A shell offers a command line interface for users. It interprets user commands and carries out them. Some shells offer simple scripting terms like if or while, and permit users to make a program that facilitates their computing life. There are two categories of shells: command line and graphical. Instinctively, command line shells offer command line user interface commonly used in Unix/Linux environments, while graphical shells give graphic user interface(GUI) like MS Windows. In this assignment, we will focus on command line shells.
Under the hood, a shell is just another user program. The files /bin/sh, /bin/bash, and /bin/tcsh are all executable files for shells. The only thing special about your login shell is that it is listed in your login record so that /bin/login (the program that prompts you for your password) knows what program to start when you log in. If you run ``cat /etc/passwd'', you'll see the login records of the machine and the login shell program listed as the last field.
A shell in Unix-like systems often supports two interesting features:
When you start a command, there are always three default file streams open, stdin (usually maps to input from the keyboard), stdout (usually maps to the normal output to the screen), andstderr (usually maps to error messages output to the screen). These, and any other open files, can be redirected, that is, they can be mapped to files or devices users specify. Below are some redirection instances:
$ cmd1 < in.txt
executes cmd1, using in.txt as the source of input, instead of the keyboard.
$ cmd2 > out.txt
executes cmd2 and places the output to file out.txt.
$ cmd3 >out.txt 2> err.txt
executes cmd3 and places the normal output to file out.txt and the error messages output to file err.txt. Here the 2 in 2> actually refers to the file descriptor 2.
The command below connects the standard output of cmd1 to the standard input of cmd2, and again connects the standard output of cmd2 to the standard input of cmd3, using the pipeline operator '|'
$ cmd1 | cmd2 | cmd3
An ex usage is:
$ sort < file.txt | uniq | wc
Which count the number of unique lines in file.txt? Without pipes, you would have to employ three commands and two intermediate files to count the unique lines in a file.
1. prepare down a simple shell. Basically your shell should read the line from standard input, parse the line with command and arguments, and execute the command with arguments. You'll need the fork() and exec*() system calls.
- The shell must find out the command in your current working directory first. If not found, it should search the directories in the shell's pathname list which will be describeed in the part 2.
- You are not allowed to use system(), as it just summons the system's /bin/sh shell to do all the work. You are not allowed to use execlp() or execvp() in the standard library, because our shell has its own path variable, describeed in part 2.
- You might suppose that arguments are separated by white spaces. You do not have to deal with special characters such as ', ", \, etc; however, you need to handle the redirection operators (<, >, and 2>) and the pipeline operator (|).
- You can assume that the command line a user types is no longer than 4096 bytes. However, you should not assume that there is a restriction on the number of arguments to a given command.
- The executable file for your shell should be named myshell, for the TAs' grading pleasure.
2. Implement the following built-in commands.
- exit: users can exit from the shell with the exit command.
- cd: cd is a command to change directories. You will need to invoke the chdir system call.
- path: path is a command not only to show the current pathname list, but also to append or remove several pathnames. In your shell implementation, you may keep a variable or data structure to deal with pathname list (referred to as "path" variable below). This list helps searching for executables when users enter specific commands.
- path (without arguments) displays the pathnames currently set. It should show pathnames separated by colons. ex) "/bin:/usr/bin"
- path + /foo/bar appends the pathname to the "path" variable. Only one pathname will be given to each invocation of the path command. It's okay to add a pathname to the "path" variable even if it already contains the same pathname, i.e., duplicates in the "path" variable are fine.
- path - /foo/bar removes the pathname from the "path" variable. It should remove all duplicates of the given pathname. Only one pathname will be given to each invocation of the path command.
3. (Extend your shell with I/O redirection (<, >, and 2>), as described above.
- stdin maps to file descriptor 0, stdout to 1, and stderr to 2.
- You'll need the open(), close(), and dup2() system calls.
- You can assume that there is no space inside the stderr redirection operator "2>"
- You should handle cases when all three streams are redirected, such as cmd < in.txt > out.txt 2> err.txt (not necessarily in this order).
- You don't need to handle other forms of redirections. For ex, you don't need to handle the following redirections : cmd 1> filename (another way to redirect standard output)), cmd 2>&1 (redirect standard error to standard output). (Most other shells do handle the above cases.)
- Your shell do not need to handle I/O redirection for built-in commands (cd, exit, path).
4. Extend your shell with pipeline (|), as described above.
- You'll need the pipe system call.
- There is no restriction on the depth of a pipeline. That is, your solution should handle arbitrary number of commands chained together with operator |
- You must handle combinations of redirection and pipeline when they can be combined. An ex is: sort < file.txt | uniq. However, if there is a conflict between redirection and pipeline, i.e. one file descriptor gets associated with two things; you must report a syntax error. An instance is: ls > 1.txt | grep FOO; we cannot redirect the stdout of ls to both 1.txt and grep at the same time.
- Your shell doesn’t need to handle built-in commands (cd, exit, path) in pipeline.
- Your shell must not accept new user commands while the pipeline is still running.