bug#44611: Prefix arg for xref-goto-xref

[Dmitry Gutov]

On 24.12.2020 19:44, Eli Zaretskii wrote:

> There's only one rule: not to make backward-incompatible changes
> without a very good reason.  That is not limited to changes in
> functions and APIs, key bindings are definitely included.  In fact, I
> think incompatible changes in key bindings are even worse than changes
> in code, since they might go against muscle memory, and nothing can
> fix that except a long and painful process of unlearning.

I'm pretty sure backward incompatible changes in code can be much worse. 
They can break the code a user has been relying on for many years, and 
the user might not have the technical ability to fix that code.

Adapting to a new binding is something that is within anybody's 
capability, though. We shouldn't do those frivolously either, but I 
believe it is something we should be less worried about.

> I've reviewed all the NEWS since v24.1, and found only one
> incompatible change in key bindings -- which started by declaring the
> existing binding obsolete.  So I think we change key bindings in
> incompatible ways only very rarely and carefully, and that is a Good
> Thing.

It has never been my impression that we are too careful about the 
bindings. Let's review the history.

NEWS.27:

  the function
  'package-menu-filter-by-keyword' has been renamed from
  'package-menu-filter'.  Its keybinding has also been changed to '/ k'
  (from 'f')

That binding was changes without an obsoletion period, whereas the 
function itself still has an obsolete-alias. Both are good things, IMO.

NEWS.26:

  *** 'vc-region-history' is now bound to 'C-x v h', replacing the older
  'vc-insert-headers' binding.

The binding changed to a function with an entirely different behavior.

NEWS.25:

  ** The old 'C-x w' bindings in 'hi-lock-mode' are officially
  deprecated in favor of the global 'M-s h' bindings introduced
  in Emacs 23.1. They'll disappear soon.

This is probably the example you're referring to. Note that it talks 
about a whole family of bindings, not just one, and that those bindings 
almost certainly have had their users. They have probably been the main 
way to use hi-lock-mode's functionality at some point in time.

NEWS.24:

  ** New Ibuffer `derived-mode' filter, bound to `/ M'.
  The old binding for `/ M' (filter by used-mode) is now bound to `/ m'.

Bindings changed in an incompatible way (the previous binding had been 
there for 12 years).

  ** Dired

  *** The minibuffer default for `=' (`dired-diff') has changed.
  It is now the backup file for the file at point, if one exists.
  In Transient Mark mode the default is the file at the active mark.

  *** `M-=' is no longer bound to `dired-backup-diff' in Dired buffers.
  The global binding for `M-=', `count-words-region' is in effect.

  ** Ruby mode

  <Lots of key binding changes without advance announcement>

I have omitted a few others too, where the changes have seemingly been 
really critical and thus unlike the present issue.

To sum up, we don't change bindings often (in part because it's hard to 
get everyone to agree on a new default, I guess), but we do that from 
time to time. And out of all examples, there was only one when we 
announced the obsoletion first.

> Given the single precedent I found, I'm fine with declaring the
> current binding of TAB obsolete and providing a replacement for it
> ('b'? 'q'?), so that we could replace it in some future version after
> 28, if that is okay with you and Juri.

Thank you, this is at least one path forward we can take, which is good 
to have. But if that is what we have to do to change that binding, it's 
pretty costly time-wise, so let's discuss some alternatives first.