Distinct persistent bash history files using emacs shell

Introduction

Many people use bash and many people use emacs but not many people use emacs shell. I decided on this latest laptop refresh to address some development ergonomics. In this case, I tend to have different emacs shell sessions in which I type different types of commands. If I reboot or restart emacs it would be nice if the jruby shell has my old JRuby command history and my mri shell has old C Ruby command history. No mixing of history between shells or complete destruction of history (e.g.last shell closed wins at writing the file).

Talk about a post made for an obscure Venn diagram of users 😃 Yet, here I am helping out the dozens of netizens who may want this.

Bash Configuration

This is the extremely easy to lookup stuff. I want a large history and I want that history to always append. For good measure let’s weed out duplicate commands to make running history show more useful context (~/.bashrc):

export HISTSIZE=100000
export HISTFILESIZE=100000
export HISTCONTROL=ignoredups:erasedups
shopt -s histappend

These are sensible settings and I would think most people use these settings if they have ever decided to configure how their shell history works.

The piece of magic on the bash side is to set HISTFILE for each named shell:

# If using emacs retain history for each named buffer so that future
# sessions will retain just that buffers history.
if [ ! -z "${EMACS_SHELL_NAME}" ]; then
  export HISTFILE=".bash_history_${EMACS_SHELL_NAME}"
fi

If we are starting a new shell and the environment variable EMACS_SHELL_NAME exists then it gets its own history file rather than using the default history file. Also since we shopt -s histappend we will just keep building that history file up. Neat…Except emacs does not set an EMACS_SHELL_NAME variable. Let’s figure that side out…

Emacs Configuration

To enter shell mode you run (shell) which, by default, makes a shell run in a buffer called *shell*. No one wants *shell* as a buffer name and so far as I can tell WE ALL make a similar interactive function (.emacs):

(defun make-shell (buffer-name) "Create a shell buffer with specific name"
  (interactive "sShell Name: ")
  (let ((buffer (generate-new-buffer buffer-name)))
    (setenv "EMACS_SHELL_NAME" buffer-name)
    (shell buffer)))

This prompts for a buffer name, then makes it, and passes it as an optional parameter to the shell function. We get a new named buffer with a shell running.

The one line I added to connect this all together is (setenv "EMACS_SHELL_NAME" buffer-name). With that line our newly created shells have the info they need to create the right HISTFILE in .bashrc.

Conclusion

This ended up being simple to wire up which made me wonder why I didn’t do this like 30 years ago? Probably because it was only a tiny bit annoying and I didn’t care enough to fix it. Maybe it was just that I had to wire this up in multiple files and it was more like making a feature than configuring one.

One thing I do know is that I have wished many many times that I still had an old command in my history only to realize it was overwritten by some inconsequential shell history.

In any case if anyone ever searches for this…you’re welcome.