How can I check if a package is installed and install it if not?

Viewed 295194

I'm working on a Ubuntu system and currently this is what I'm doing:

if ! which command > /dev/null; then
   echo -e "Command not found! Install? (y/n) \c"
   read
   if "$REPLY" = "y"; then
      sudo apt-get install command
   fi
fi

Is this what most people would do? Or is there a more elegant solution?

27 Answers

dpkg-query --showformat='${db:Status-Status}'

This produces a small output string which is unlikely to change and is easy to compare deterministically without grep:

pkg=hello
status="$(dpkg-query -W --showformat='${db:Status-Status}' "$pkg" 2>&1)"
if [ ! $? = 0 ] || [ ! "$status" = installed ]; then
  sudo apt install $pkg
fi

The $? = 0 check is needed because if you've never installed a package before, and after you remove certain packages such as hello, dpkg-query exits with status 1 and outputs to stderr:

dpkg-query: no packages found matching hello

instead of outputting not-installed. The 2>&1 captures that error message too when it comes preventing it from going to the terminal.

For multiple packages:

pkgs='hello certbot'
install=false
for pkg in $pkgs; do
  status="$(dpkg-query -W --showformat='${db:Status-Status}' "$pkg" 2>&1)"
  if [ ! $? = 0 ] || [ ! "$status" = installed ]; then
    install=true
    break
  fi
done
if "$install"; then
  sudo apt install $pkgs
fi

The possible statuses are documented in man dpkg-query as:

   n = Not-installed
   c = Config-files
   H = Half-installed
   U = Unpacked
   F = Half-configured
   W = Triggers-awaiting
   t = Triggers-pending
   i = Installed

The single letter versions are obtainable with db:Status-Abbrev, but they come together with the action and error status, so you get 3 characters and would need to cut it.

So I think it is reliable enough to rely on the uncapitalized statuses (Config-files vs config-files) not changing instead.

dpkg -s exit status

This unfortunately doesn't do what most users want:

pkgs='qemu-user pandoc'
if ! dpkg -s $pkgs >/dev/null 2>&1; then
  sudo apt-get install $pkgs
fi

because for some packages, e.g. certbot, doing:

sudo apt install certbot
sudo apt remove certbot

leaves certbot in state config-files, which means that config files were left in the machine. And in that state, dpkg -s still returns 0, because the package metadata is still kept around so that those config files can be handled more nicely.

To actually make dpkg -s return 1 as desired, --purge would be needed:

sudo apt remove --purge certbot

which actually moves it into not-installed/dpkg-query: no packages found matching.

Note that only certain packages leave config files behind. A simpler package like hello goes directly from installed to not-installed without --purge.

Tested on Ubuntu 20.10.

Python apt package

There is a pre-installed Python 3 package called apt in Ubuntu 18.04 which exposes an Python apt interface!

A script that checks if a package is installed and installs it if not can be seen at: How to install a package using the python-apt API

Here is a copy for reference:

#!/usr/bin/env python
# aptinstall.py

import apt
import sys

pkg_name = "libjs-yui-doc"

cache = apt.cache.Cache()
cache.update()
cache.open()

pkg = cache[pkg_name]
if pkg.is_installed:
    print "{pkg_name} already installed".format(pkg_name=pkg_name)
else:
    pkg.mark_install()

    try:
        cache.commit()
    except Exception, arg:
        print >> sys.stderr, "Sorry, package installation failed [{err}]".format(err=str(arg))

Check if an executable is in PATH instead

See: How can I check if a program exists from a Bash script?

See also

I've settled on one based on Nultyi's answer:

MISSING=$(dpkg --get-selections $PACKAGES 2>&1 | grep -v 'install$' | awk '{ print $6 }')
# Optional check here to skip bothering with apt-get if $MISSING is empty
sudo apt-get install $MISSING

Basically, the error message from dpkg --get-selections is far easier to parse than most of the others, because it doesn't include statuses like "deinstall". It also can check multiple packages simultaneously, something you can't do with just error codes.

Explanation/example:

$ dpkg --get-selections  python3-venv python3-dev screen build-essential jq
dpkg: no packages found matching python3-venv
dpkg: no packages found matching python3-dev
screen                                          install
build-essential                                 install
dpkg: no packages found matching jq

So grep removes installed packages from the list, and awk pulls the package names out from the error message, resulting in MISSING='python3-venv python3-dev jq', which can be trivially inserted into an install command.

I'm not blindly issuing an apt-get install $PACKAGES, because as mentioned in the comments, this can unexpectedly upgrade packages you weren't planning on; not really a good idea for automated processes that are expected to be stable.

It seems that nowadays apt-get has an option --no-upgrade that just does what the OP wants:

--no-upgrade Do not upgrade packages. When used in conjunction with install, no-upgrade will prevent packages listed from being upgraded if they are already installed.

Manpage from https://linux.die.net/man/8/apt-get

Therefore you can use

apt-get install --no-upgrade package

and package will be installed only if it's not.

Inspired by Chris' answer:

#! /bin/bash

installed() {
    return $(dpkg-query -W -f '${Status}\n' "${1}" 2>&1|awk '/ok installed/{print 0;exit}{print 1}')
}

pkgs=(libgl1-mesa-dev xorg-dev vulkan-tools libvulkan-dev vulkan-validationlayers-dev spirv-tools)
missing_pkgs=""

for pkg in ${pkgs[@]}; do
    if ! $(installed $pkg) ; then
        missing_pkgs+=" $pkg"
    fi
done

if [ ! -z "$missing_pkgs" ]; then
    cmd="sudo apt install -y $missing_pkgs"
    echo $cmd
fi

This explicitly prints 0 if installed else 1 using only awk:

dpkg-query -W -f '${Status}\n' 'PKG' 2>&1|awk '/ok installed/{print 0;exit}{print 1}'

or if you prefer the other way around where 1 means installed and 0 otherwise:

dpkg-query -W -f '${Status}\n' 'PKG' 2>&1|awk '/ok installed/{print 1;exit}{print 0}'

** replace PKG with your package name

Convenience function:

installed() {
    return $(dpkg-query -W -f '${Status}\n' "${1}" 2>&1|awk '/ok installed/{print 0;exit}{print 1}')
}


# usage:
installed gcc && echo Yes || echo No

#or

if installed gcc; then
    echo yes
else
    echo no
fi

I had a similar requirement when running test locally instead of in Docker. Basically I only wanted to install any .deb files found if they weren't already installed.

# If there are .deb files in the folder, then install them
if [ `ls -1 *.deb 2> /dev/null | wc -l` -gt 0 ]; then
  for file in *.deb; do
    # Only install if not already installed (non-zero exit code)
    dpkg -I ${file} | grep Package: | sed -r 's/ Package:\s+(.*)/\1/g' | xargs dpkg -s
    if [ $? != 0 ]; then
        dpkg -i ${file}
    fi;
  done;
else
  err "No .deb files found in '$PWD'"
fi

I guess the only problem I can see is that it doesn't check the version number of the package so if .deb file is a newer version. Then this wouldn't overwrite the currently installed package.

Kinda based off yours just a little more 'elegant'. Just because I'm bored.

#!/bin/bash
FOUND=("\033[38;5;10m")
NOTFOUND=("\033[38;5;9m")
PKG="${@:1}"
command ${PKG} &>/dev/null
if [[ $? != 0 ]]; then
    echo -e "${NOTFOUND}[!] ${PKG} not found [!]"
    echo -e "${NOTFOUND}[!] Would you like to install ${PKG} now ? [!]"
    read -p "[Y/N] >$ " ANSWER
    if [[ ${ANSWER} == [yY] || ${ANSWER} == [yY][eE][sS] ]]; then
        if grep -q "bian" /etc/os-release; then
            sudo apt-get install ${PKG}
        elif grep -q "arch" /etc/os-release; then
            if [[ -f /bin/yay ]] || [[ -f /bin/yaourt ]]; then
                yaourt -S ${PKG} 2>./err || yay -S ${PKG} 2>./err
            else
                sudo pacman -S ${PKG}
            fi
        elif grep -q "fedora" /etc/os-release; then
             sudo dnf install ${PKG}
        else
            echo -e "${NOTFOUND}[!] This script couldn't detect your package manager [!]"
            echo -e "${NOTFOUND}[!] Manually install it [!]"
        fi
    elif [[ ${ANSWER} == [nN] || ${ANSWER} == [nN][oO] ]]; then
        echo -e "${NOTFOUND}[!] Exiting [!]"
    fi
else
    echo -e "${FOUND}[+] ${PKG} found [+]"
fi

The answers that suggest to use something along the lines of:

dpkg-query --showformat '${db:Status-Status}\n' --show $package | grep -q '^installed$'
dpkg-query --showformat '${Status}\n' --show $package | grep -q '^install ok installed$'

are correct.

But if you have the package dpkg-dev installed and you do not just want to check whether a package is installed but you also:

  • want to know whether a package is installed in a certain version
  • want to have a package in a certain architecture
  • want to see if a virtual package is provided

then you can abuse the dpkg-checkbuilddeps tool for this job:

dpkg-checkbuilddeps -d apt /dev/null

This will check whether apt is installed.

The following will check whether apt is installed in at least version 2.3.15 and grep is installed as amd64 and the virtual package x-window-manager is provided by some of the installed packages:

dpkg-checkbuilddeps -d 'apt (>= 2.3.15), grep:amd64, x-window-manager' /dev/null

The exit status of dpkg-checkbuilddeps will tell the script whether the dependencies are satisfied or not. Since this method supports passing multiple packages, you only have to run dpkg-checkbuilddeps once even if you want to check whether more than one package is installed.

Since you mentioned Ubuntu, and you want to do this programmatically(although dpkg variations can also be used but would be more complex to implement), this(which) will definitely work:

#!/bin/bash

pkgname=mutt
which $pkgname > /dev/null;isPackage=$?
if [ $isPackage != 0 ];then
        echo "$pkgname not installed"
        sleep 1
        read -r -p "${1:-$pkgname will be installed. Are you sure? [y/N]} " response
        case "$response" in
            [yY][eE][sS]|[yY]) 
                sudo apt-get install $pkgname
                ;;
            *)
                false
                ;;
        esac

else
        echo "$pkgname is installed"
        sleep 1
fi

Although for POSIX compatibility, you would want to use command -v instead as mentioned in another similar question.

In that case, which $pkgname > /dev/null should be replaced by command -v $pkgname in the above code sample.

I use the following way:

which mySQL 2>&1|tee 1> /dev/null
  if [[ "$?" == 0 ]]; then
                echo -e "\e[42m MySQL already installed. Moving on...\e[0m"
        else
        sudo apt-get install -y mysql-server
                if [[ "$?" == 0 ]]; then
                        echo -e "\e[42mMy SQL installed\e[0m"
                else
                        echo -e "\e[42Installation failed\e[0m"
                fi
        fi

I use this solution as I find it most straightforward.

function must_install(){
   return "$(apt -qq list $var --installed 2> /dev/null |wc -l)"
}

function install_if() {
    unset install
    for var in "$@"
    do
        if $(must_install $var)
        then
            install+="${var} "
        fi
    done
    if [ -n "$install" ];
    then
        sudo apt-get install -qy $install
    fi
}

The neat thing is, must_install returns 1 or 0 which is then interpreted as true or false from the calling if, so we don't need any test using [].

install_if takes any number of packages separated by space.

The problem is apt is not meant to be used in scripts, so this might stop working at any time. 8)

All the answers are good but seem to be complex for beginners like myself to understand. so here is the solution that worked for me. My Linux environment is centOS but can't confirm it works for all distributions

PACKAGE_NAME=${PACKAGE_NAME:-node}

if ! command -v $PACKAGE_NAME > /dev/null; then
    echo "Installing $PACKAGE_NAME ..."
else
    echo "$PACKAGE_NAME already installed"
fi
Related