Plasma GitLab Archive
Projects Blog Knowledge


The shell library: Pipelines and redirections

The shell library provides some of the functionality of a Unix shell; in particular the library supports starting simple commands, executing pipelined commands, and arranging redirections.

Module Shell: The comfortable layer

This module is designed for the average user who does not know very much about the Unix process model, and just wants to start external commands. A simple command can be executed by a statement like

 call [ cmd "ls" [ "/dir1"; "/dir2" ]] 

This statement searches "ls" in the current search PATH, launches the new process, and passes the arguments "/dir1" and /dir2" to the process. Nothing special is done with the file descriptors; the new process shares stdin, stdout and stderr with the current process (all other descriptors are automatically closed).

A pipeline can be constructed by call as well. For example:

 call [ cmd "find" [ "/dir" ]; cmd "grep" [ "something" ] ] 

The output of "find" is redirected to the input of "grep".

You can redirect stdin, stdout, stderr (and every other descriptor) by assigning them to other descriptors, or by opening and reading from or writing to them. In the latter case, multiple descriptors can be served parallely. For example:

let s = "d\na\nc\nb\n" in
let b = Buffer.create 20 in
call ~stdin:(from_string s) ~stdout:(to_buffer b) [ cmd "sort" [] ]

Here, "sort" reads the contents of s and writes the result to b.

Unlike the Unix shell, this module reports errors from all components of a pipeline. For example:

call [ cmd "cat" [ "/notfound" ]; cmd "ls" [ "/notfound.too" ] ]

This will raise an exception

Subprocess_error [ "/bin/cat", Unix.WEXITED 1; "/bin/ls", Unix.WEXITED 1 ]

There is another subtle difference to many Unix shells (and normally also the system function in libc). This module reports errors occuring between fork and exec; for instance if the file "fail" refers to a non-existing interpreter

#! /not/found

but is executable, this special error can only be detected by the "exec" call. Unix shells print an error message to stderr, and return an exit code of 127 (which is reserved for this case):

Sys.command "fail";;
sh: ./fail: No such file or directory
~ : int = 127

However, the true reason isn't reported. In contrast to this, the Shell module is able to pass the real error condition back to the calling program:

call [ command "fail" ];;
Uncaught exception: Unix.Unix_error(20, "execve", "./fail").

Module Shell_sys: The fundamental layer

The module Shell is a simple application of the functions defined in Shell_sys, the more fundamental module. Shell_sys allows a more fine-grained control of the execution of external commands; however, it is more difficult to use.

Shell_sys allows it to run processes both synchronously and asynchronously, to form pipelines with arbitrary topology, to create user-defined handlers serving file descriptors, and to control signal handling.

Module Unix_exts: The missing system calls

This module contains some system calls missing in the Unix library distributed with O'Caml.

This web site is published by Informatikbüro Gerd Stolpmann
Powered by Caml