inferior-shell

2016-09-29

INFERIOR-SHELL: spawning shell commands and pipes from Common Lisp

What is inferior-shell?

This CL library allows you to spawn local or remote processes and shell pipes. It lets me use CL in many cases where I would previously write shell scripts. The name is a pun, in that this library can both let you spawn inferior (children processes) shells, and serve itself as an inferior (not so featureful) shell. Because so many features of a shell are missing, inferior-shell only brings down the low-hanging fruits of shell scripting; yet CL is such a better programming language than the shell (or other "scripting" languages) that it is already a great pleasure to be able to write things in CL rather than in these languages. More features will come, and/or you can use other CL libraries as a complement.

Inferior-shell recognizes a small domain-specific language to describe commands or pipelines of commands, and some functions to actually run these pipelines either locally or remotely (via ssh). It will implicitly invoke ssh when asked to run a command on a remote host; for best results, be sure to have passphrase-less access to these hosts via e.g. ssh-agent.

The name inferior-shell was suggested by Michael Livshin, as inspired by the equivalent notion in GNU Emacs.

Example use of inferior-shell, from the rpm system:

(defun rpms-installed (&key (packagenames t) host)
  (run/lines
   `(pipe (rpm -qa)
          ,@(unless (eq packagenames t)
              `((egrep ("^(" ,@(loop :for (name . more) :on packagenames
                                 :collect name :when more :collect "|")
                             ")-[^-]+(-[^-]+)?$")))))
   :host host))

Limitations

By default, inferior-shell uses uiop:run-program as its universal execution backend, and has its limitations, which are as follows.

First, inferior-shell at this point only supports synchronous execution of sub-processes. For asynchronous execution, please use IOlib or executor. IOlib requires C compilation and linking, and may or may not support Windows. executor only supports select implementations. A future extension to inferior-shell may use IOlib as a backend.

Second, supported platforms at this time include: ABCL, Allegro, CLISP, ClozureCL, CMUCL, ECL, LispWorks, RMCL, SBCL, SCL, XCL. Platforms NOT (yet) supported include: CormanLisp (untested), GCL (untested), Genera (unimplemented), MKCL (untested). On supported platforms, inferior-shell works on both Unix and Windows.

Exported Functionality

The inferior-shell library creates a package INFERIOR-SHELL, that exports the following macros and functions:

  • PARSE-PROCESS-SPEC SPEC

parse an expression in the process-spec mini-language into objects specifying a pipeline of processes to be executed. See the PROCESS-SPEC mini-language below.

  • PRINT-PROCESS-SPEC SPEC &OPTIONAL OUTPUT

print a process specification to given OUTPUT into a portable form usable by a Unix shell. OUTPUT is as per FORMAT's stream output argument, defaults to NIL for returning the result as a string. SPEC can be a parsed PROCESS-SPEC object, a CONS to be parsed by PARSE-PROCESS-SPEC, or a string for a process-spec that has already been formatted.

  • *CURRENT-HOST-NAMES*

a variable, a list of strings, the aliases for the localhost. You may need to initialize it if the defaults don't suffice.

  • CURRENT-HOST-NAME-P X

a function, returns true if X is a string member of CURRENT-HOST-NAMES

  • INITIALIZE-CURRENT-HOST-NAMES

function that initializes the CURRENT-HOST-NAMES with "localhost" and the results from $(hostname -s) and $(hostname -f).

  • RUN CMD &KEY ON-ERROR TIME SHOW HOST OUTPUT

RUN will execute the given command CMD, which can be a CONS to be parsed by PARSE-PROCESS-SPEC, a PROCESS-SPEC object already parsed, or a string to be passed to a Unix shell. ON-ERROR specifies behavior in case the command doesn't successfully exit with exit code 0. NIL means continue and return the error code; T (default) means signal the error (same as :ON-ERROR 'SIGNAL); any other value is called as by UIOP:CALL-FUNCTION with the condition as argument, and if it returns, its return values are returned by RUN. TIME is a boolean which if true causes the execution to be timed as per TIME. SHOW is a boolean which if true causes a message to be sent to the *TRACE-OUTPUT* before execution. HOST is either NIL (execute on localhost) or a string specifying a host on which to run the command using ssh if it's not an alias for localhost as recognized by CURRENT-HOST-NAME-P (be sure to have passphraseless login using ssh-agent). The INPUT, OUTPUT and ERROR-OUTPUT arguments are as for UIOP:RUN-PROGRAM, except that they default to NIL, T, T respectively instead of NIL, NIL, NIL. In particular, OUTPUT is as per UIOP:SLURP-OUTPUT-STREAM one of NIL for no output (redirect to /dev/null), a stream for itself, T (default) for the current *standard-output*, :INTERACTIVE for inheriting the parent process's stdout, :LINES for returning one result per line, :STRING for returning the output as one big string, :STRING/STRIPPED is like :STRING but strips any line-ending at the end of the results, just like a shell's cmd or $(cmd) would, and more options are accepted and you can define your own, as per uiop's slurp-input-stream protocol. On Windows, RUN will not succeed for pipes, only for simple commands. On Unix, simple commands on localhost are executed directly, but remote commands and pipes are executed by spawning a shell.

  • RUN/NIL CMD &KEY ON-ERROR TIME SHOW HOST

RUN/NIL is a shorthand for RUN with :INPUT :OUTPUT :ERROR-OUTPUT bound to NIL.

  • RUN/S CMD &KEY ON-ERROR TIME SHOW HOST

RUN/S is a shorthand for RUN with :OUTPUT bound to :STRING, returning as a string what the inferior command sent to its standard output.

  • RUN/SS CMD &KEY ON-ERROR TIME SHOW HOST

RUN/S is a shorthand for RUN :OUTPUT :STRING/STRIPPED, just like a shell's cmd or $(cmd) would do.

  • RUN/INTERACTIVE CMD &KEY ON-ERROR TIME SHOW HOST

RUN/INTERACTIVE is a shorthand for RUN with :INPUT :OUTPUT :ERROR-OUTPUT all bound to :INTERACTIVE, so you may run commands that interact with users, inheritting the stdin, stdout and stderr of the current process.

  • RUN/I CMD &KEY KEYS

RUN/I is a shorthand alias for RUN/INTERACTIVE

  • RUN/LINES CMD &KEY ON-ERROR TIME SHOW HOST

RUN/LINES is a shorthand for RUN :OUTPUT :LINES, returning as a list of one string per line (stripped of line-ending) what the inferior command sent to its standard output.

The PROCESS-SPEC mini-language

This library offers a SEXP syntax to specify processes and pipelines of processes in the manner of Unix shells, including support for file descriptor redirection. Process specifications can be printed, to be executed by a local or remote shell, or directly executed by your Lisp implementation, depending on its capabilities and on the complexity of the pipeline.

SEXP mini-language

;; A process is a pipe or a command
process := pipe | or | and | progn | fork | command

;; A pipe is a list of processes, each of whose output is connected to the next one's input.
pipe := ( pipe process* )

;; OR is a list of processes which will be executed in sequence until one returns exit code 0.
or := ( or processes )

;; AND is a list of processes which will be executed in sequence until one does not return exit code 0.
and := ( and processes )

;; PROGN is a list of processes which will be executed sequentially.
progn := ( progn processes )

;; FORK is a list of processes which will be forked and executed in parallel.
fork := ( fork processes )

;; A command is a list of tokens and redirections.
;; Tokens specify the argv, redirections specify modifications of the inherited file descriptors.
command := ( [redirection|token|tokens]* )

;; A token is a string, to be used literally,
;; a keyword, to be downcased and prefixed with -- as in :foo ==> "--foo"
;; a symbol, to be downcased, or a list of tokens to be concatenated.
token := string | keyword | symbol | (token*)

;; A list starting with * is actually to be spliced in the token stream.
tokens := (\* [token|tokens]*)

;; Redirections mimic those redirections available to a shell, for instance zsh.
redirection := (
 ! fd pathname flags |   ;; open a file with given flags redirect to specified fd
 < fd? pathname | ;; open a file for input, redirect to specified fd (default: 0)
 [>|>>|<>|>!|>>!] fd? pathname | ;; open a file for (respectively) output, append, io, output clobbering, append clobbering, redirect to specified fd (default: 1)
 - fd | <& fd - | >& fd - | ;; close a fd
 <& - | >& - | ;; close fd 0, respectively fd 1.
 <& fd fd | >& fd fd | ;; redirect fds: the left one is the new number, the right one the old number.
 >& pn | >&! | ;; redirect both fd 1 and 2 to pathname (respectively, clobbering)
 >>& pn | >>&! ) ;; redirect both fd 1 and 2 to append to pathname (respectively, clobbering)

Note that these are all exported symbols from the INFERIOR-SHELL package, except that a few of them are also inherited from COMMON-LISP: < > - Therefore the other ones will only work if you either use the INFERIOR-SHELL package or use a package prefix INFERIOR-SHELL: where appropriate.

A SEXP in the minilanguage can be parsed with parse-process-spec, into an object of class process-spec. print-process-spec will print a process-spec object; in this context, a string represents itself (assuming it's already a printed process spec), and a cons is a specification in the minilanguage to be parsed with parse-process-spec first.

TO DO

Better document it.

Have a complementary inferior-shell-watcher library that uses iolib to spawn pipes locally, and watch the subprocesses as part of the iolib event loop.

Categories
library