auto-save-visited-file-name

auto-save-visited-file-name

[Stefan Monnier]

Has anyone ever tried to use this variable ?

It has some really nasty effects.  Let's see a session:

% emacs -q --no-site-file
M-: (setq auto-save-visited-file-name t)
C-x C-f ~/tmp/foo.txt RET
type in some text with é or à
C-x C-s
# Now foo.txt has size 30, for instance, using latin-1.
SPC DEL
# Wait until the buffer is auto-saved.
# BUG-1: Now foo.txt has size 32, using emacs-mule encoding.
C-x C-s
# BUG-2: Emacs prompts because the file was "changed since saved".
# let's not save, but kill the buffer instead.
C-g
C-x k RET
# BUG-3: foo.txt does not exist any more because Emacs thought it
#        was the auto-save file and should thus be removed.

I think we need to do something.
Maybe fix the above bugs, which might uglify a good bit of the code.
Or maybe remove the option instead.
Or at least make it into a `defvar' rather than `defcustom'.


        Stefan

Re: auto-save-visited-file-name

[Kim F. Storm]

Stefan Monnier <monnier@IRO.UMontreal.CA> writes:

> Has anyone ever tried to use this variable ?

Not me.  Is this a new option?

> 
> It has some really nasty effects.  Let's see a session:
> 
> % emacs -q --no-site-file
> M-: (setq auto-save-visited-file-name t)
> C-x C-f ~/tmp/foo.txt RET
> type in some text with é or à
> C-x C-s
> # Now foo.txt has size 30, for instance, using latin-1.
> SPC DEL
> # Wait until the buffer is auto-saved.
> # BUG-1: Now foo.txt has size 32, using emacs-mule encoding.
> C-x C-s
> # BUG-2: Emacs prompts because the file was "changed since saved".
> # let's not save, but kill the buffer instead.
> C-g
> C-x k RET
> # BUG-3: foo.txt does not exist any more because Emacs thought it
> #        was the auto-save file and should thus be removed.
> 
> I think we need to do something.
> Maybe fix the above bugs, which might uglify a good bit of the code.

Re. bug 1, I haven't a clue.

.. but doesn't that mean that auto-save files are always saved 
in emacs-mule rather than the current coding of the buffer?
That sounds like a more generic error that needs fixing anyway.


Re. bug 2, checking the varibale in basic-save-buffer seems to be 
trivial.

Re. bug 3, checking the variable in delete-auto-save-file-if-necessary
seems to be trivial too.

> Or maybe remove the option instead.
> Or at least make it into a `defvar' rather than `defcustom'.

I think we should try to fix it...

-- 
Kim F. Storm <storm@cua.dk> http://www.cua.dk

Re: auto-save-visited-file-name

[Eli Zaretskii]

> From: Stefan Monnier <monnier@IRO.UMontreal.CA>
> Date: 15 Oct 2003 14:59:33 -0400
> 
> SPC DEL
> # Wait until the buffer is auto-saved.
> # BUG-1: Now foo.txt has size 32, using emacs-mule encoding.

A real bug, IMHO: we write auto-save files in emacs-mule, but should
use the buffer's file encoding when auto-save-visited-file-name is
set.  That might mean dealing with characters that cannot be encoded
in the buffer-file-coding-system.

Re: auto-save-visited-file-name

[Stefan Monnier]

> .. but doesn't that mean that auto-save files are always saved 
> in emacs-mule rather than the current coding of the buffer?
> That sounds like a more generic error that needs fixing anyway.

No, I think it's a very reasonable choice (and I know it's been discussed
here and more recently on XEmacs's lists).
This is the only encoding that we know for sure will not lose any
information, will succeed without needing to prompt the user for more
details, and it is also more efficient which I still consider relevant in
the context of auto-saving.  It's not perfect, but the other options aren't
much more attractive.
IIUC, the internal format in emacs-unicode is utf-8, so the auto-save
format will naturally use utf-8 at that point.


        Stefan

Re: auto-save-visited-file-name

[Kim F. Storm]

Stefan Monnier <monnier@IRO.UMontreal.CA> writes:

> > .. but doesn't that mean that auto-save files are always saved 
> > in emacs-mule rather than the current coding of the buffer?
> > That sounds like a more generic error that needs fixing anyway.
> 
> No, I think it's a very reasonable choice (and I know it's been discussed
> here and more recently on XEmacs's lists).
> This is the only encoding that we know for sure will not lose any
> information, will succeed without needing to prompt the user for more
> details, and it is also more efficient which I still consider relevant in
> the context of auto-saving.  It's not perfect, but the other options aren't
> much more attractive.

Well, we could use the buffer's coding if it suffices -- and only revert
to emacs-mule if it doesn't.  That would not require any prompting...

I assume that the original buffer's coding is used if the
auto-save file is restored.

But I have no strong opinions, so let's not discuss it further.


-- 
Kim F. Storm <storm@cua.dk> http://www.cua.dk

Re: auto-save-visited-file-name

[Richard Stallman]

Does this solve the problems you found with auto-save-visited-file-name?

*** buffer.c.~1.438.~	Thu Sep 11 09:51:05 2003
--- buffer.c	Fri Oct 17 11:37:36 2003
***************
*** 1429,1435 ****
    if (STRINGP (b->auto_save_file_name)
        && b->auto_save_modified != 0
        && BUF_SAVE_MODIFF (b) < b->auto_save_modified
!       && BUF_SAVE_MODIFF (b) < BUF_MODIFF (b))
      {
        Lisp_Object tem;
        tem = Fsymbol_value (intern ("delete-auto-save-files"));
--- 1429,1436 ----
    if (STRINGP (b->auto_save_file_name)
        && b->auto_save_modified != 0
        && BUF_SAVE_MODIFF (b) < b->auto_save_modified
!       && BUF_SAVE_MODIFF (b) < BUF_MODIFF (b)
!       && NILP (Fsymbol_value (intern ("auto-save-visited-file-name"))))
      {
        Lisp_Object tem;
        tem = Fsymbol_value (intern ("delete-auto-save-files"));


*** fileio.c.~1.494.~	Thu Sep 11 09:51:19 2003
--- fileio.c	Fri Oct 17 11:35:48 2003
***************
*** 4684,4690 ****
  {
    Lisp_Object val;
  
!   if (auto_saving)
      {
        /* We use emacs-mule for auto saving... */
        setup_coding_system (Qemacs_mule, coding);
--- 4684,4691 ----
  {
    Lisp_Object val;
  
!   if (auto_saving
!       && ! NILP (Fsymbol_value (intern ("auto-save-visited-file-name"))))
      {
        /* We use emacs-mule for auto saving... */
        setup_coding_system (Qemacs_mule, coding);
***************
*** 5212,5218 ****
      error ("IO error writing %s: %s", SDATA (filename),
  	   emacs_strerror (save_errno));
  
!   if (visiting)
      {
        SAVE_MODIFF = MODIFF;
        XSETFASTINT (current_buffer->save_length, Z - BEG);
--- 5213,5222 ----
      error ("IO error writing %s: %s", SDATA (filename),
  	   emacs_strerror (save_errno));
  
!   if (visiting
!       || (auto_saving
! 	  && ! NILP (Fsymbol_value (intern ("auto-save-visited-file-name")))))
! 
      {
        SAVE_MODIFF = MODIFF;
        XSETFASTINT (current_buffer->save_length, Z - BEG);
***************
*** 5775,5785 ****
    minibuffer_auto_raise = 0;
    auto_saving = 1;
  
!   /* First, save all files which don't have handlers.  If Emacs is
!      crashing, the handlers may tweak what is causing Emacs to crash
!      in the first place, and it would be a shame if Emacs failed to
!      autosave perfectly ordinary files because it couldn't handle some
!      ange-ftp'd file.  */
    for (do_handled_files = 0; do_handled_files < 2; do_handled_files++)
      for (tail = Vbuffer_alist; GC_CONSP (tail); tail = XCDR (tail))
        {
--- 5779,5792 ----
    minibuffer_auto_raise = 0;
    auto_saving = 1;
  
!   /* On first pass, save all files that don't have handlers.
!      On second pass, save all files that do have handlers.
! 
!      If Emacs is crashing, the handlers may tweak what is causing
!      Emacs to crash in the first place, and it would be a shame if
!      Emacs failed to autosave perfectly ordinary files because it
!      couldn't handle some ange-ftp'd file.  */
! 
    for (do_handled_files = 0; do_handled_files < 2; do_handled_files++)
      for (tail = Vbuffer_alist; GC_CONSP (tail); tail = XCDR (tail))
        {

Re: auto-save-visited-file-name

[Stefan Monnier]

> Does this solve the problems you found with auto-save-visited-file-name?

It helps, but see below.

> *** buffer.c.~1.438.~	Thu Sep 11 09:51:05 2003
> --- buffer.c	Fri Oct 17 11:37:36 2003
> ***************
> *** 1429,1435 ****
>     if (STRINGP (b->auto_save_file_name)
>         && b->auto_save_modified != 0
>         && BUF_SAVE_MODIFF (b) < b->auto_save_modified
> !       && BUF_SAVE_MODIFF (b) < BUF_MODIFF (b))
>       {
>         Lisp_Object tem;
>         tem = Fsymbol_value (intern ("delete-auto-save-files"));
> --- 1429,1436 ----
>     if (STRINGP (b->auto_save_file_name)
>         && b->auto_save_modified != 0
>         && BUF_SAVE_MODIFF (b) < b->auto_save_modified
> !       && BUF_SAVE_MODIFF (b) < BUF_MODIFF (b)
> !       && NILP (Fsymbol_value (intern ("auto-save-visited-file-name"))))
>       {
>         Lisp_Object tem;
>         tem = Fsymbol_value (intern ("delete-auto-save-files"));

I think the above has a problem: auto-save-visited-file-name is only
consulted when setting up buffer-auto-save-file-name, so it can have a value
inconsistent with the actual situation.  Better check whether
buffer-auto-save-file-name is equal to buffer-file-name.


        Stefan

Re: auto-save-visited-file-name

[Kevin Rodgers]

Richard Stallman wrote:

> Does this solve the problems you found with auto-save-visited-file-name?

That patch checks the vaule of auto-save-visited-file-name.  But what if the
user instead explicitly sets buffer-auto-save-file-name to buffer-file-name (or
to any name with the same true name as buffer-file-name)?

-- 
Kevin Rodgers

Re: auto-save-visited-file-name

[Richard Stallman]

    That patch checks the vaule of auto-save-visited-file-name.  But what if the
    user instead explicitly sets buffer-auto-save-file-name to buffer-file-name (or
    to any name with the same true name as buffer-file-name)?

the code as actually installed compares the file names.