bug#23203: 25.0.91; some loaddefs files have auto-save remnants after building (and install doesn't ignore them)

[Eli Zaretskii]

> Date: Sun, 3 Apr 2016 21:11:34 +0100
> From: "Phillip Lord" <phillip.lord@russet.org.uk>
> Cc: "Phillip Lord" <phillip.lord@russet.org.uk>,
>  23203@debbugs.gnu.org,
>  stromeko@nexgo.de
> 
> Actually, my first batch did not exempt autoload files. It only did that
> when called using batch-update-autoloads: i.e. in batch, and in a function
> which is commented (but not documented) to be for use during the Emacs
> build process. Interactive calls to "update-directory-autoloads" and the
> like would be unaffected.
> 
> I think that the comment is wrong, as I use this function outside of the
> Emacs build process (and I am sure others do also), so I think we should
> remove that. Alternatively, if it is right, we should move it do the
> docstring.

Fixing the doc string is always a good thing, but it's orthogonal to
the issue at hand.  (I do agree that the comment is wrong.)

> But, I think having a "no backup files" default is still sensible for a
> function used in batch; if it is always used this way in the Emacs build,
> then it it telling us something. So, I would change it, and update the
> docstring. We could also split the function, so that there is a version
> which does not suppress backup files.

I don't think I agree that batch mode should behave differently wrt
backups.

How about this alternative: only disable backing up the initial
(effectively empty) contents of the autoloads file?  AFAICT, the
backup files are created during a bootstrap only because we first
write the initial "rubric" into it, using write-region, and only after
that visit it.  This defeats the normal mechanism of backing up just
once per session, and leaves a backup file whose contents are not
interesting.

So an alternative would be to modify autoload-ensure-default-file so
that it returns some indication about the fact it created the file,
and then change its caller to set buffer-backed-up after it visits the
file, thus preventing the backup _only_ when the file is first
created.

This should at least solve Achim's problem, but without affecting
anything else.

WDYT?