Re: goto-line-history should not be buffer local.

[Matt Armstrong <matt@rfc20.org>]

Alan Mackenzie <acm@muc.de> writes:

> Hello, Juri.
>
> On Tue, Feb 16, 2021 at 19:13:36 +0200, Juri Linkov wrote:
>> > -(defvar-local goto-line-history nil
>> > +(defcustom goto-line-history-local nil
>> > +  "If this option is nil, `goto-line-history' is shared between all buffers.
>> > +if it is non-nil, each buffer has its own value of this history list.
>> > +
>> > +Note that on changing from non-nil to nil, the former contents of
>> > +`goto-line-history' for each buffer are discarded on use of
>> > +`goto-line' in that buffer."
>> > +  :group 'editing
>> > +  :type 'boolean
>> > +  :safe #'booleanp
>> > +  :version "28.1")
>
>> > Anybody have any objections to this?
>
>> This is not an objection, just sharing a tip that to make
>> Isearch history buffer-local like it's in web browsers,
>> where each web page tab uses own local search history,
>> is possible with a small code snippet:
>
>>   (make-variable-buffer-local 'search-ring)
>>   (make-variable-buffer-local 'regexp-search-ring)
>
>> OTOH, with using defcustom it's easier to customize this, indeed.
>
> The point about the approach I've taken is that once a variable has
> undergone make-variable-buffer-local, it cannot be changed back to a
> non-local variable, except by restarting Emacs (I'm not sure whether or
> not this is a design bug).  So I've avoided the use of that primitive.

Why the option at all?  Why not just make the history global?

I'm not strongly opposed, but I also don't think "make an option for it"
is always the right thing.

Who are the people who would set the option back to buffer local
history?

Note: the feature hasn't escaped to a stable release yet, so no strong
need for an option for backward compatibility reasons.

This is the strongest statement in favor of buffer locality I found:

https://lists.gnu.org/r/bug-gnu-emacs/2019-11/msg01580.html:
> I thought that it might be even better to have the history be
> buffer-local as well; because line numbers from one buffer don't
> really make sense on another.

"it might be better".  We've now got people who actively found it
annoying.

FYI we've got a similar, not yet fixed, bug about goto-char's history:
https://debbugs.gnu.org/cgi/bugreport.cgi?bug=6980.  I'd say both should
work a similar way, but I'm unconvinced that every such case needs an
option covering a local vs. global choice.