On 25.05.2020 18:17, Eli Zaretskii wrote:

>>> Yes, this will fix the reported problem of customizability.
>>> Maybe Eli will agree to fix this in emacs-27.
>>
>> I can post the corresponding patch, if it helps.
> 
> It's okay to add a defcustom and a new function, but the other part of
> the patch changes the default behavior, and that is less okay.  Can we
> do one without the other?

The defcustom already exists. It currently defaults to #'ignore (do 
nothing). It is called in case #1.

The attached patch moves case #2 into a new function and makes it the 
default value of the said defcustom (thus case #2 effectively moves into 
case #1). As a result, the default behavior doesn't change, but the user 
will have a much easier time turning off case #2.

> (Btw, the textual descriptions of the options both in the patch and
> those already in the code are confusingly obscure, so much so that I
> don't think I could understand what each one does.)

Knowing the subject matter somewhat, I think the descriptions are 
meaningful enough, but to make sense of them one has to understand how 
the whole feature comes together. E.g. at what times 
next-error-find-buffer is called.

> All in all, I feel (for quite some time) that this area is
> over-engineered and keeps bumping into more and more unintended
> consequences.  Maybe it's time to take a step back and rethink the
> entire subject?  (But definitely not on the release branch.)

That's what we're doing here.

If the attached patch is accepted for emacs-27, then on master we'll 
change next-error-find-buffer-function's default back to #'ignore (thus 
reverting to the previous path that I sent). And see if we find problems 
with how the result is working.

So in the end, next-error-find-buffer-function will have three possible 
values:

- Pre Emacs 27 behavior.
- Emacs 27 behavior (extracted in the attached patch).
- Simplified behavior with less buffer-local state (Emacs 28, hopefully).