I got TRAMP 2.0.25 to work under NT Emacs 21.2.1 using the latest
installation of the Cygwin tools (1.3.12-2) using Cygwin's ssh, scp, and
ssh-agent. The remote host was running Debian GNU/Linux. I haven't
been brave enough to try a Cygwin-to-Cygwin configuration yet -- I think
Cygwin-to-UNIX is more common anyway. Also, I didn't try getting the
"telnet" method to work, on the theory that using scp is more common.
Here's what I had to do to make it work:
1. In my .emacs startup file, I do:
(require 'tramp)
(setq tramp-default-method "scpx")
This is required when using TRAMP under NT Emacs as described in the
manual.
to load up TRAMP. I have load-path set appropriately.
2. Next, my .emacs file executes this horror:
(let ((shell-file-name "c:/apps/cygwin/bin/bash.exe"))
(shell-command "echo /tmp/ssh-*/agent.* 2>/dev/null" " *temp*"))
(save-excursion
(set-buffer " *temp*")
(end-of-line)
(if (> (buffer-size) 5)
(setenv "SSH_AUTH_SOCK" (buffer-substring (point-min) (point)))
(ding)
(message "my-general.el: WARNING: ssh-agent is not running!")
(sit-for 3)))
The above mess is to cope with what I think is a fairly common
situation -- namely, where the user launches NT Emacs using a
shortcut on the desktop but starts ssh-agent from a bash shell also
launched from the desktop. Since Emacs is not a descendant of
ssh-agent, it does not inherit the SSH_AUTH_SOCK environment
variable (and neither does scp).
My solution (above) is a total hack. It malfunctions if the echo
command outputs more than one pathname, which happens if you have old
ssh-agent socket directories lying around. If TRAMP had were to call
a hook at just the right moment, I could write some Elisp to find the
most recent ssh-agent socket file and setenv the SSH_AUTH_SOCK
environment variable within Emacs.
Maybe the general solution is to create a tramp-methods key named
tramp-pre-rcp-hook, which is run just before tramp-rcp-program
executes?
3. Next, I had to make the following change to lisp/tramp.el.
***************
*** 3113,3118 ****
--- 3113,3119 ----
"Like `insert-file-contents' for tramp files."
(barf-if-buffer-read-only)
(setq filename (expand-file-name filename))
+ (setq buffer-file-name filename)
(with-parsed-tramp-file-name filename nil
(when (tramp-ange-ftp-file-name-p multi-method method user host)
(tramp-invoke-ange-ftp 'insert-file-contents
This may not be the right thing to do, but it prevents a
wrong-type-argument error from vc-file-not-found-hook and from
find-file-noselect-1 when they each evaluate (file-truename
buffer-file-name) with buffer-file-name set to nil.
4. Next, I had to make the following change to lisp/tramp.el.
*** 3024,3030 ****
(unless (file-exists-p filename)
(error "Cannot make local copy of non-existing file `%s'"
filename))
! (setq tmpfil (tramp-make-temp-file))
(cond (rcp-program
;; Use rcp-like program for file transfer.
(tramp-message-for-buffer
--- 3024,3030 ----
(unless (file-exists-p filename)
(error "Cannot make local copy of non-existing file `%s'"
filename))
! (setq tmpfil (substring (tramp-make-temp-file) 2))
(cond (rcp-program
;; Use rcp-like program for file transfer.
(tramp-message-for-buffer
This is to cope with the fact that, by default, NT Emacs has
temporary-file-directory set to C:/DOCUME~1/username/LOCALS~1/Temp/,
which is unacceptable to Cygwin's scp as the second argument, because
scp interprets the "c:" as a hostname! The substring call chops off
the "C:", leaving tmpfil set to a value that is acceptable to both
Cygwin's scp and NT Emacs (and, more importantly, means the same
thing to both of them).
This change doesn't pass the "smell test" to me, but it probably
works 99% of the time. I haven't thought of a more general solution
yet.
I hope this helps.
--
Francis Litterio
[EMAIL PROTECTED]
http://world.std.com/~franl/
GPG and PGP public keys available on keyservers.
_______________________________________________
Tramp-devel mailing list
[EMAIL PROTECTED]
http://mail.freesoftware.fsf.org/mailman/listinfo/tramp-devel
