VC and select-safe-coding-system

VC and select-safe-coding-system

[Andre Spiegel]

I've been told that on a file under VC control, a user abort in
select-safe-coding-system can lose data.  A few days ago, I asked on
this list whether anyone could tell me the precise circumstances when
this happens.  I got no reply.  So I've tried to create such a situation
myself, but couldn't spot any such problem.  I also inspected the code
carefully, but couldn't see any potential difficulties.

If anyone has any more information on this matter, please let me know.

Re: VC and select-safe-coding-system

[Richard Stallman]

    I've been told that on a file under VC control, a user abort in
    select-safe-coding-system can lose data.  A few days ago, I asked on
    this list whether anyone could tell me the precise circumstances when
    this happens.  I got no reply.

I have seen this happen, too.  What happened, I think,
is that the file had already been renamed to a backup file.
So there was no longer any file with the visited name.

To see this problem, I think you need to set vc-make-backup-files to t.

Re: VC and select-safe-coding-system

[Stefan Monnier]

> I've been told that on a file under VC control, a user abort in
> select-safe-coding-system can lose data.  A few days ago, I asked on
> this list whether anyone could tell me the precise circumstances when
> this happens.  I got no reply.  So I've tried to create such a situation
> myself, but couldn't spot any such problem.  I also inspected the code
> carefully, but couldn't see any potential difficulties.

The only think I remember related to this is a problem that's
independent from VC:

If select-safe-coding-system prompts the user and the user
aborts with C-g, the file might end up "lost".

IIRC, the problem was that when `write-region' is called, the pre-existing
file has already been renamed to its backup name, so when write-region
calls select-safe-coding-system, the file is temporarily absent.

Maybe this bug was "fixed" since my original report by renaming the backup
file back to its non-backup name in case of an abort by the user.

But I think the problem of having a "temporarily missing file" should
be fixed as well (there will always be such a time-window anyway, but
we should make it shorter by making sure we don't prompt the user during
this time).


	Stefan

Re: VC and select-safe-coding-system

[Richard Stallman]

    Maybe this bug was "fixed" since my original report by renaming the backup
    file back to its non-backup name in case of an abort by the user.

I had forgotten making that change.
That might indeed have fixed it.  

    But I think the problem of having a "temporarily missing file" should
    be fixed as well (there will always be such a time-window anyway, but
    we should make it shorter by making sure we don't prompt the user during
    this time).

That may require substantial rearrangement.  Anyone want to see how much?

Re: VC and select-safe-coding-system

[Andre Spiegel]

On Tue, 2002-09-10 at 01:34, Richard Stallman wrote:

> I had forgotten making that change.
> That might indeed have fixed it.  

So I will consider the issue as closed.  If anything along these lines
happens again, let me know.