5.6 Integration with other Emacs packages

TRAMP supports starting new running processes on the remote host for discovering remote file names. Emacs packages on the remote host need no specific modifications for TRAMP’s use.

This type of integration does not work with the ftp method, and does not support the pty association as specified in start-file-process.

process-file and start-file-process work on the remote host when the variable default-directory is remote:

(let ((default-directory "/ssh:remote.host:"))
  (start-file-process "grep" (get-buffer-create "*grep*")
                      "/bin/sh" "-c" "grep -e tramp *"))

For a local process, process-file returns either the exit code of the process, or a string describing a signal, when the process has been interrupted. Since it cannot be determined reliably whether a remote process has been interrupted, process-file will always returns the exit code for it. When the user option process-file-return-signal-string is non-nil, process-file treats all exit codes greater than 128 as an indication that the process has been interrupted, and returns a corresponding string.

This remote process handling does not apply to GVFS (see GVFS-based external methods) because the remote file system is mounted on the local host and TRAMP accesses it by changing the default-directory.

TRAMP starts a remote process when a command is executed in a remote file or directory buffer. As of now, these packages have been integrated to work with TRAMP: shell.el, eshell.el, compile.el (commands like compile and grep) and gud.el (gdb or perldb).

TRAMP always modifies the INSIDE_EMACS environment variable for remote processes. By default, this environment variable shows the Emacs version. TRAMP adds its own version string, so it looks like ‘27.2,tramp:’. However, other packages might also add their name to this environment variable, like ‘27.2,comint,tramp:’.

For TRAMP to find the command on the remote, it must be accessible through the default search path as setup by TRAMP upon first connection. Alternatively, use an absolute path or extend tramp-remote-path (see How TRAMP finds and uses programs on the remote host):

(add-to-list 'tramp-remote-path "~/bin")
(add-to-list 'tramp-remote-path "/appli/pub/bin")

Customize user option tramp-remote-process-environment to suit the remote program’s environment for the remote host. tramp-remote-process-environment is a list of strings structured similar to process-environment, where each element is a string of the form ‘ENVVARNAME=VALUE’.

To avoid any conflicts with local host environment variables set through local configuration files, such as ~/.profile, use ‘ENVVARNAME=’ to unset them for the remote environment.

Use add-to-list to add entries:

(add-to-list 'tramp-remote-process-environment "JAVA_HOME=/opt/java")

Modifying or deleting already existing values in the tramp-remote-process-environment list may not be feasible on restricted remote hosts. For example, some system administrators disallow changing HISTORY environment variable. To accommodate such restrictions when using TRAMP, fix the tramp-remote-process-environment by the following code in the local .emacs file:

(let ((process-environment tramp-remote-process-environment))
  (setenv "HISTORY" nil)
  (setq tramp-remote-process-environment process-environment))

Setting the ENV environment variable instructs some shells to read an initialization file. By default, TRAMP disables this. You can override this behavior by evaluating

(let ((process-environment tramp-remote-process-environment))
  (setenv "ENV" "$HOME/.profile")
  (setq tramp-remote-process-environment process-environment))

In addition to tramp-remote-process-environment, you can set environment variables for individual remote process calls by let-binding process-environment. TRAMP applies any entries not present in the global default value of process-environment (overriding tramp-remote-process-environment settings, if they conflict). For example:

(let ((process-environment (cons "HGPLAIN=1" process-environment)))
  (process-file …))

Let-binding in this way works regardless of whether the process to be called is local or remote, since TRAMP would add just the HGPLAIN setting and local processes would take whole value of process-environment along with the new value of HGPLAIN.

For integrating other Emacs packages so TRAMP can execute remotely, please file a bug report. See Reporting Bugs and Problems.

5.6.1 Running remote programs that create local X11 windows

To allow a remote program to create an X11 window on the local host, set the DISPLAY environment variable for the remote host as follows in the local .emacs file:

(add-to-list 'tramp-remote-process-environment
             (format "DISPLAY=%s" (getenv "DISPLAY")))

(getenv "DISPLAY") should return a recognizable name for the local host that the remote host can redirect X11 window interactions. If querying for a recognizable name is not possible for whatever reason, then replace (getenv "DISPLAY") with a hard-coded, fixed name. Note that using :0 for X11 display name here will not work as expected.

An alternate approach is specify ForwardX11 yes or ForwardX11Trusted yes in ~/.ssh/config on the local host.

5.6.2 Running shell on a remote host

Set explicit-shell-file-name to the appropriate shell name when using TRAMP between two hosts with different operating systems, such as ‘windows-nt’ and ‘gnu/linux’. This option ensures the correct name of the remote shell program.

When explicit-shell-file-name is equal to nil, calling shell interactively will prompt for a shell name.

You could use connection-local variables for setting different values of explicit-shell-file-name for different remote hosts.

  '((explicit-shell-file-name . "/bin/bash")
    (explicit-bash-args . ("-i"))))

  '((explicit-shell-file-name . "/bin/ksh")
    (explicit-ksh-args . ("-i"))))

  '(:application tramp :protocol "ssh" :machine "localhost")

  `(:application tramp :protocol "sudo"
    :user "root" :machine ,(system-name))

5.6.3 Running shell-command on a remote host

shell-command executes commands synchronously or asynchronously on remote hosts and displays output in buffers on the local host. Example:

C-x C-f /sudo:: RET
M-& tail -f /var/log/syslog.log RET

tail command outputs continuously to the local buffer whose name is the value of the variable shell-command-buffer-name-async.

M-x auto-revert-tail-mode RET runs similarly showing continuous output.

shell-command uses the user option shell-file-name and the variable shell-command-switch in order to determine which shell to run. For remote hosts, their default values are /bin/sh and -c, respectively (except for the adb method, which uses /system/bin/sh). Like the variables in the previous section, these variables can be changed via connection-local variables.

If Emacs supports the user option async-shell-command-width (since Emacs 27), TRAMP cares about its value for asynchronous shell commands. It specifies the number of display columns for command output. For synchronous shell commands, a similar effect can be achieved by adding the environment variable COLUMNS to tramp-remote-process-environment.

5.6.4 Running eshell on a remote host

TRAMP is integrated into eshell.el, which enables interactive eshell sessions on remote hosts at the command prompt. You must add the module eshell-tramp to eshell-modules-list. Here’s a sample interaction after opening M-x eshell RET on a remote host:

~ $ cd /sudo::/etc RET
/sudo:root@host:/etc $ hostname RET
/sudo:root@host:/etc $ id RET
uid=0(root) gid=0(root) groups=0(root)
/sudo:root@host:/etc $ find-file shadow RET
#<buffer shadow>
/sudo:root@host:/etc $

eshell added custom su and sudo commands that set the default directory correctly for the *eshell* buffer. TRAMP silently updates tramp-default-proxies-alist with an entry for this directory (see Connecting to a remote host using multiple hops):

~ $ cd /ssh:user@remotehost:/etc RET
/ssh:user@remotehost:/etc $ find-file shadow RET
File is not readable: /ssh:user@remotehost:/etc/shadow
/ssh:user@remotehost:/etc $ sudo find-file shadow RET
#<buffer shadow>

/ssh:user@remotehost:/etc $ su - RET
/su:root@remotehost:/root $ id RET
uid=0(root) gid=0(root) groups=0(root)
/su:root@remotehost:/root $

5.6.5 Running a debugger on a remote host

gud.el provides a unified interface to symbolic debuggers. TRAMP can run debug on remote hosts by calling gdb with a remote file name:

M-x gdb RET
Run gdb (like this): gdb -i=mi /ssh:host:~/myprog RET

Since the remote gdb and gdb-inferior processes do not belong to the same process group on the remote host, there will be a warning, which can be ignored:

&"warning: GDB: Failed to set controlling terminal: Operation not permitted\n"

As consequence, there will be restrictions in I/O of the process to be debugged.

Relative file names are based on the remote default directory. When myprog.pl exists in /ssh:host:/home/user, valid calls include:

M-x perldb RET
Run perldb (like this): perl -d myprog.pl RET

Just the local part of a remote file name, such as perl -d /home/user/myprog.pl, is not possible.

Arguments of the program to be debugged must be literal, can take relative or absolute paths, but not remote paths.

5.6.6 Running remote processes on MS Windows hosts

winexe runs processes on a remote MS Windows host, and TRAMP can use it for process-file and start-file-process.

tramp-smb-winexe-program specifies the local winexe command. Powershell V2.0 on the remote host is required to run processes triggered from TRAMP.

explicit-shell-file-name and explicit-*-args have to be set properly so M-x shell RET can open a proper remote shell on a MS Windows host. To open cmd, set it as follows:

(setq explicit-shell-file-name "cmd"
      explicit-cmd-args '("/q"))

To open powershell as a remote shell, use this:

(setq explicit-shell-file-name "powershell"
      explicit-powershell-args '("-file" "-"))

5.6.7 Remote process connection type

Asynchronous processes behave differently based on whether they use a pseudo tty or not. This is controlled by the variable process-connection-type, which can be t or pty (use a pseudo tty), or nil or pipe (don’t use one). TRAMP is based on running shells on the remote host, which requires a pseudo tty. Therefore, it declares the variable tramp-process-connection-type, which carries this information for remote processes. Its default value is t, and there is no need to change it. The name of the remote pseudo tty is returned by the function process-tty-name.

If a remote process, started by start-file-process, should not use a pseudo tty, this can be requested by setting process-connection-type to nil or pipe. There is still a pseudo tty for the started process, but some terminal properties are changed, like suppressing translation of carriage return characters into newline.

The function make-process allows controlling this explicitly by using the :connection-type keyword. If this keyword is not used, the value of process-connection-type is applied instead.

5.6.8 Process properties of asynchronous remote processes

When available, TRAMP adds process properties to process objects of asynchronous properties. However, it is not guaranteed that all these properties are set.

The functions list-system-processes and process-attributes return information about system processes on the respective remote host. In order to retrieve this information, they use the command ps, driven by the following constants:

Constant: tramp-process-attributes-ps-args

This is a list of arguments (strings) ps is called with. The default value is appropriate for GNU/Linux remote hosts.

Constant: tramp-process-attributes-ps-format

This is a list of cons cells (key . type) for interpretation of the ps output. key is a key used in the process-attributes output plus the key pid, and type is the respective value returned by ps. It can be

numberp— a number
stringp— a string without spaces
number— a string of number width, could contain spaces
nil— a string until end of line

The default value is appropriate for GNU/Linux remote hosts.

If, for example, tramp-process-attributes-ps-args is declared as ("-eww" "-o" "pid,euid,euser,egid,egroup,comm:40,state"), the output of the respective ps command would look like

    PID  EUID EUSER     EGID EGROUP   COMMAND                                  S
      1     0 root         0 root     systemd                                  S
   1610     0 root         0 root     NFSv4 callback                           S

The corresponding tramp-process-attributes-ps-format has the value

((pid . numberp) (euid . numberp) (user . stringp)
 (egid . numberp) (group . stringp) (comm . 40) (state . stringp))

The default values for tramp-process-attributes-ps-args and tramp-process-attributes-ps-format can be overwritten by connection-local variables. This is already done by TRAMP for the adb method, see tramp-adb-connection-local-default-ps-profile and tramp-adb-connection-local-default-ps-variables.

There are three further predefined sets of connection-local variables for remote BSD systems, for remote macOS systems, and for a remote ps command implemented with busybox. These are called tramp-connection-local-*-ps-profile and tramp-connection-local-*-ps-variables. Use them like

 '(:application tramp :machine "mybsdhost")

If you want to see a listing of remote system processes when calling proced, set user option proced-show-remote-processes to non-nil, or invoke that command with a negative argument like C-u - M-x proced RET when your buffer has a remote default-directory.

5.6.9 Improving performance of asynchronous remote processes

TRAMP’s implementation of make-process and start-file-process requires a serious overhead for initialization, every process invocation. This is needed for handling interactive dialogs when connecting the remote host (like providing a password), and initial environment setup.

Sometimes, this is not needed. Instead of starting a remote shell and running the command afterwards, it is sufficient to run the command directly. TRAMP supports this by an alternative implementation of make-process and start-file-process. This is triggered by the connection property "direct-async-process", See Setting own connection related information, which must be set to a non-nil value. Example:

(add-to-list 'tramp-connection-properties
             (list (regexp-quote "/ssh:user@host:")
                   "direct-async-process" t))

Using direct asynchronous processes in TRAMP is not possible, if the remote host is connected via multiple hops (see Connecting to a remote host using multiple hops). In this case, TRAMP falls back to its classical implementation.

Furthermore, this approach has the following limitations:

In order to gain even more performance, it is recommended to bind tramp-verbose to 0 when running make-process or start-file-process. Furthermore, you might set tramp-use-ssh-controlmaster-options to nil in order to bypass TRAMP’s handling of the ControlMaster options, and use your own settings in ~/.ssh/config, Using ssh connection sharing.