A Lisp server executing software on demand

Alessandro Serra db00a63fcc add asd installer for the client 5 gadi atpakaļ
archlinux 5c11e3680b add archlinux package 5 gadi atpakaļ
client db00a63fcc add asd installer for the client 5 gadi atpakaļ
dot.lserver 8832654aa1 create default dot.lserver directory 5 gadi atpakaļ
roswell 6fa1fa8343 add roswell script 5 gadi atpakaļ
server ec4a0ccff4 Fixed an off-by-one error 6 gadi atpakaļ
shell-completion d0e227a260 add bash completion 5 gadi atpakaļ
.gitignore 2fa97e40bf .gitignore added 7 gadi atpakaļ
Makefile 27444d73e4 General Makefile 6 gadi atpakaļ
README.md 383ccc318d Link to the SBCL site 6 gadi atpakaļ
quick-setup.sh 8832654aa1 create default dot.lserver directory 5 gadi atpakaļ

README.md

lserver

Using a standalone server executable

Build prerequisites

Building and installing

  1. Download the tarball and unpack it under ~/quicklisp/local-projects.

    $ cd ~/quicklisp/local-projects; wget -O lserver.tar.gz https://notabug.org/quasus/lserver/archive/master.tar.gz && tar xvf lserver.tar.gz && cd lserver
    
  2. As you are in the lserver directory, build the server and the client.

    $ make
    
  3. Install the server and client binaries. You can specify a location by means of prefix, e. g.

    $ make install prefix="$HOME/usr"
    

will install both under ~/usr/bin/.

Running

By default, the home directory of the server is $HOME/.lserver. This can be overridden by means of the LSERVER_HOME environment variable.

The LSERVER_SOCKET variable overrides the socket name, which is default by default.

Using from Portacle

Instead of building a standalone executable, you can install the portable Common Lisp development environment Portacle and run the server from there. This way you let Portacle take care of SBCL and quicklisp and moreover, you get live introspection and hacking with Slime.

In this case you don’t need to build the server.

Installation

  1. Install Portacle.

  2. cd into the Portacle directory.

  3. Download the tarball and unpack it under quicklisp/local-projects:

    $ cd all/quicklisp/local-projects; wget -O lserver.tar.gz https://notabug.org/quasus/lserver/archive/master.tar.gz && tar xvf lserver.tar.gz && cd lserver
    
  4. Build and install the client:

    $ cd client
    $ make
    $ make install prefix="$HOME/usr"
    

Of course, you can use a different prefix or no prefix altogether if you install it under /usr/local.

Running

To run the server, start Portacle, load lserver by writing

    CL-USER> (ql:quickload "lserver")

at the prompt. When you do this for the first time, quicklisp will fetch dependencies, so you will need an internet connection. Then run the server by calling

    CL-USER> (lserver:run-server :background t)

You can specify an alternative home directory and/or socket name like this:

    CL-USER> (lserver:run-server :background t :home "~/alternative-home" :socket "mysock")

Using from a Lisp environment

The system definition is found in the server directory. You build the client and run the server as explained in the section Using from Portacle.

Client

    $ lcli --list-commands
    $ lcli [command] [command-arguments]
  • command is the name of the piece of software to run.
  • command-arguments are CLI options for the command.

Configuring

No commands are defined out of the box, so the server must be configured in order to be of any use. You do that by editing the lserverrc.lisp file residing in the server home directory.

You can run the quick-setup.sh script in order to obtain a simple file-based command system described below together with the eval and say commands. This doesn’t mean that this is the preferable way of managing commands, that the eval and say commands must be generally available or that lserver::file-command is an ‘official’ way of distributing software.

The lserverrc.lisp file is a regular Lisp file loaded when the server starts. Single-line comments start with a semicolon and multi-line comments go between #| and |#.

You can use the add-command function to directly define commands in the configuration file. Here is a small configuration:

(in-package #:lserver)

(add-command "eval" (lambda (args)
                      (dolist (arg args t)
                        (eval (with-standard-io-syntax
                                (let ((*package* (find-package "LSERVER")))
                                  (read-from-string arg))))))
             "Evaluate the arguments as Lisp expressions")

(add-command "say" (lambda (args)
                     (loop for value in (multiple-value-list (eval (with-standard-io-syntax
                                                                     (let ((*package* (find-package "LSERVER")))
                                                                       (read-from-string (first args))))))
                           do (format t "~A~%" value)
                           finally (finish-output) (return t))))

The first line specifies the namespace containing lserver-specific tools. Put it at the top of the configuration file unless you know what you are doing. The add-command function accepts a command name, a main-type function accepting a list of command-line options, and an optional description. The commands defined in the snippet evaluate arbitrary Lisp expressions and optionally print the results, e. g.

$ lcli eval '(write-line "Hello, world!")'
Hello, world!
$ lcli say '(+ 1 2)'
3

Of course, you don’t want that if lserver is to be used as a system-wide daemon.

If you want to define commands in individual files, you can use the following setup:

(defun file-command (function &optional description)
  (add-command (pathname-name *load-truename*) function description))

(defvar *path* (list (merge-pathnames #p"commands/" (lserver-homedir-pathname))))

(defun commands-from-path ()
  (dolist (directory *path*)
    (dolist (file (uiop:directory-files directory "*.lisp"))
      (let ((*package* (find-package "LSERVER")))
        (load file)))))

(commands-from-path)

Instead of defining the eval and say commands directly in the configuration file, you can create the following files eval.lisp and say.lisp under the commands subdirectory:

;;;; eval.lisp

(in-package #:lserver)

(file-command (lambda (args)
                (dolist (arg args t)
                  (eval (with-standard-io-syntax
                          (let ((*package* (find-package "LSERVER")))
                            (read-from-string arg))))))
              "Evaluate arguments as Lisp forms.")


;;;; say.lisp

(in-package #:lserver)

(file-command (lambda (args)
                (loop for value in (multiple-value-list (eval (with-standard-io-syntax
                                                                (let ((*package* (find-package "LSERVER")))
                                                                  (read-from-string (first args))))))
                      do (format t "~A~%" value)
                      finally (finish-output) (return t)))
              "Evaluate the argument as a Lisp form and print the values on separate lines.")

Observe that file-command names commands after files, so you can easily rename commands by just renaming files.

If you add new command files while the server is running, you can run

$ lcli eval '(commands-from-path)'

for a live update.

Writing software

lserver commands can be defined on top of lserver-unaware functions. When the command is executed, the associated function is applied to the list of provided command arguments with the following dynamic bindings:

  • *standard-input*, *standard-output*, and *standard-error* are identified with client’s standard streams;
  • *query-io* corresponds to stdin and stderr;
  • *default-pathname-defaults* is client’s working directory,
  • *package* is the LSERVER package.

The exit code is determined by the returned value: if the function returns an integer, it is used as the exit code, if the function returns a non-integer true value, the exit code is 0 and 1 otherwise. If something goes wrong, the client dies with a nonzero exit code, so such codes are not quite reliable.