bug#3702: Emacs fails to respect the intent in `C-u C-u C-x C-s' of making a backup

bug#3702: Emacs fails to respect the intent in `C-u C-u C-x C-s' of making a backup

[Taylor R Campbell]

Often, when saving a file, I want to make a backup.  The only way I
know to express this intent is `C-u C-u C-x C-s'.  When I type `C-u
C-u C-x C-s', I mean to tell Emacs:  `I want you to make a backup and
then save this file.'  The documentation for the `save-buffer' command
says:

   Prefixed by two C-u's,
    unconditionally makes the previous version into a backup file.
    ^^^^^^^^^^^^^^^
Unfortunately, Emacs fails to respect my intent in many circumstances.
First, I set `version-control' to t in order to keep numbered backups.
Then in order to make backups in directories such as /tmp, I set
`backup-enable-predicate' to be a function that always returns t.
Next, in order to make backups when `vc' is involved, I set
`vc-make-backup-files' to t, which confused me because it didn't take
effect immediately -- I had to reload all my buffers before it did.
This is because some vc hook set another variable, `backup-inhibited',
when I first loaded those buffers.

There are plenty of other documented ways for arbitrary modes to
interfere with my intent when I type `C-u C-u' -- there are several
hooks that are run during `save-buffer' before `backup-buffer' and
conditions under which backups will be made.  This contradicts the
documentation of `save-buffer' which explicitly says that `C-u C-u'
means a backup will be made *unconditionally*.  I don't mind that
there may be various heuristics for choosing whether to make a backup
file when the user has not indicated a preference either way (by
typing `C-u 0' or `C-u C-u' or whatnot), but it is very frustrating
when the documentation says `unconditionally' and the implementation
has many documented ways to interfere and add conditions.  Of course,
the documentation could simply omit the word `unconditionally', but I
don't want that either -- I want it really to be unconditional (as far
as the documented hooks in Emacs go; of course someone could redefine
`save-buffer', but it's the documented hooks and conditions such as
`backup-inhibited' that bother me).

I am using

GNU Emacs 22.2.1 (i386-apple-darwin8.11.1, GTK+ Version 2.10.13) of 2008-07-06 on Oberon.local.

In my .emacs file, relevant to backups, I have

(defun always-backup (filename) filename t)
(setq-default backup-enable-predicate 'always-backup)
(setq-default version-control t)
(setq-default vc-make-backup-files t)
(setq-default backup-inhibited nil)
(setq-default make-backup-files t)

bug#3702: Emacs fails to respect the intent in `C-u C-u C-x C-s' of making a backup

[Andrew Hyatt]

Sorry for the late reply on this one.  This problem still exists in
Emacs 25.  I agree that the current behavior doesn't match up with the
documentation.

This probably is fixable by adding another free variable set by
save-buffer and read in various functions, but there's a lot of those
kind of variables already - the logic here, IMHO, could use a rewrite to
make the interactions clear.  Maybe someone familiar with this code
might give their opinion.

Taylor R Campbell <campbell@mumble.net> writes:

> Often, when saving a file, I want to make a backup.  The only way I
> know to express this intent is `C-u C-u C-x C-s'.  When I type `C-u
> C-u C-x C-s', I mean to tell Emacs:  `I want you to make a backup and
> then save this file.'  The documentation for the `save-buffer' command
> says:
>
>    Prefixed by two C-u's,
>     unconditionally makes the previous version into a backup file.
>     ^^^^^^^^^^^^^^^
> Unfortunately, Emacs fails to respect my intent in many circumstances.
> First, I set `version-control' to t in order to keep numbered backups.
> Then in order to make backups in directories such as /tmp, I set
> `backup-enable-predicate' to be a function that always returns t.
> Next, in order to make backups when `vc' is involved, I set
> `vc-make-backup-files' to t, which confused me because it didn't take
> effect immediately -- I had to reload all my buffers before it did.
> This is because some vc hook set another variable, `backup-inhibited',
> when I first loaded those buffers.
>
> There are plenty of other documented ways for arbitrary modes to
> interfere with my intent when I type `C-u C-u' -- there are several
> hooks that are run during `save-buffer' before `backup-buffer' and
> conditions under which backups will be made.  This contradicts the
> documentation of `save-buffer' which explicitly says that `C-u C-u'
> means a backup will be made *unconditionally*.  I don't mind that
> there may be various heuristics for choosing whether to make a backup
> file when the user has not indicated a preference either way (by
> typing `C-u 0' or `C-u C-u' or whatnot), but it is very frustrating
> when the documentation says `unconditionally' and the implementation
> has many documented ways to interfere and add conditions.  Of course,
> the documentation could simply omit the word `unconditionally', but I
> don't want that either -- I want it really to be unconditional (as far
> as the documented hooks in Emacs go; of course someone could redefine
> `save-buffer', but it's the documented hooks and conditions such as
> `backup-inhibited' that bother me).
>
> I am using
>
> GNU Emacs 22.2.1 (i386-apple-darwin8.11.1, GTK+ Version 2.10.13) of 2008-07-06 on Oberon.local.
>
> In my .emacs file, relevant to backups, I have
>
> (defun always-backup (filename) filename t)
> (setq-default backup-enable-predicate 'always-backup)
> (setq-default version-control t)
> (setq-default vc-make-backup-files t)
> (setq-default backup-inhibited nil)
> (setq-default make-backup-files t)

bug#3702: Emacs fails to respect the intent in `C-u C-u C-x C-s' of making a backup

[Eli Zaretskii]

> From: Andrew Hyatt <ahyatt@gmail.com>
> Date: Fri, 17 Jun 2016 23:37:28 -0400
> Cc: 3702@debbugs.gnu.org
> 
> 
> Sorry for the late reply on this one.  This problem still exists in
> Emacs 25.  I agree that the current behavior doesn't match up with the
> documentation.
> 
> This probably is fixable by adding another free variable set by
> save-buffer and read in various functions, but there's a lot of those
> kind of variables already - the logic here, IMHO, could use a rewrite to
> make the interactions clear.  Maybe someone familiar with this code
> might give their opinion.

It sounds like this report boils down to interpreting the single word
"unconditionally" in the doc string of save-buffer.  All the rest is
actually a request for a feature that never existed in Emacs.  (FWIW,
I don't object to adding such a feature, if someone submits patches to
that effect.)

bug#3702: Emacs fails to respect the intent in `C-u C-u C-x C-s' of making a backup

[Eli Zaretskii]

> From: Andrew Hyatt <ahyatt@gmail.com>
> Date: Sat, 25 Jun 2016 23:22:42 -0400
> 
> 
> Eli Zaretskii <eliz@gnu.org> writes:
> 
> >> From: Andrew Hyatt <ahyatt@gmail.com>
> >> Date: Fri, 17 Jun 2016 23:37:28 -0400
> >> Cc: 3702@debbugs.gnu.org
> >> 
> >> 
> >> Sorry for the late reply on this one.  This problem still exists in
> >> Emacs 25.  I agree that the current behavior doesn't match up with the
> >> documentation.
> >> 
> >> This probably is fixable by adding another free variable set by
> >> save-buffer and read in various functions, but there's a lot of those
> >> kind of variables already - the logic here, IMHO, could use a rewrite to
> >> make the interactions clear.  Maybe someone familiar with this code
> >> might give their opinion.
> >
> > It sounds like this report boils down to interpreting the single word
> > "unconditionally" in the doc string of save-buffer.  All the rest is
> > actually a request for a feature that never existed in Emacs.  (FWIW,
> > I don't object to adding such a feature, if someone submits patches to
> > that effect.)
> 
> Maybe the solution then is just to change the wording in the
> documentation?  But it might take someone who is very familiar with the
> various exceptions to craft the new wording.

OK, I modified the doc string to the above effect on the emacs-25
branch, and I'm marking this bug done.

Thanks.