Re: New names for VC local version backups

[Andre Spiegel]

Eli Zaretskii <address@hidden> writes:

> Alternatively, instead of making the version part of the file name, we 
> could make it a directory:
> 
>     version x.y.z or foobar.c -> versions/x_y_z/foobar.c
> 

Hmm, this doesn't appeal to me.  It would result in an explosion of
subdirectories under `versions'.  It would be difficult to find a
version of a particular file if you don't happen to know the version
number.  Completely unrelated files would end up in the same
subdirectory only because they happen to have the same version number.

I rather think that using some kind of tilde-suffix is a very natural
extension of the Emacs backup facilities.  This has already been done
by C-x v ~ for a long time, and I think it is natural and conceptually
clean to build on that.

It means that we have three kinds of backups: ordinary backups (file~,
might also be called "session backups", because that's what they
technically are), numbered backups (file.~num~) and version backups
(file.~rev~), defined as a copy of some version rev that already
exists in the repository.

A special kind of version backup is one that was created automatically
(and should be temporary), this is identified by putting another `.'
before rev, as Stefan suggested.  Gerd seems to agree with the idea,
how about you, Eli?  If it's ok with you I'll go ahead and make the
change (it's a one- or two-liner now).

Concerning 8.3 file systems --

> The simplest solution would be to run the file name through
> convert-standard-filename (it's a no-op on Unix).  The hard thing to
> do is to make Emacs recognize such munged file names when it sees
> them in the directory.  [...] It is possible to decide that this
> feature won't work on 8+3 filesystems (e.g., numbered backups don't
> work in that case), if it turns out too expensive to support.

I think that's what we should do.  However, I think we need to make
sure that Emacs does not even attempt to create a version backup on
such a file system, effectively disabling the feature.  How could that
be checked?  Are numbered backups disabled automatically on MS-DOS?
Or is it the user's responsibility not to enable them?