Build your own shell

Goal

Write a working interactive shell that runs external commands, wires pipelines, redirects I/O, and handles signals and background jobs. Building it turns the process model from theory into reflex: fork/exec/wait, file descriptors and dup2, process groups, and the signal-handling that makes Ctrl-C and job control work.

Subject: full brief & instructions

Practices

• Linux system deep dive — fork/exec, process states, signals, and file descriptors are exactly what this exercises.
• Concurrency and parallelism — processes, pipes, and waiting are concurrency at the OS level.
• Networking fundamentals — the same fd/epoll machinery underlies socket servers.

Milestones

1. REPL + external commands. Read a line, tokenize it, fork + execvp the command, and waitpid for it. Handle the empty line and a not-found command cleanly.
2. Built-ins. Implement cd, exit, and pwd — the ones that must run in the shell process itself (a forked cd changes nothing), which teaches why the distinction exists.
3. Redirection. Parse >, >>, < and rewire stdin/stdout with open + dup2 before exec; confirm the child inherits the redirected descriptors.
4. Pipelines. Support a | b | c: create pipes, fork each stage, wire read/write ends with dup2, close unused fds (the leaked-fd bug is a rite of passage), and wait on all.
5. Signals & job control. Handle SIGINT/SIGTSTP so they hit the foreground job not the shell; add & background execution, process groups, and a jobs listing.

Stretch goals

• Add environment-variable expansion, globbing, and quoting rules.
• Add command history and line editing (readline or a raw-mode implementation).

Related

• Build your own shell — subject — the full feature spec and acceptance checks for this exercise.
• Linux system deep dive — cites this exercise in its # Practice section.
• Write Yourself a Git — the other "build the tool you use daily" systems exercise.