make-auto-save-file-name: not using name transforms for mere buffers

make-auto-save-file-name: not using name transforms for mere buffers

[Florian von Savigny]

Hi,

I am using Linux on a home directory that is physically identical with
a vfat partition (don't discuss this please; I am well aware it's
technically markedly inferior, but I nevertheless have my reasons).

I also like the auto-save feature and have also switched it on for
buffers not visiting files. However, whenever I am writing a mail,
emacs tries to save it into an auto-save file with asterisks in its
name, which doesn't work since this is a forbidden character on VFAT.

I have tried customising auto-save-file-name-transforms, but this does
not work, apparently because make-auto-save-file-name only uses it if
the buffer is visiting a file (does anybody know why on earth?).

Is there any standard way to make make-auto-save-file-name use it also
for buffer names, or do I simply have to hack the function?

Any help would be greatly appreciated,


-- 


Florian v. Savigny

If you are going to reply in private, please be patient, as I only
check for mail something like once a week. - Si vous allez répondre
personellement, patientez s.v.p., car je ne lis les courriels
qu'environ une fois par semaine.

Re: make-auto-save-file-name: not using name transforms for mere buffers

[Eli Zaretskii]

> From: Florian von Savigny <florian265@uboot.com>
> Newsgroups: gnu.emacs.help
> Date: 10 Dec 2003 17:07:36 +0100
> 
> I have tried customising auto-save-file-name-transforms, but this does
> not work, apparently because make-auto-save-file-name only uses it if
> the buffer is visiting a file (does anybody know why on earth?).

Because its doc string says, in its very first line:

   *Transforms to apply to buffer file name before making auto-save file name.
                        ^^^^^^^^^^^^^^^^^^^
So buffers with no file name are not subject to such transforms.

> Is there any standard way to make make-auto-save-file-name use it also
> for buffer names, or do I simply have to hack the function?

Instead of hacking make-auto-save-file-name, I'd suggest to hack
convert-standard-file-name.  On GNU and Unix systems, this function
simply returns its argument.  What you need is to redefine it so that
it recognizes that its argument file resides on a VFAT volume, and
then applies the same transformations done by the MS-Windows variant
of convert-standard-file-name (see w32-fns.el).

Re: make-auto-save-file-name: not using name transforms for mere buffers

[Kevin Rodgers]

Florian von Savigny wrote:

> I also like the auto-save feature and have also switched it on for
> buffers not visiting files. However, whenever I am writing a mail,
> emacs tries to save it into an auto-save file with asterisks in its
> name, which doesn't work since this is a forbidden character on VFAT.
> 
> I have tried customising auto-save-file-name-transforms, but this does
> not work, apparently because make-auto-save-file-name only uses it if
> the buffer is visiting a file (does anybody know why on earth?).
> 
> Is there any standard way to make make-auto-save-file-name use it also
> for buffer names, or do I simply have to hack the function?

Why not just redefine make-auto-save-file-name (and auto-save-file-name-p)?

-- 
Kevin Rodgers

Re: make-auto-save-file-name: not using name transforms for mere buffers

[Florian von Savigny]

"Eli Zaretskii" <eliz@elta.co.il> writes:

> Because its doc string says, in its very first line:
> 
>    *Transforms to apply to buffer file name before making auto-save file name.
>                         ^^^^^^^^^^^^^^^^^^^
> So buffers with no file name are not subject to such transforms.

Ok, so it doesn't do it because it is not told so ;-) but I still
wonder why it is told not to do so (I mean why, in the first place,
the author decided not to apply this feature to file names derived
from mere buffer names).

> Instead of hacking make-auto-save-file-name, I'd suggest to hack
> convert-standard-file-name.  On GNU and Unix systems, this function
> simply returns its argument.  What you need is to redefine it so that
> it recognizes that its argument file resides on a VFAT volume, and
> then applies the same transformations done by the MS-Windows variant
> of convert-standard-file-name (see w32-fns.el).

Wow, thanks for this hint; it does sound like a sane approach. But I
still need to understand when, in the particular case of auto-save
file names derived from buffer names, convert-standard-file-name takes
effect. AFAICS, it is not called directly from within
make-auto-save-file-name. I couldn't make out where, or if, it is
called indirectly.

However, I've made a start. I've decided that no direct function
to the kernel to query mounted filesystems seems available, and
between looking at the output of /proc/mounts and calling mount with
no arguments, I've decided to do the latter (I don't know if any
approach is favourable over the other). I thought it would be more
feasible to cache this information:

(setq file-systems nil)  ; alist of ((MOUNTPOINT . FS-TYPE) ... etc.

(defun cache-file-systems ()
  "Function to set `file-systems' automatically."
  (interactive)  ; useful after mounts, could also be called by gnuclient
  (call-process "mount" nil "*mount*" nil)
  (save-excursion
    (set-buffer "*mount*")
    (goto-char (point-min))
    (while
	(search-forward-regexp " on \\([^ ]+\\) type \\([^ ]+\\)" nil t)
      (setq file-systems 
	    (cons 
	     (cons (match-string 1) (match-string 2))
		  file-systems)))
    (kill-buffer "*mount*")))


So far, this works. Now we need to query it:


(defun filename-get-fs (filename)
  "Returns, as a string, the type of file system within the realm of
which a filename falls. Depends on the correct setting of `file-systems',
which can be guaranteed by setting it with `cache-file-systems'."
  ;; for this to be really sane, file-systems must be sorted such that
  ;; those pairs with the longest mount points come first
  (setq filename (expand-file-name filename))
  (let ((ind 0)
	mount-point)
    (catch 'result
      (while (setq mount-point (car (nth ind file-systems)))
	(if (< (length mount-point)
	       (length filename))
	    (if (string-equal mount-point 
			      (substring filename 0 (length mount-point)))
		(throw 'result (cdr (nth ind file-systems)))))
	  (setq ind (+ ind 1))))))


But here I'm stuck. For the function to be sane, the list must be
sorted longest mountpoint first (notably, "/" must be the last
entry, because it matches everything). How to do this?


This function working, it could be used inside
convert-standard-file-name like: 

        (cond ((equal (filename-get-fs filename) "vfat")
            (;; apply certain transforms))
              (( ...

(haven't checked the syntax, but you get the idea)


For looking up transforms, I'd like to have a list of alists of a
value and a list of alists each ;-) ... is that possible? This one
here doesn't work:

  (defvar file-name-char-transforms
    '(("vfat" . (("*" . "_") (":" . "%") 
		  ("ä" . "ae") ("ö" . "oe") ("ü" . "ue") ("ß" "ss")))
      ("fat" . (("*" . "_") (":" "%"))))
    "Complexly structured list containing rules how to translate
filename characters in the context of given file systems.")



-- 


Florian v. Savigny

If you are going to reply in private, please be patient, as I only
check for mail something like once a week. - Si vous allez répondre
personellement, patientez s.v.p., car je ne lis les courriels
qu'environ une fois par semaine.

-- 


Florian v. Savigny

If you are going to reply in private, please be patient, as I only
check for mail something like once a week. - Si vous allez répondre
personellement, patientez s.v.p., car je ne lis les courriels
qu'environ une fois par semaine.

Re: make-auto-save-file-name: not using name transforms for mere buffers

[Eli Zaretskii]

> From: Florian von Savigny <florian265@uboot.com>
> Newsgroups: gnu.emacs.help
> Date: 11 Dec 2003 12:46:02 +0100
> 
> > Because its doc string says, in its very first line:
> > 
> >    *Transforms to apply to buffer file name before making auto-save file name.
> >                         ^^^^^^^^^^^^^^^^^^^
> > So buffers with no file name are not subject to such transforms.
> 
> Ok, so it doesn't do it because it is not told so ;-) but I still
> wonder why it is told not to do so (I mean why, in the first place,
> the author decided not to apply this feature to file names derived
> from mere buffer names).

Probably because the rules that are supposed to be applied to file
names aren't fit for buffer names.  In other words, the applications
of auto-save-file-name-transforms that the programmer had in mind did
not include the kind of situation in which you've found yourself.

> Wow, thanks for this hint; it does sound like a sane approach. But I
> still need to understand when, in the particular case of auto-save
> file names derived from buffer names, convert-standard-file-name takes
> effect. AFAICS, it is not called directly from within
> make-auto-save-file-name.

It _is_ called directly, at least in my version of Emacs, albeit only
for non-Unix platforms.  But that should be easy to fix: simply
modify the condition under which make-auto-save-file-name calls
convert-standard-file-name.

If your version of make-auto-save-file-name doesn't call
convert-standard-file-name on any OS, perhaps you have an old Emacs
(or else I have a too new Emacs ;-).

> However, I've made a start. I've decided that no direct function
> to the kernel to query mounted filesystems seems available, and
> between looking at the output of /proc/mounts and calling mount with
> no arguments, I've decided to do the latter (I don't know if any
> approach is favourable over the other).

Why not call `df' instead?  I think "df ." should produce a line where
one field gives you the type of the filesystem that you want.

> For looking up transforms, I'd like to have a list of alists of a
> value and a list of alists each ;-) ... is that possible?

I'd try to use regexps and replace-match instead.

> This one
> here doesn't work:
> 
>   (defvar file-name-char-transforms
>     '(("vfat" . (("*" . "_") (":" . "%") 
> 		  ("ä" . "ae") ("ö" . "oe") ("ü" . "ue") ("ß" "ss")))
>       ("fat" . (("*" . "_") (":" "%"))))
>     "Complexly structured list containing rules how to translate
> filename characters in the context of given file systems.")

What doesn't work, exactly?

Re: make-auto-save-file-name: not using name transforms for mere buffers

[Florian von Savigny]

Newsgroups: gnu.emacs.help
Subject: Re: make-auto-save-file-name: not using name transforms for mere	buffers
References: <m3y8tkvepj.fsf@uboot.com> <mailman.1505.1071080061.399.help-gnu-emacs@gnu.org> <m3smjrtw5h.fsf@uboot.com> <mailman.1581.1071159027.399.help-gnu-emacs@gnu.org>
From: Florian von Savigny <florian265@uboot.com>
--text follows this line--

Eli Zaretskii <eliz@elta.co.il> writes:

> > Wow, thanks for this hint; it does sound like a sane approach. But I
> > still need to understand when, in the particular case of auto-save
> > file names derived from buffer names, convert-standard-file-name takes
> > effect. AFAICS, it is not called directly from within
> > make-auto-save-file-name.
> 
> It _is_ called directly, at least in my version of Emacs, albeit only
> for non-Unix platforms.  

So is it one of msdos-long-file-names or dos-8+3-filename which calls
it? These were non-operational, undefined functions (presumably not
loaded) when running under Linux.

> But that should be easy to fix: simply modify the condition under
> which make-auto-save-file-name calls convert-standard-file-name.

I couldn't spot it called under any condition (neither directly nor
indirectly), but maybe I am overlooking something:

(defun make-auto-save-file-name ()
  "Return file name to use for auto-saves of current buffer.
Does not consider `auto-save-visited-file-name' as that variable is checked
before calling this function.  You can redefine this for customization.
See also `auto-save-file-name-p'."
  (if buffer-file-name
      (let ((list auto-save-file-name-transforms)
	    (filename buffer-file-name)
	    result)
	;; Apply user-specified translations
	;; to the file name.
	(while (and list (not result))
	  (if (string-match (car (car list)) filename)
	      (setq result (replace-match (cadr (car list)) t nil
					  filename)))
	  (setq list (cdr list)))
	(if result (setq filename result))

	(if (and (eq system-type 'ms-dos)
		 (not (msdos-long-file-names)))
	    ;; We truncate the file name to DOS 8+3 limits before
	    ;; doing anything else, because the regexp passed to
	    ;; string-match below cannot handle extensions longer than
	    ;; 3 characters, multiple dots, and other atrocities.
	    (let ((fn (dos-8+3-filename
		       (file-name-nondirectory buffer-file-name))))
	      (string-match "\\`\\([^.]+\\)\\(\\.\\(..?\\)?.?\\|\\)\\'" fn)
	      (concat (file-name-directory buffer-file-name)
		      "#" (match-string 1 fn)
		      "." (match-string 3 fn) "#"))
	  (concat (file-name-directory filename)
		  "#"
		  (file-name-nondirectory filename)
		  "#")))
    ;; else part of the if buffer-file-name:
    ;; Deal with buffers that don't have any associated files.  (Mail
    ;; mode tends to create a good number of these.)

    (let ((buffer-name (buffer-name))
	  (limit 0))
      ;; Eliminate all slashes and backslashes by
      ;; replacing them with sequences that start with %.
      ;; Quote % also, to keep distinct names distinct.
      (while (string-match "[/\\%]" buffer-name limit)
	(let* ((character (aref buffer-name (match-beginning 0)))
	       (replacement
		(cond ((eq character ?%) "%%")
		      ((eq character ?/) "%+")
		      ((eq character ?\\) "%-"))))
	  (setq buffer-name (replace-match replacement t t buffer-name))
	  (setq limit (1+ (match-end 0)))))
      ;; Generate the file name.
      (expand-file-name
       (format "#%s#%s#" buffer-name (make-temp-name ""))
       ;; Try a few alternative directories, to get one we can write it.
       (cond
	((file-writable-p default-directory) default-directory)
	((file-writable-p "/var/tmp/") "/var/tmp/")
	("~/"))))))


> If your version of make-auto-save-file-name doesn't call
> convert-standard-file-name on any OS, perhaps you have an old Emacs
> (or else I have a too new Emacs ;-).

Oops, I wouldn't know how to run the same Emacs under Win. In any
case, while I do use Emacs both under Linux and Windows, they are
completely independent, and different major versions (gnu/linux:
21.2.1; windows-nt: 20.7.1).

I have had a look at how my Windows version does it ("*scratch*" will
become "#!scratch!#-16740779UIZ#"), and it appears that in w32-fns.el,
which is presumably loaded only under windows (I have just managed to
get the same behaviour by loading this library file under Linux, BTW,
though it hasn't loaded without error), make-auto-save-file-name is
renamed to 'original-make-auto-save-file-name and then redefined to

  (convert-standard-filename (original-make-auto-save-file-name))

convert-standard-filename is also defined to do something less
trivial; it converts all forbidden characters to "!", which is somehow
quite symbolic.


> > However, I've made a start. I've decided that no direct function
> > to the kernel to query mounted filesystems seems available, and
> > between looking at the output of /proc/mounts and calling mount with
> > no arguments, I've decided to do the latter (I don't know if any
> > approach is favourable over the other).
> 
> Why not call `df' instead?  I think "df ." should produce a line where
> one field gives you the type of the filesystem that you want.

df version 4.1 doesn't do so on my Debian woody, though it has an
option to exclude certain types of filesystems from being
considered. It sounds like it could be done on demand each time, if it
works. This would certainly be highly preferable, because nothing
could get out of date like that, and it would be less of a hassle to
program.

> > For looking up transforms, I'd like to have a list of alists of a
> > value and a list of alists each ;-) ... is that possible?
> 
> I'd try to use regexps and replace-match instead.

I have actually not yet made up my mind about the method of
transforming (I don't expect this to be problematic); what I want to
use the "list of alists of a value and a list of alists each" for is
to store rules on what type of fs what character to replace with what,
so read

  ("vfat" . (("*" . "_") ...

to mean "on a vfat fs, replace "*" with "_" (and so on)". I wanted to
make this user-customisable, hence the variable. Of course, come to
think of it, the car, i.e. "*" in this case, could just as well be a
regexp instead, i.e. "\\*" in this case, iirc.

> What doesn't work, exactly?

Oops, I tested it again, and, in fact, it works as expected. Probably
I simply messed up the cars and cdrs yesterday.

To sum up, I gather one would have to make changes similar to those in
w32-fns.el, up to the redefinition of make-auto-save-file-name, but
then define convert-standard-filename in the way you have suggested
(make it behave file system sensitive).

Would you agree, or do you think I have missed a point (does your
emacs version actually do it differently)?


And one call for help: could anyone please suggest a way of sorting
the list of alists I called "file-systems" demonstrated earlier in
this thread such that those alists whose car is the longer string come
first?

-- 


Florian v. Savigny

If you are going to reply in private, please be patient, as I only
check for mail something like once a week. - Si vous allez répondre
personellement, patientez s.v.p., car je ne lis les courriels
qu'environ une fois par semaine.

-- 


Florian v. Savigny

If you are going to reply in private, please be patient, as I only
check for mail something like once a week. - Si vous allez répondre
personellement, patientez s.v.p., car je ne lis les courriels
qu'environ une fois par semaine.

Re: make-auto-save-file-name: not using name transforms for mere buffers

[Eli Zaretskii]

> From: Florian von Savigny <florian265@uboot.com>
> Newsgroups: gnu.emacs.help
> Date: 12 Dec 2003 02:19:30 +0100
> 
> > > Wow, thanks for this hint; it does sound like a sane approach. But I
> > > still need to understand when, in the particular case of auto-save
> > > file names derived from buffer names, convert-standard-file-name takes
> > > effect. AFAICS, it is not called directly from within
> > > make-auto-save-file-name.
> > 
> > It _is_ called directly, at least in my version of Emacs, albeit only
> > for non-Unix platforms.  
> 
> So is it one of msdos-long-file-names or dos-8+3-filename which calls
> it? These were non-operational, undefined functions (presumably not
> loaded) when running under Linux.

It sounds like the code I've been looking at exists only in the CVS
version of Emacs.  So it seems like in your case the best option is
to redefine make-auto-save-file-name with your own version.

> > Why not call `df' instead?  I think "df ." should produce a line where
> > one field gives you the type of the filesystem that you want.
> 
> df version 4.1 doesn't do so on my Debian woody

Sorry, I shopuld have written "df -T .".  The -T switch is important.

Re: make-auto-save-file-name: not using name transforms for mere buffers

[Kevin Rodgers]

Florian von Savigny wrote:

> And one call for help: could anyone please suggest a way of sorting
> the list of alists I called "file-systems" demonstrated earlier in
> this thread such that those alists whose car is the longer string come
> first?

;; Sort ((MOUNT-POINT . FILESYSTEM-TYPE) ...) alist by MOUNT-POINT:
(sort file-systems (lambda (cons-1 cons-2)
		     (string-lessp (car cons-1) (car cons-2))))


-- 
Kevin Rodgers

Re: make-auto-save-file-name: not using name transforms for mere buffers

[Florian von Savigny]

Kevin Rodgers <ihs_4664@yahoo.com> writes:

> ;; Sort ((MOUNT-POINT . FILESYSTEM-TYPE) ...) alist by MOUNT-POINT:
> (sort file-systems (lambda (cons-1 cons-2)
> 		     (string-lessp (car cons-1) (car cons-2))))

Thanks a lot, Kevin. I am sorry that meanwhile this has turned out to
be unnecessary for the present problem, as Eli's suggestion of using
'df -T' at the time of creating the file name has made my caching
mechanism superfluous (and the whole construction more failsafe).

I am, however, grateful anyway, as I use alists a lot, and have saved
your solution away. I'll be sure to use it.

-- 


Florian v. Savigny

If you are going to reply in private, please be patient, as I only
check for mail something like once a week. - Si vous allez répondre
personellement, patientez s.v.p., car je ne lis les courriels
qu'environ une fois par semaine.

Re: make-auto-save-file-name: not using name transforms for mere buffers

[Florian von Savigny]

"Eli Zaretskii" <eliz@elta.co.il> writes:

> It sounds like the code I've been looking at exists only in the CVS
> version of Emacs.  So it seems like in your case the best option is
> to redefine make-auto-save-file-name with your own version.

Hmm. I'll look into where convert-file-name is used. I still find
intriguing the idea of making emacs foreign-fs-proof even if running
under GNU/Linux. Maybe I'll also copy the scheme Emacs for Win32 uses. 

(What's the CVS version of Emacs? Is it the latest cutting-edge
developer version, so one can expect its features to turn up in one of
the future official versions? Or is it more of a specialised version?)

> Sorry, I should have written "df -T .".  The -T switch is important.

Rats, another one I simply overlooked, since of course I did look
at the manpage ...

Thanks a lot, Eli. That facilitates matters considerably.

-- 


Florian v. Savigny

If you are going to reply in private, please be patient, as I only
check for mail something like once a week. - Si vous allez répondre
personellement, patientez s.v.p., car je ne lis les courriels
qu'environ une fois par semaine.

Re: make-auto-save-file-name: not using name transforms for mere buffers

[Eli Zaretskii]

> From: Florian von Savigny <florian265@uboot.com>
> Newsgroups: gnu.emacs.help
> Date: 13 Dec 2003 12:51:55 +0100
> 
> (What's the CVS version of Emacs? Is it the latest cutting-edge
> developer version, so one can expect its features to turn up in one of
> the future official versions? Or is it more of a specialised version?)

It's the development code of Emacs.  CVS is a configuration-control
package that lets you access the development sources via the Internet.
Emacs development is done thru CVS.

Re: make-auto-save-file-name: not using name transforms for mere buffers

[Florian von Savigny]

"Eli Zaretskii" <eliz@elta.co.il> writes:

> > (What's the CVS version of Emacs? Is it the latest cutting-edge
> > developer version, so one can expect its features to turn up in one of
> > the future official versions? Or is it more of a specialised version?)
> 
> It's the development code of Emacs.  

I see; so it's rational to expect that future official versions of
Emacs will use convert-standard-filename the way your version does,
isn't it?

-- 


Florian v. Savigny

If you are going to reply in private, please be patient, as I only
check for mail something like once a week. - Si vous allez répondre
personellement, patientez s.v.p., car je ne lis les courriels
qu'environ une fois par semaine.

Re: make-auto-save-file-name: not using name transforms for mere buffers

[Eli Zaretskii]

> From: Florian von Savigny <florian265@uboot.com>
> Newsgroups: gnu.emacs.help
> Date: 13 Dec 2003 20:52:21 +0100
> 
> so it's rational to expect that future official versions of
> Emacs will use convert-standard-filename the way your version does,
> isn't it?

Yes.  However, the code I have still calls convert-standard-filename
only on non-Posix platforms.  So perhaps it's a good idea to suggest a
change whereby convert-standard-filename is called on all platforms
regardless, for the cases such as yours.  If you decide to post such a
suggestion, please do so on gnu.emacs.bug or emacs-devel@gnu.org.

TIA

Re: make-auto-save-file-name: not using name transforms for mere buffers

[David Combs]

In article <m34qw5gckk.fsf@uboot.com>,
Florian von Savigny  <florian265@uboot.com> wrote:
SNIP
>
>> Sorry, I should have written "df -T .".  The -T switch is important.

What's this -T option to df?

And why is it "important" here?

My isp (panix.com, netbsd) doesn't have it,
nor does Solaris 9.

Thanks

David

Re: make-auto-save-file-name: not using name transforms for mere buffers

[Eli Zaretskii]

> From: dkcombs@panix.com (David Combs)
> Newsgroups: gnu.emacs.help
> Date: Sun, 4 Jan 2004 03:57:53 +0000 (UTC)
> >
> >> Sorry, I should have written "df -T .".  The -T switch is important.
> 
> What's this -T option to df?

It causes `df' to print the filesystem's type.

> And why is it "important" here?

Because that's what the OP wanted: the type of the filesystem where
Emacs was going to save a file (or was it auto-save?).  A certain
type of filesystem needed a special treatment on the OP's system.

> My isp (panix.com, netbsd) doesn't have it,
> nor does Solaris 9.

It's an extension of GNU `df'.  The OP was using a GNU/Linux system,
so I was certain his `df' had that option.  If you install GNU
Fileutils, so will yours.