Emacs/Tramp - remote files via shell

Michael Forster writes:
 > I am using WinXP, XEmacs native, no Cygwin, SSH 2 (binary from
 > http://www.ssh.com) and got into the same problems as Pete. Using the
 > following patch Tramp works for me:

Does dired (C-x d) work?
-- 
Pete Forman                -./\.-  Disclaimer: This post is originated
WesternGeco                  -./\.-   by myself and does not represent
pete.forman@...    -./\.-   opinion of Schlumberger, Baker
http://petef.port5.com  (new)    -./\.-   Hughes or their divisions.

Hi,

> Does dired (C-x d) work?

I have not tested much, because I do not normally use Dired, but
displaying a directory and opening files in it works.

Mike

"Michael Forster" <forster@...> writes:

> I am using WinXP, XEmacs native, no Cygwin, SSH 2 (binary from
> http://www.ssh.com) and got into the same problems as Pete. Using the following
> patch Tramp works for me:

Hm.  Maybe tramp-handle-expand-file-name should do the work itself for
the remote file name, rather than calling on the standard
expand-file-name.

Hm.

What does (file-name-absolute-p "/foo/bar") return on your XEmacs?  I
wonder if I can use it.

Also, what does (file-name-as-directory "foo/bar") return on your
XEmacs?  I'm afraid it might append a backslash rather than a slash :-(

And finally, is there a variable like directory-sep-char in Emacs?
Does it work to do

    (let ((directory-sep-char ?/))
      (expand-file-name "/foo/bar/./baz"))

rather than calling it directly?

kai
-- 
Simplification good!  Oversimplification bad!  (Larry Wall)

In principle, I like your patch, but what happens for people using
"\" in their filenames on Unix?

Could you look at the value of directory-sep-char on your XEmacs?  If
it is ?\\, then all is proper and we could just replace
directory-sep-char with "/" always.  Right?

kai
-- 
Simplification good!  Oversimplification bad!  (Larry Wall)

Hi,

Kai wrote:
> In principle, I like your patch, but what happens for people 
> using "\" in their filenames on Unix?

I'm well aware of this, but this file names will probably never work
with the windows native expand-file-name.

> Hm.  Maybe tramp-handle-expand-file-name should do the work 
> itself for the remote file name, rather than calling on the 
> standard expand-file-name.

That is probably the clean way to do this.

> Could you look at the value of directory-sep-char on your 
> XEmacs?  

==> ?\\

> If it is ?\\, then all is proper and we could just 
> replace directory-sep-char with "/" always.  Right?

Yes. Using this:

--- tramp.el	2002/01/06 14:02:38	2.68
+++ tramp.el	2002/01/10 07:51:37
@@ -2315,8 +2315,9 @@
 	;; expand-file-name (this does "/./" and "/../").
 	(tramp-make-tramp-file-name
 	 multi-method method user host
-	 (tramp-drop-volume-letter
-	  (tramp-run-real-handler 'expand-file-name (list path))))))))
+	 (let ((directory-sep-char ?/))
+	  (tramp-drop-volume-letter
+	   (tramp-run-real-handler 'expand-file-name (list path)))))))))
 
 ;; Remote commands.
 
instead of my original patch works. Please note that this does not solve
the above problem.

> What does (file-name-absolute-p "/foo/bar") return on your 
> XEmacs?  I wonder if I can use it.

==> t

> Also, what does (file-name-as-directory "foo/bar") return on 
> your XEmacs?  I'm afraid it might append a backslash rather 
> than a slash :-(

==> "foo/bar\\"

> And finally, is there a variable like directory-sep-char in 
> Emacs? Does it work to do
> 
>     (let ((directory-sep-char ?/))
>       (expand-file-name "/foo/bar/./baz"))
> 
> rather than calling it directly?

(let ((directory-sep-char ?/))
  (expand-file-name "/foo/bar/./baz"))

==> "//dionysos/br.lehre/foo/bar/baz"

(expand-file-name "/foo/bar/./baz")

==> "\\\\dionysos\\br.lehre\\foo\\bar\\baz"

(\\dionysos\br.lehre is the current directory. In case you do not know
the windows UNC path syntax: this does mean: host "dionysos", share
"br.lehre", subdirectory foo/bar/baz)


Mike

"Michael Forster" <forster@...> writes:

> Kai wrote:
>> In principle, I like your patch, but what happens for people 
>> using "\" in their filenames on Unix?
>
> I'm well aware of this, but this file names will probably never work
> with the windows native expand-file-name.
>
>> Hm.  Maybe tramp-handle-expand-file-name should do the work 
>> itself for the remote file name, rather than calling on the 
>> standard expand-file-name.
>
> That is probably the clean way to do this.
>
>> Could you look at the value of directory-sep-char on your 
>> XEmacs?  
>
> ==> ?\\
>
>> If it is ?\\, then all is proper and we could just 
>> replace directory-sep-char with "/" always.  Right?
>
> Yes. Using this:
>
> --- tramp.el	2002/01/06 14:02:38	2.68
> +++ tramp.el	2002/01/10 07:51:37
> @@ -2315,8 +2315,9 @@
>  	;; expand-file-name (this does "/./" and "/../").
>  	(tramp-make-tramp-file-name
>  	 multi-method method user host
> -	 (tramp-drop-volume-letter
> -	  (tramp-run-real-handler 'expand-file-name (list path))))))))
> +	 (let ((directory-sep-char ?/))
> +	  (tramp-drop-volume-letter
> +	   (tramp-run-real-handler 'expand-file-name (list path)))))))))
>  
>  ;; Remote commands.
>  
> instead of my original patch works. Please note that this does not solve
> the above problem.

Ah.  This seems to be safe enough, so I'll do that.  [time passes]
Done.  Will commit in a jiffy.

>> What does (file-name-absolute-p "/foo/bar") return on your 
>> XEmacs?  I wonder if I can use it.
>
> ==> t

Looks good.

>> Also, what does (file-name-as-directory "foo/bar") return on 
>> your XEmacs?  I'm afraid it might append a backslash rather 
>> than a slash :-(
>
> ==> "foo/bar\\"

Ah.  So does (let ((directory-sep-char ?/)) (file-name-as-directory
"foo/bar")) do the right thing then?

>> And finally, is there a variable like directory-sep-char in 
>> Emacs? Does it work to do
>> 
>>     (let ((directory-sep-char ?/))
>>       (expand-file-name "/foo/bar/./baz"))
>> 
>> rather than calling it directly?
>
> (let ((directory-sep-char ?/))
>   (expand-file-name "/foo/bar/./baz"))
>
> ==> "//dionysos/br.lehre/foo/bar/baz"

Ick.  So it's interpreting the absolute(!) file name "/foo/bar/./baz"
as a relative file name?!?!  How can we circumvent that?

> (expand-file-name "/foo/bar/./baz")
>
> ==> "\\\\dionysos\\br.lehre\\foo\\bar\\baz"

As above.

> (\\dionysos\br.lehre is the current directory. In case you do not know
> the windows UNC path syntax: this does mean: host "dionysos", share
> "br.lehre", subdirectory foo/bar/baz)

Hm.  But it's still completely unclear to me why does
expand-file-name consider the directory as relative :-(

kai
-- 
Simplification good!  Oversimplification bad!  (Larry Wall)

"Michael Forster" <forster@...> writes:

> Kai wrote:
>> In principle, I like your patch, but what happens for people 
>> using "\" in their filenames on Unix?
>
> I'm well aware of this, but this file names will probably never work
> with the windows native expand-file-name.

Please don't forget that Tramp is also used on Unix!  Adding your
code means it breaks on Unix.  At the _very_ least, the execution of
the function should conditionalize on system-type or suchlike.

Suppose I say (expand-file-name "foo\\bar") on my Unix system with
your patch -- wrong result, right?

kai
-- 
Simplification good!  Oversimplification bad!  (Larry Wall)

Hi,

Kai wrote:
> >> Also, what does (file-name-as-directory "foo/bar") return on
> >> your XEmacs?  I'm afraid it might append a backslash rather 
> >> than a slash :-(
> >
> > ==> "foo/bar\\"
> 
> Ah.  So does (let ((directory-sep-char ?/)) (file-name-as-directory
> "foo/bar")) do the right thing then?

Yes, it does:

(let ((directory-sep-char ?/))
  (file-name-as-directory "foo/bar"))

==> "foo/bar/"

> >> And finally, is there a variable like directory-sep-char in
> >> Emacs? Does it work to do
> >> 
> >>     (let ((directory-sep-char ?/))
> >>       (expand-file-name "/foo/bar/./baz"))
> >> 
> >> rather than calling it directly?
> >
> > (let ((directory-sep-char ?/))
> >   (expand-file-name "/foo/bar/./baz"))
> >
> > ==> "//dionysos/br.lehre/foo/bar/baz"
> 
> Ick.  So it's interpreting the absolute(!) file name 
> "/foo/bar/./baz" as a relative file name?!?!  How can we 
> circumvent that?

After some investigation I must correct my statement: Xemacs does not
prepend the current directory, but the current drive, which is correct.
On Windows /foo/bar means an absolute Path _on the current drive_, and
"\\dionysos\br.lehre" is considered as a drive (it is a share on a file
server, which is roughly equivalent). If I am working on a local file,
expand-file-name returns "C:/foo/bar/baz".

Mike

"Michael Forster" <forster@...> writes:

> Hi,
>
> Kai wrote:
>> >> Also, what does (file-name-as-directory "foo/bar") return on
>> >> your XEmacs?  I'm afraid it might append a backslash rather 
>> >> than a slash :-(
>> >
>> > ==> "foo/bar\\"
>> 
>> Ah.  So does (let ((directory-sep-char ?/)) (file-name-as-directory
>> "foo/bar")) do the right thing then?
>
> Yes, it does:
>
> (let ((directory-sep-char ?/))
>   (file-name-as-directory "foo/bar"))
>
> ==> "foo/bar/"

I've made that change.

Could you please look whether tramp-handle-directory-files needs the
same change?

>> >> And finally, is there a variable like directory-sep-char in
>> >> Emacs? Does it work to do
>> >> 
>> >>     (let ((directory-sep-char ?/))
>> >>       (expand-file-name "/foo/bar/./baz"))
>> >> 
>> >> rather than calling it directly?
>> >
>> > (let ((directory-sep-char ?/))
>> >   (expand-file-name "/foo/bar/./baz"))
>> >
>> > ==> "//dionysos/br.lehre/foo/bar/baz"
>> 
>> Ick.  So it's interpreting the absolute(!) file name 
>> "/foo/bar/./baz" as a relative file name?!?!  How can we 
>> circumvent that?
>
> After some investigation I must correct my statement: Xemacs does not
> prepend the current directory, but the current drive, which is correct.
> On Windows /foo/bar means an absolute Path _on the current drive_, and
> "\\dionysos\br.lehre" is considered as a drive (it is a share on a file
> server, which is roughly equivalent). If I am working on a local file,
> expand-file-name returns "C:/foo/bar/baz".

So, how can we achieve that (expand-file-name "/foo/bar/./baz")
produces "/foo/bar/baz" on your system?  I'm sure it must be
possible, but I don't have XEmacs on Windows to try.

kai
-- 
Simplification good!  Oversimplification bad!  (Larry Wall)