at the logic of backtrace, here is a reproducible test case:
>>
>> 0. emacs-29 -Q
>> 1. C-x b foo RET
>> 2. Type: abc SPC abd
>> 3. C-x b *scratch* RET
>> 4. Type: ab M-/
>> 5. C-x k foo RET
>> 6. Type: SPC ab M-/
>
> Sorry, I meant emacs-30, not emacs-29.

Thanks, I can reproduce it.  With the attached patch (against emacs-30)
I don't get the error; instead, at step 6 "ab" expands to "abc", which I
suppose is what's expected, and then typing `M-/' again shows the
message "No further dynamic expansion for ‘ab’ found", which also seems
as expected.  And with the patch, all current dabbrev regression tests
pass with ert-run-tests-batch-and-exit.  Does anyone see a problem with
the patch?

Steve Berman