This is the first in a series of posts about Bash history configuration.

Bash can keep a history of the commands you’ve issued and make them available via multiple methods, including Readline history commands and history expansion.

An example of a Readline history command is reverse-search-history, which is usually bound to C-r (Ctrl+R), and incrementally searches the history backwards.

An example for history expansion is !!, an event designator for the previous command.

Having your history around is very useful, but the default settings leave a few things on the table:

• History entries seem to disappear
• History contains duplicates and useless junk like ls, which you’re not very likely ever to feel like retrieving
• Multi-line entries are folded into a single line, or split into multiple separate entries, or even seem to change how they’re stored
• Parallel shell sessions overwrite each other’s history

In these posts, I explain my Bash history setup, which addresses all of these points. “Perfect” is of course in jest—I don’t think I’ve seen many topics where people are as opinionated as when it comes to customizing a command-line setup. So, “perfect for me”, probably.

Note that I’m not going to talk about settings that change the behaviour of history expansion, but only how history is stored.

Enabling history

Bash can be compiled without history support (--enable-history option for configure), but it’s enabled by default unless not supported by the operating system.

Given history support, interactive shells have history turned on by default, but this can also be controlled using the history shell option:

set -o history

to enable, and

set +o history

to disable. If enabled, the SHELLOPTS variable contains history, and the status is also reported by set -o.

Infinite history

The first nice-to-have is history that goes back all the way.

To understand how this works, we have to know a bit about how Bash stores history. Command history is stored in multiple places: each session has an in-memory history, which is initialized from a history file; when the session is finished, the in-memory history is written back to the history file.

Configuring this in a way so multiple parallel sessions don’t wipe out each other’s histories will be subject of a separate post; for now, we’re just thinking about a single session.

The history file by default lives at ~/.bash_history; to put it elsewhere, set the HISTFILE variable. I like my home directory uncluttered, and I try to follow the XDG Base Directory Specification, so I have this in my .bashrc:

HISTFILE="$HOME/.local/share/bash/history"

where ~/.local/share is the default value of $XDG_DATA_HOME.

When the session history gets written to the history file, the file is either overwritten, or the history is appended. This is controlled with the histappend shell option:

shopt -s histappend

to append, and

shopt -u histappend

to overwrite the history file.

Since we only care about an individual shell session so far, we don’t need to append; we can overwrite the history file at the end of each session without losing anything.

The size of the history is controlled with the HISTSIZE variable (defaults to 500), and the size of the history file with HISTFILESIZE. Common advice is to just set these to large numbers; however, since Bash 4.3, they can be set to a negative value for unlimited history. (Watch out, setting to 0 disables history and truncates the history file, respectively!)

So, for an infinite history (and so far ignoring implications of parallel shell session), we’d want something like this in our .bashrc:

# No clutter in the home directory
HISTFILE="$HOME/.local/share/bash/history"

# Overwrite when storing
shopt -u histappend

# Infinite session history
HISTSIZE=-1

# Infinite history file
HISTFILESIZE=-1

In the next post, we’ll look at controlling what exactly goes into the history.