I implemented a “debugger” for Bash in six lines of Bash. It kind of behaves like JavaScript’s debugger keyword. Here’s how it works:

debugger() {
  echo "Stopped in REPL. Press ^D to resume, or ^C to abort."
  local line
  while read -r -p "> " line; do
    eval "$line"
  done
  echo
}

And there it is. Add this to a script, insert a call to debugger somewhere, and run the script. It’ll pause right execution right there. Once paused, we can do things like:

print the contents of variables with echo
run commands that are on our PATH (e.g., pwd, ls, …)
call functions defined in the script

… and pretty much everything that we could have done if we were editing the script directly. Here’s a short session demonstrating how it can be used:

#!/usr/bin/env bash

debugger() {
  # ... implemented above ...
}

foo=1
debugger
echo "foo: $foo"

❯ foo.sh
Stopped in REPL. Press ^D to resume, or ^C to abort.
> pwd
/Users/jez
> echo $foo
1
> foo=42
> ^D
foo: 42

Stopping on failures

I find that most of the time this is useful when a script is failing for some reason. Rather than put a debugger call right before the failing command, I can just add this at the top of the file:

trap 'debugger' ERR

When any command has a non-zero exit code, Bash will run debugger and pause the program.

I’ve been keeping this function and trap call commented out at the top of my scripts and uncommenting them when needed (It uses eval, which is not the best from a security perspective, which is why it’s commented by default).

Future work

Of course, I said “debugger” in quotes earlier because it’s not really a debugger:

Using it requires editing the script we want to debug to include these lines, and then calling debugger somewhere. It doesn’t launch an inferior process and control it, like gdb or lldb would.
There’s no break command to edit breakpoints while stopped. All breakpoints must have been written into the program up front.
There’s also no step or next commands for stepping into or over the next function or command.
When it stops, it doesn’t show the text content of the last line that executed, or even the line number.

But I have some thoughts on how to implement these, too… Bash’s trap builtin has a way to trap DEBUG, which runs after every command. I think I could make clever use of traps to implementat least one of step or next, and definitely something that says “stopped on line X” and maybe even use that to print the source text of that line. Implementing break seems to be the hardest—I don’t have any ideas for that one right now.

I’m releasing this code into the public domain. If you want to change it to implement any of these features, I’d be more than interested to hear about it!