1473 lines
46 KiB
Bash
Executable File
1473 lines
46 KiB
Bash
Executable File
#!/bin/sh
|
||
#
|
||
# pfetch - Simple POSIX sh fetch script.
|
||
|
||
log() {
|
||
# End here if no data was found.
|
||
[ "$2" ] || return
|
||
|
||
# Store the value of '$1' as we reset the argument list below.
|
||
name=$1
|
||
|
||
# Use 'set --' as a means of stripping all leading and trailing
|
||
# white-space from the info string. This also normalizes all
|
||
# white-space inside of the string.
|
||
#
|
||
# Disable the shellcheck warning for word-splitting
|
||
# as it's safe and intended ('set -f' disables globbing).
|
||
# shellcheck disable=2046,2086
|
||
{
|
||
set -f
|
||
set +f -- $2
|
||
info=$*
|
||
}
|
||
|
||
# Construct the information string.
|
||
out="[3${PF_COL1-4};1m${name}[m"
|
||
out="$out${PF_SEP}[$((info_length-${#name}))C"
|
||
out="$out[3${PF_COL2-7}m${info}"
|
||
}
|
||
|
||
get_title() {
|
||
# Username is retrieved by first checking '$USER' with a fallback
|
||
# to the 'id -un' command.
|
||
user=${USER:-$(id -un)}
|
||
|
||
# Hostname is retrieved by first checking '$HOSTNAME' with a fallback
|
||
# to the 'hostname' command.
|
||
#
|
||
# Disable the warning about '$HOSTNAME' being undefined in POSIX sh as
|
||
# the intention for using it is allowing the user to overwrite the
|
||
# value on invocation.
|
||
# shellcheck disable=SC2039
|
||
hostname=${HOSTNAME:-${hostname:-$(hostname)}}
|
||
|
||
log "[3${PF_COL3:-1}m${user}${c7}@[3${PF_COL3:-1}m${hostname}" " "
|
||
}
|
||
|
||
get_os() {
|
||
# This function is called twice, once to detect the distribution name
|
||
# for the purposes of picking an ascii art early and secondly to display
|
||
# the distribution name in the info output (if enabled).
|
||
#
|
||
# On first run, this function displays _nothing_, only on the second
|
||
# invocation is 'log()' called.
|
||
[ "$distro" ] && {
|
||
log os "$distro"
|
||
return
|
||
}
|
||
|
||
case $os in
|
||
Linux*)
|
||
# Some Linux distributions (which are based on others)
|
||
# fail to identify as they **do not** change the upstream
|
||
# distribution's identification packages or files.
|
||
#
|
||
# It is senseless to add a special case in the code for
|
||
# each and every distribution (which _is_ technically no
|
||
# different from what it is based on) as they're either too
|
||
# lazy to modify upstream's identification files or they
|
||
# don't have the know-how (or means) to ship their own
|
||
# lsb-release package.
|
||
#
|
||
# This causes users to think there's a bug in system detection
|
||
# tools like neofetch or pfetch when they technically *do*
|
||
# function correctly.
|
||
#
|
||
# Exceptions are made for distributions which are independent,
|
||
# not based on another distribution or follow different
|
||
# standards.
|
||
#
|
||
# This applies only to distributions which follow the standard
|
||
# by shipping unmodified identification files and packages
|
||
# from their respective upstreams.
|
||
if command -v lsb_release; then
|
||
distro=$(lsb_release -sd)
|
||
|
||
# Android detection works by checking for the existence of
|
||
# the follow two directories. I don't think there's a simpler
|
||
# method than this.
|
||
elif [ -d /system/app ] && [ -d /system/priv-app ]; then
|
||
distro="Android $(getprop ro.build.version.release)"
|
||
|
||
else
|
||
# This used to be a simple '. /etc/os-release' but I believe
|
||
# this is insecure as we blindly executed whatever was in the
|
||
# file. This parser instead simply handles 'key=val', treating
|
||
# the file contents as plain-text.
|
||
while IFS='=' read -r key val; do
|
||
case $key in
|
||
PRETTY_NAME) distro=$val ;;
|
||
esac
|
||
done < /etc/os-release
|
||
fi
|
||
|
||
# 'os-release' and 'lsb_release' sometimes add quotes
|
||
# around the distribution name, strip them.
|
||
distro=${distro##[\"\']}
|
||
distro=${distro%%[\"\']}
|
||
|
||
# Special cases for (independent) distributions which
|
||
# don't follow any os-release/lsb standards whatsoever.
|
||
command -v crux && distro=$(crux)
|
||
command -v guix && distro='Guix System'
|
||
|
||
# Check to see if we're running Bedrock Linux which is
|
||
# very unique. This simply checks to see if the user's
|
||
# PATH contais a Bedrock specific value.
|
||
case $PATH in
|
||
*/bedrock/cross/*) distro='Bedrock Linux'
|
||
esac
|
||
|
||
# Check to see if Linux is running in Windows 10 under
|
||
# WSL1 (Windows subsystem for Linux [version 1]) and
|
||
# append a string accordingly.
|
||
#
|
||
# If the kernel version string ends in "-Microsoft",
|
||
# we're very likely running under Windows 10 in WSL1.
|
||
[ "${kernel%%*-Microsoft}" ] ||
|
||
distro="$distro on Windows 10 [WSL1]"
|
||
|
||
# Check to see if Linux is running in Windows 10 under
|
||
# WSL2 (Windows subsystem for Linux [version 2]) and
|
||
# append a string accordingly.
|
||
#
|
||
# This checks to see if '$WSLENV' is defined. This
|
||
# appends the Windows 10 string even if '$WSLENV' is
|
||
# empty. We only need to check that is has been _exported_.
|
||
distro="${distro}${WSLENV+ on Windows 10 [WSL2]}"
|
||
;;
|
||
|
||
Darwin*)
|
||
# Parse the SystemVersion.plist file to grab the macOS
|
||
# version. The file is in the following format:
|
||
#
|
||
# <key>ProductVersion</key>
|
||
# <string>10.14.6</string>
|
||
#
|
||
# 'IFS' is set to '<>' to enable splitting between the
|
||
# keys and a second 'read' is used to operate on the
|
||
# next line directly after a match.
|
||
#
|
||
# '_' is used to nullify a field. '_ _ line _' basically
|
||
# says "populate $line with the third field's contents".
|
||
while IFS='<>' read -r _ _ line _; do
|
||
case $line in
|
||
# Match 'ProductVersion' and read the next line
|
||
# directly as it contains the key's value.
|
||
ProductVersion)
|
||
IFS='<>' read -r _ _ mac_version _
|
||
break
|
||
;;
|
||
esac
|
||
done < /System/Library/CoreServices/SystemVersion.plist
|
||
|
||
# Use the ProductVersion to determine which macOS/OS X codename
|
||
# the system has. As far as I'm aware there's no "dynamic" way
|
||
# of grabbing this information.
|
||
case $mac_version in
|
||
10.4*) distro='Mac OS X Tiger' ;;
|
||
10.5*) distro='Mac OS X Leopard' ;;
|
||
10.6*) distro='Mac OS X Snow Leopard' ;;
|
||
10.7*) distro='Mac OS X Lion' ;;
|
||
10.8*) distro='OS X Mountain Lion' ;;
|
||
10.9*) distro='OS X Mavericks' ;;
|
||
10.10*) distro='OS X Yosemite' ;;
|
||
10.11*) distro='OS X El Capitan' ;;
|
||
10.12*) distro='macOS Sierra' ;;
|
||
10.13*) distro='macOS High Sierra' ;;
|
||
10.14*) distro='macOS Mojave' ;;
|
||
10.15*) distro='macOS Catalina' ;;
|
||
*) distro='macOS' ;;
|
||
esac
|
||
|
||
distro="$distro $mac_version"
|
||
;;
|
||
|
||
Haiku)
|
||
# Haiku uses 'uname -v' for version information
|
||
# instead of 'uname -r' which only prints '1'.
|
||
distro=$(uname -sv)
|
||
;;
|
||
|
||
Minix|DragonFly)
|
||
distro="$os $kernel"
|
||
|
||
# Minix and DragonFly don't support the escape
|
||
# sequences used, clear the exit trap.
|
||
trap '' EXIT
|
||
;;
|
||
|
||
SunOS)
|
||
# Grab the first line of the '/etc/release' file
|
||
# discarding everything after '('.
|
||
IFS='(' read -r distro _ < /etc/release
|
||
;;
|
||
|
||
*)
|
||
# Catch all to ensure '$distro' is never blank.
|
||
# This also handles the BSDs.
|
||
distro="$os $kernel"
|
||
;;
|
||
esac
|
||
}
|
||
|
||
get_kernel() {
|
||
case $os in
|
||
# Don't print kernel output on some systems as the
|
||
# OS name includes it.
|
||
*BSD*|Haiku|Minix|IRIX) ;;
|
||
|
||
*)
|
||
# '$kernel' is the cached output of 'uname -r'.
|
||
log kernel "$kernel"
|
||
;;
|
||
esac
|
||
}
|
||
|
||
get_host() {
|
||
case $os in
|
||
Linux*)
|
||
# Despite what these files are called, version doesn't
|
||
# always contain the version nor does name always contain
|
||
# the name.
|
||
read -r name < /sys/devices/virtual/dmi/id/product_name
|
||
read -r version < /sys/devices/virtual/dmi/id/product_version
|
||
read -r model < /sys/firmware/devicetree/base/model
|
||
|
||
host="$name $version $model"
|
||
;;
|
||
|
||
Darwin*|FreeBSD*|DragonFly*)
|
||
host=$(sysctl -n hw.model)
|
||
;;
|
||
|
||
NetBSD*)
|
||
host=$(sysctl -n machdep.dmi.system-vendor \
|
||
machdep.dmi.system-product)
|
||
;;
|
||
|
||
*BSD*|Minix)
|
||
host=$(sysctl -n hw.vendor hw.product)
|
||
;;
|
||
esac
|
||
|
||
# Turn the host string into an argument list so we can iterate
|
||
# over it and remove OEM strings and other information which
|
||
# shouldn't be displayed.
|
||
#
|
||
# Disable the shellcheck warning for word-splitting
|
||
# as it's safe and intended ('set -f' disables globbing).
|
||
# shellcheck disable=2046,2086
|
||
{
|
||
set -f
|
||
set +f -- $host
|
||
host=
|
||
}
|
||
|
||
# Iterate over the host string word by word as a means of stripping
|
||
# unwanted and OEM information from the string as a whole.
|
||
#
|
||
# This could have been implemented using a long 'sed' command with
|
||
# a list of word replacements, however I want to show that something
|
||
# like this is possible in pure sh.
|
||
#
|
||
# This string reconstruction is needed as some OEMs either leave the
|
||
# identification information as "To be filled by OEM", "Default",
|
||
# "undefined" etc and we shouldn't print this to the screen.
|
||
for word; do
|
||
# This works by reconstructing the string by excluding words
|
||
# found in the "blacklist" below. Only non-matches are appended
|
||
# to the final host string.
|
||
case $word in
|
||
To | [Bb]e | [Ff]illed | [Bb]y | O.E.M. | OEM |\
|
||
Not | Applicable | Specified | System | Product | Name |\
|
||
Version | Undefined | Default | string | INVALID | <20> | os )
|
||
continue
|
||
;;
|
||
esac
|
||
|
||
host="$host$word "
|
||
done
|
||
|
||
# '$arch' is the cached output from 'uname -m'.
|
||
log host "${host:-$arch}"
|
||
}
|
||
|
||
get_uptime() {
|
||
# Uptime works by retrieving the data in total seconds and then
|
||
# converting that data into days, hours and minutes using simple
|
||
# math.
|
||
case $os in
|
||
Linux*|Minix*)
|
||
IFS=. read -r s _ < /proc/uptime
|
||
;;
|
||
|
||
Darwin*|*BSD*|DragonFly*)
|
||
s=$(sysctl -n kern.boottime)
|
||
|
||
# Extract the uptime in seconds from the following output:
|
||
# [...] { sec = 1271934886, usec = 667779 } Thu Apr 22 12:14:46 2010
|
||
s=${s#*=}
|
||
s=${s%,*}
|
||
|
||
# The uptime format from 'sysctl' needs to be subtracted from
|
||
# the current time in seconds.
|
||
s=$(($(date +%s) - s))
|
||
;;
|
||
|
||
Haiku)
|
||
# The boot time is returned in microseconds, convert it to
|
||
# regular seconds.
|
||
s=$(($(system_time) / 1000000))
|
||
;;
|
||
|
||
SunOS)
|
||
# Split the output of 'kstat' on '.' and any white-space
|
||
# which exists in the command output.
|
||
#
|
||
# The output is as follows:
|
||
# unix:0:system_misc:snaptime 14809.906993005
|
||
#
|
||
# The parser extracts: ^^^^^
|
||
IFS=' .' read -r _ s _ <<-EOF
|
||
$(kstat -p unix:0:system_misc:snaptime)
|
||
EOF
|
||
;;
|
||
|
||
IRIX)
|
||
# Grab the uptime in a pretty format. Usually,
|
||
# 00:00:00 from the 'ps' command.
|
||
t=$(LC_ALL=POSIX ps -o etime= -p 1)
|
||
|
||
# Split the pretty output into days or hours
|
||
# based on the uptime.
|
||
case $t in
|
||
*-*) d=${t%%-*} t=${t#*-} ;;
|
||
*:*:*) h=${t%%:*} t=${t#*:} ;;
|
||
esac
|
||
|
||
h=${h#0} t=${t#0}
|
||
|
||
# Convert the split pretty fields back into
|
||
# seconds so we may re-convert them to our format.
|
||
s=$((${d:-0}*86400 + ${h:-0}*3600 + ${t%%:*}*60 + ${t#*:}))
|
||
;;
|
||
esac
|
||
|
||
# Convert the uptime from seconds into days, hours and minutes.
|
||
d=$((s / 60 / 60 / 24))
|
||
h=$((s / 60 / 60 % 24))
|
||
m=$((s / 60 % 60))
|
||
|
||
# Only append days, hours and minutes if they're non-zero.
|
||
[ "$d" = 0 ] || uptime="${uptime}${d}d "
|
||
[ "$h" = 0 ] || uptime="${uptime}${h}h "
|
||
[ "$m" = 0 ] || uptime="${uptime}${m}m "
|
||
|
||
log uptime "${uptime:-0m}"
|
||
}
|
||
|
||
get_pkgs() {
|
||
# This is just a simple wrapper around 'command -v' to avoid
|
||
# spamming '>/dev/null' throughout this function.
|
||
has() { command -v "$1" >/dev/null; }
|
||
|
||
# This works by first checking for which package managers are
|
||
# installed and finally by printing each package manager's
|
||
# package list with each package one per line.
|
||
#
|
||
# The output from this is then piped to 'wc -l' to count each
|
||
# line, giving us the total package count of whatever package
|
||
# managers are installed.
|
||
#
|
||
# Backticks are *required* here as '/bin/sh' on macOS is
|
||
# 'bash 3.2' and it can't handle the following:
|
||
#
|
||
# var=$(
|
||
# code here
|
||
# )
|
||
#
|
||
# shellcheck disable=2006
|
||
packages=`
|
||
case $os in
|
||
Linux*)
|
||
# Commands which print packages one per line.
|
||
has bonsai && bonsai list
|
||
has crux && pkginfo -i
|
||
has pacman-key && pacman -Qq
|
||
has dpkg && dpkg-query -f '.\n' -W
|
||
has rpm && rpm -qa
|
||
has xbps-query && xbps-query -l
|
||
has apk && apk info
|
||
has guix && guix package --list-installed
|
||
has opkg && opkg list-installed
|
||
|
||
# Directories containing packages.
|
||
has kiss && printf '%s\n' /var/db/kiss/installed/*/
|
||
has brew && printf '%s\n' "$(brew --cellar)/"*
|
||
has emerge && printf '%s\n' /var/db/pkg/*/*/
|
||
has pkgtool && printf '%s\n' /var/log/packages/*
|
||
has eopkg && printf '%s\n' /var/lib/eopkg/package/*
|
||
|
||
# 'nix' requires two commands.
|
||
has nix-store && {
|
||
nix-store -q --requisites /run/current-system/sw
|
||
nix-store -q --requisites ~.nix-profile
|
||
}
|
||
;;
|
||
|
||
Darwin*)
|
||
# Commands which print packages one per line.
|
||
has pkgin && pkgin list
|
||
|
||
# Directories containing packages.
|
||
has brew && printf '%s\n' /usr/local/Cellar/*
|
||
|
||
# 'port' prints a single line of output to 'stdout'
|
||
# when no packages are installed and exits with
|
||
# success causing a false-positive of 1 package
|
||
# installed.
|
||
#
|
||
# 'port' should really exit with a non-zero code
|
||
# in this case to allow scripts to cleanly handle
|
||
# this behavior.
|
||
has port && {
|
||
pkg_list=$(port installed)
|
||
|
||
[ "$pkg_list" = "No ports are installed." ] ||
|
||
printf '%s\n' "$pkg_list"
|
||
}
|
||
;;
|
||
|
||
FreeBSD*|DragonFly*)
|
||
pkg info
|
||
;;
|
||
|
||
OpenBSD*)
|
||
printf '%s\n' /var/db/pkg/*/
|
||
;;
|
||
|
||
NetBSD*)
|
||
pkg_info
|
||
;;
|
||
|
||
Haiku)
|
||
printf '%s\n' /boot/system/package-links/*
|
||
;;
|
||
|
||
Minix)
|
||
printf '%s\n' /usr/pkg/var/db/pkg/*/
|
||
;;
|
||
|
||
SunOS)
|
||
has pkginfo && pkginfo -i
|
||
has pkg && pkg list
|
||
;;
|
||
|
||
IRIX)
|
||
versions -b
|
||
;;
|
||
esac | wc -l
|
||
`
|
||
|
||
case $os in
|
||
# IRIX's package manager adds 3 lines of extra
|
||
# output which we must account for here.
|
||
IRIX) packages=$((packages - 3)) ;;
|
||
esac
|
||
|
||
[ "$packages" -gt 1 ] && log pkgs "$packages"
|
||
}
|
||
|
||
get_memory() {
|
||
case $os in
|
||
# Used memory is calculated using the following "formula":
|
||
# MemUsed = MemTotal + Shmem - MemFree - Buffers - Cached - SReclaimable
|
||
# Source: https://github.com/KittyKatt/screenFetch/issues/386
|
||
Linux*)
|
||
# Parse the '/proc/meminfo' file splitting on ':' and 'k'.
|
||
# The format of the file is 'key: 000kB' and an additional
|
||
# split is used on 'k' to filter out 'kB'.
|
||
while IFS=':k ' read -r key val _; do
|
||
case $key in
|
||
MemTotal)
|
||
mem_used=$((mem_used + val))
|
||
mem_full=$val
|
||
;;
|
||
|
||
Shmem)
|
||
mem_used=$((mem_used + val))
|
||
;;
|
||
|
||
MemFree|Buffers|Cached|SReclaimable)
|
||
mem_used=$((mem_used - val))
|
||
;;
|
||
esac
|
||
done < /proc/meminfo
|
||
|
||
mem_used=$((mem_used / 1024))
|
||
mem_full=$((mem_full / 1024))
|
||
;;
|
||
|
||
# Used memory is calculated using the following "formula":
|
||
# (wired + active + occupied) * 4 / 1024
|
||
Darwin*)
|
||
mem_full=$(($(sysctl -n hw.memsize) / 1024 / 1024))
|
||
|
||
# Parse the 'vmstat' file splitting on ':' and '.'.
|
||
# The format of the file is 'key: 000.' and an additional
|
||
# split is used on '.' to filter it out.
|
||
while IFS=:. read -r key val; do
|
||
case $key in
|
||
*' wired'*|*' active'*|*' occupied'*)
|
||
mem_used=$((mem_used + ${val:-0}))
|
||
;;
|
||
esac
|
||
|
||
# Using '<<-EOF' is the only way to loop over a command's
|
||
# output without the use of a pipe ('|').
|
||
# This ensures that any variables defined in the while loop
|
||
# are still accessible in the script.
|
||
done <<-EOF
|
||
$(vm_stat)
|
||
EOF
|
||
|
||
mem_used=$((mem_used * 4 / 1024))
|
||
;;
|
||
|
||
OpenBSD*)
|
||
mem_full=$(($(sysctl -n hw.physmem) / 1024 / 1024))
|
||
|
||
# This is a really simpler parser for 'vmstat' which grabs
|
||
# the used memory amount in a lazy way. 'vmstat' prints 3
|
||
# lines of output with the needed value being stored in the
|
||
# final line.
|
||
#
|
||
# This loop simply grabs the 3rd element of each line until
|
||
# the EOF is reached. Each line overwrites the value of the
|
||
# previous one so we're left with what we wanted. This isn't
|
||
# slow as only 3 lines are parsed.
|
||
while read -r _ _ line _; do
|
||
mem_used=${line%%M}
|
||
|
||
# Using '<<-EOF' is the only way to loop over a command's
|
||
# output without the use of a pipe ('|').
|
||
# This ensures that any variables defined in the while loop
|
||
# are still accessible in the script.
|
||
done <<-EOF
|
||
$(vmstat)
|
||
EOF
|
||
;;
|
||
|
||
# Used memory is calculated using the following "formula":
|
||
# mem_full - ((inactive + free + cache) * page_size / 1024)
|
||
FreeBSD*|DragonFly*)
|
||
mem_full=$(($(sysctl -n hw.physmem) / 1024 / 1024))
|
||
|
||
# Use 'set --' to store the output of the command in the
|
||
# argument list. POSIX sh has no arrays but this is close enough.
|
||
#
|
||
# Disable the shellcheck warning for word-splitting
|
||
# as it's safe and intended ('set -f' disables globbing).
|
||
# shellcheck disable=2046
|
||
{
|
||
set -f
|
||
set +f -- $(sysctl -n hw.pagesize \
|
||
vm.stats.vm.v_inactive_count \
|
||
vm.stats.vm.v_free_count \
|
||
vm.stats.vm.v_cache_count)
|
||
}
|
||
|
||
# Calculate the amount of used memory.
|
||
# $1: hw.pagesize
|
||
# $2: vm.stats.vm.v_inactive_count
|
||
# $3: vm.stats.vm.v_free_count
|
||
# $4: vm.stats.vm.v_cache_count
|
||
mem_used=$((mem_full - (($2 + $3 + $4) * $1 / 1024 / 1024)))
|
||
;;
|
||
|
||
NetBSD*)
|
||
mem_full=$(($(sysctl -n hw.physmem64) / 1024 / 1024))
|
||
|
||
# NetBSD implements a lot of the Linux '/proc' filesystem,
|
||
# this uses the same parser as the Linux memory detection.
|
||
while IFS=':k ' read -r key val _; do
|
||
case $key in
|
||
MemFree)
|
||
mem_free=$((val / 1024))
|
||
break
|
||
;;
|
||
esac
|
||
done < /proc/meminfo
|
||
|
||
mem_used=$((mem_full - mem_free))
|
||
;;
|
||
|
||
Haiku)
|
||
# Read the first line of 'sysinfo -mem' splitting on
|
||
# '(', ' ', and ')'. The needed information is then
|
||
# stored in the 5th and 7th elements. Using '_' "consumes"
|
||
# an element allowing us to proceed to the next one.
|
||
#
|
||
# The parsed format is as follows:
|
||
# 3501142016 bytes free (used/max 792645632 / 4293787648)
|
||
IFS='( )' read -r _ _ _ _ mem_used _ mem_full <<-EOF
|
||
$(sysinfo -mem)
|
||
EOF
|
||
|
||
mem_used=$((mem_used / 1024 / 1024))
|
||
mem_full=$((mem_full / 1024 / 1024))
|
||
;;
|
||
|
||
Minix)
|
||
# Minix includes the '/proc' filesystem though the format
|
||
# differs from Linux. The '/proc/meminfo' file is only a
|
||
# single line with space separated elements and elements
|
||
# 2 and 3 contain the total and free memory numbers.
|
||
read -r _ mem_full mem_free _ < /proc/meminfo
|
||
|
||
mem_used=$(((mem_full - mem_free) / 1024))
|
||
mem_full=$(( mem_full / 1024))
|
||
;;
|
||
|
||
SunOS)
|
||
hw_pagesize=$(pagesize)
|
||
|
||
# 'kstat' outputs memory in the following format:
|
||
# unix:0:system_pages:pagestotal 1046397
|
||
# unix:0:system_pages:pagesfree 885018
|
||
#
|
||
# This simply uses the first "element" (white-space
|
||
# separated) as the key and the second element as the
|
||
# value.
|
||
#
|
||
# A variable is then assigned based on the key.
|
||
while read -r key val; do
|
||
case $key in
|
||
*total) pages_full=$val ;;
|
||
*free) pages_free=$val ;;
|
||
esac
|
||
done <<-EOF
|
||
$(kstat -p unix:0:system_pages:pagestotal \
|
||
unix:0:system_pages:pagesfree)
|
||
EOF
|
||
|
||
mem_full=$((pages_full * hw_pagesize / 1024 / 1024))
|
||
mem_free=$((pages_free * hw_pagesize / 1024 / 1024))
|
||
mem_used=$((mem_full - mem_free))
|
||
;;
|
||
|
||
IRIX)
|
||
# Read the memory information from the 'top' command. Parse
|
||
# and split each line until we reach the line starting with
|
||
# "Memory".
|
||
#
|
||
# Example output: Memory: 160M max, 147M avail, .....
|
||
while IFS=' :' read -r label mem_full _ mem_free _; do
|
||
case $label in
|
||
Memory)
|
||
mem_full=${mem_full%M}
|
||
mem_free=${mem_free%M}
|
||
break
|
||
;;
|
||
esac
|
||
done <<-EOF
|
||
$(top -n)
|
||
EOF
|
||
|
||
mem_used=$((mem_full - mem_free))
|
||
;;
|
||
esac
|
||
|
||
log memory "${mem_used:-?}M / ${mem_full:-?}M"
|
||
}
|
||
|
||
get_wm() {
|
||
case $os in
|
||
# Don't display window manager on macOS.
|
||
Darwin*) ;;
|
||
|
||
*)
|
||
# xprop can be used to grab the window manager's properties
|
||
# which contains the window manager's name under '_NET_WM_NAME'.
|
||
#
|
||
# The upside to using 'xprop' is that you don't need to hardcode
|
||
# a list of known window manager names. The downside is that
|
||
# not all window managers conform to setting the '_NET_WM_NAME'
|
||
# atom..
|
||
#
|
||
# List of window managers which fail to set the name atom:
|
||
# catwm, fvwm, dwm, 2bwm, monster, wmaker and sowm [mine! ;)].
|
||
#
|
||
# The final downside to this approach is that it does _not_
|
||
# support Wayland environments. The only solution which supports
|
||
# Wayland is the 'ps' parsing mentioned below.
|
||
#
|
||
# A more naive implementation is to parse the last line of
|
||
# '~/.xinitrc' to extract the second white-space separated
|
||
# element.
|
||
#
|
||
# The issue with an approach like this is that this line data
|
||
# does not always equate to the name of the window manager and
|
||
# could in theory be _anything_.
|
||
#
|
||
# This also fails when the user launches xorg through a display
|
||
# manager or other means.
|
||
#
|
||
#
|
||
# Another naive solution is to parse 'ps' with a hardcoded list
|
||
# of window managers to detect the current window manager (based
|
||
# on what is running).
|
||
#
|
||
# The issue with this approach is the need to hardcode and
|
||
# maintain a list of known window managers.
|
||
#
|
||
# Another issue is that process names do not always equate to
|
||
# the name of the window manager. False-positives can happen too.
|
||
#
|
||
# This is the only solution which supports Wayland based
|
||
# environments sadly. It'd be nice if some kind of standard were
|
||
# established to identify Wayland environments.
|
||
#
|
||
# pfetch's goal is to remain _simple_, if you'd like a "full"
|
||
# implementation of window manager detection use 'neofetch'.
|
||
#
|
||
# Neofetch use a combination of 'xprop' and 'ps' parsing to
|
||
# support all window managers (including non-conforming and
|
||
# Wayland) though it's a lot more complicated!
|
||
|
||
# Don't display window manager if X isn't running.
|
||
[ "$DISPLAY" ] || return
|
||
|
||
# This is a two pass call to xprop. One call to get the window
|
||
# manager's ID and another to print its properties.
|
||
command -v xprop && {
|
||
# The output of the ID command is as follows:
|
||
# _NET_SUPPORTING_WM_CHECK: window id # 0x400000
|
||
#
|
||
# To extract the ID, everything before the last space
|
||
# is removed.
|
||
id=$(xprop -root -notype _NET_SUPPORTING_WM_CHECK)
|
||
id=${id##* }
|
||
|
||
# The output of the property command is as follows:
|
||
# _NAME 8t
|
||
# _NET_WM_PID = 252
|
||
# _NET_WM_NAME = "bspwm"
|
||
# _NET_SUPPORTING_WM_CHECK: window id # 0x400000
|
||
# WM_CLASS = "wm", "Bspwm"
|
||
#
|
||
# To extract the name, everything before '_NET_WM_NAME = \"'
|
||
# is removed and everything after the next '"' is removed.
|
||
wm=$(xprop -id "$id" -notype -len 25 -f _NET_WM_NAME 8t)
|
||
|
||
# Handle cases of a window manager _not_ populating the
|
||
# '_NET_WM_NAME' atom. Display nothing in this case.
|
||
case $wm in
|
||
*'_NET_WM_NAME = '*)
|
||
wm=${wm##*_NET_WM_NAME = \"}
|
||
wm=${wm%%\"*}
|
||
;;
|
||
|
||
*)
|
||
# Fallback to checking the process list
|
||
# for the select few window managers which
|
||
# don't set '_NET_WM_NAME'.
|
||
#
|
||
# TODO: This is currently limited to 'grep'
|
||
# implementations providing the '-o'
|
||
# flag. This needs to be replaced with
|
||
# a command which searches for a list
|
||
# of strings and returns _only_ the
|
||
# first match's contents (also ensuring
|
||
# the search itself isn't matched).
|
||
#
|
||
# A generic parser isn't possible as
|
||
# the output of 'ps' is _not_ the same
|
||
# between implementations and across
|
||
# operating systems.
|
||
#
|
||
# The simple search method above works
|
||
# regardless of 'ps' implementation.
|
||
#
|
||
# Disable the shellcheck warning about using
|
||
# 'pgrep' instead of 'ps | grep' as 'pgrep'
|
||
# is not always available.
|
||
# shellcheck disable=2009
|
||
wm=$(ps x | grep -o \
|
||
-e '[c]atwm' \
|
||
-e '[f]vwm' \
|
||
-e '[d]wm' \
|
||
-e '[2]bwm' \
|
||
-e '[m]onsterwm' \
|
||
-e '[w]maker' \
|
||
-e '[s]owm')
|
||
;;
|
||
esac
|
||
}
|
||
;;
|
||
esac
|
||
|
||
log wm "$wm"
|
||
}
|
||
|
||
|
||
get_de() {
|
||
# This only supports Xorg related desktop environments though
|
||
# this is fine as knowing the desktop envrionment on Windows,
|
||
# macOS etc is useless (they'll always report the same value).
|
||
#
|
||
# Display the value of '$XDG_CURRENT_DESKTOP', if it's empty,
|
||
# display the value of '$DESKTOP_SESSION'.
|
||
log de "${XDG_CURRENT_DESKTOP:-$DESKTOP_SESSION}"
|
||
}
|
||
|
||
get_shell() {
|
||
# Display the basename of the '$SHELL' environment variable.
|
||
log shell "${SHELL##*/}"
|
||
}
|
||
|
||
get_editor() {
|
||
# Display the value of '$VISUAL', if it's empty, display the
|
||
# value of '$EDITOR'.
|
||
log editor "${VISUAL:-$EDITOR}"
|
||
}
|
||
|
||
get_palette() {
|
||
# Print the first 8 terminal colors. This uses the existing
|
||
# sequences to change text color with a sequence prepended
|
||
# to reverse the foreground and background colors.
|
||
#
|
||
# This allows us to save hardcoding a second set of sequences
|
||
# for background colors.
|
||
palette="[7m$c1 $c1 $c2 $c2 $c3 $c3 $c4 $c4 $c5 $c5 $c6 $c6 [m"
|
||
|
||
# Print the palette with a new-line before and afterwards.
|
||
printf '\n' >&6
|
||
log "$palette
|
||
" " "
|
||
}
|
||
|
||
get_ascii() {
|
||
# This is a simple function to read the contents of
|
||
# an ascii file from 'stdin'. It allows for the use
|
||
# of '<<-EOF' to prevent the break in indentation in
|
||
# this source code.
|
||
#
|
||
# This function also sets the text colors according
|
||
# to the ascii color.
|
||
read_ascii() {
|
||
# 'PF_COL1': Set the info name color according to ascii color.
|
||
# 'PF_COL3': Set the title color to some other color. ¯\_(ツ)_/¯
|
||
PF_COL1=${PF_COL1:-${1:-7}}
|
||
PF_COL3=${PF_COL3:-$((${1:-7}%8+1))}
|
||
|
||
# POSIX sh has no 'var+=' so 'var=${var}append' is used. What's
|
||
# interesting is that 'var+=' _is_ supported inside '$(())'
|
||
# (arithmetic) though there's no support for 'var++/var--'.
|
||
#
|
||
# There is also no $'\n' to add a "literal"(?) newline to the
|
||
# string. The simplest workaround being to break the line inside
|
||
# the string (though this has the caveat of breaking indentation).
|
||
while IFS= read -r line; do
|
||
ascii="$ascii$line
|
||
"
|
||
done
|
||
}
|
||
|
||
# This checks for ascii art in the following order:
|
||
# '$1': Argument given to 'get_ascii()' directly.
|
||
# '$PF_ASCII': Environment variable set by user.
|
||
# '$distro': The detected distribution name.
|
||
# '$os': The name of the operating system/kernel.
|
||
#
|
||
# NOTE: Each ascii art below is indented using tabs, this
|
||
# allows indentation to continue naturally despite
|
||
# the use of '<<-EOF'.
|
||
case ${1:-${PF_ASCII:-${distro:-$os}}} in
|
||
[Aa]lpine*)
|
||
read_ascii 4 <<-EOF
|
||
${c4} /\\ /\\
|
||
/${c7}/ ${c4}\\ \\
|
||
/${c7}/ ${c4}\\ \\
|
||
/${c7}// ${c4}\\ \\
|
||
${c7}// ${c4}\\ \\
|
||
${c4}\\
|
||
EOF
|
||
;;
|
||
|
||
[Aa]ndroid*)
|
||
read_ascii 2 <<-EOF
|
||
${c2} ;, ,;
|
||
${c2} ';,.-----.,;'
|
||
${c2} ,' ',
|
||
${c2} / O O \\
|
||
${c2}| |
|
||
${c2}'-----------------'
|
||
EOF
|
||
;;
|
||
|
||
[Aa]rch*)
|
||
read_ascii 4 <<-EOF
|
||
${c6} /\\
|
||
${c6} / \\
|
||
${c6} /\\ \\
|
||
${c4} / \\
|
||
${c4} / ,, \\
|
||
${c4} / | | -\\
|
||
${c4} /_-'' ''-_\\
|
||
EOF
|
||
;;
|
||
|
||
[Aa]rco*)
|
||
read_ascii 4 <<-EOF
|
||
${c4} /\\
|
||
${c4} / \\
|
||
${c4} / /\\ \\
|
||
${c4} / / \\ \\
|
||
${c4} / / \\ \\
|
||
${c4} / / _____\\ \\
|
||
${c4}/_/ \`----.\\_\\
|
||
EOF
|
||
;;
|
||
|
||
[Aa]rtix*)
|
||
read_ascii 6 <<-EOF
|
||
${c4} /\\
|
||
${c4} / \\
|
||
${c4} /\`'.,\\
|
||
${c4} / ',
|
||
${c4} / ,\`\\
|
||
${c4} / ,.'\`. \\
|
||
${c4}/.,'\` \`'.\\
|
||
EOF
|
||
;;
|
||
|
||
[Bb]edrock*)
|
||
read_ascii 4 <<-EOF
|
||
${c7}__
|
||
${c7}\\ \\___
|
||
${c7} \\ _ \\
|
||
${c7} \\___/
|
||
EOF
|
||
;;
|
||
|
||
[Cc]ent[Oo][Ss]*)
|
||
read_ascii 5 <<-EOF
|
||
${c2} ____${c3}^${c5}____
|
||
${c2} |\\ ${c3}|${c5} /|
|
||
${c2} | \\ ${c3}|${c5} / |
|
||
${c5}<---- ${c4}---->
|
||
${c4} | / ${c2}|${c3} \\ |
|
||
${c4} |/__${c2}|${c3}__\\|
|
||
${c2} v
|
||
EOF
|
||
;;
|
||
|
||
[Dd]ebian*)
|
||
read_ascii 1 <<-EOF
|
||
${c1} _____
|
||
${c1} / __ \\
|
||
${c1}| / |
|
||
${c1}| \\___-
|
||
${c1}-_
|
||
${c1} --_
|
||
EOF
|
||
;;
|
||
|
||
[Dd]ragon[Ff]ly*)
|
||
read_ascii 1 <<-EOF
|
||
,${c1}_${c7},
|
||
('-_${c1}|${c7}_-')
|
||
>--${c1}|${c7}--<
|
||
(_-'${c1}|${c7}'-_)
|
||
${c1}|
|
||
${c1}|
|
||
${c1}|
|
||
EOF
|
||
;;
|
||
|
||
[Ee]lementary*)
|
||
read_ascii <<-EOF
|
||
${c7} _______
|
||
${c7} / ____ \\
|
||
${c7}/ | / /\\
|
||
${c7}|__\\ / / |
|
||
${c7}\\ /__/ /
|
||
${c7}\\_______/
|
||
EOF
|
||
;;
|
||
|
||
[Ee]ndeavour*)
|
||
read_ascii 4 <<-EOF
|
||
${c1}/${c4}\\
|
||
${c1}/${c4}/ \\${c6}\\
|
||
${c1}/${c4}/ \\ ${c6}\\
|
||
${c1}/ ${c4}/ _) ${c6})
|
||
${c1}/_${c4}/___-- ${c6}__-
|
||
${c6}/____--
|
||
EOF
|
||
;;
|
||
|
||
[Ff]edora*)
|
||
read_ascii 4 <<-EOF
|
||
${c7} _____
|
||
/ __)${c4}\\${c7}
|
||
| / ${c4}\\ \\${c7}
|
||
${c4}__${c7}_| |_${c4}_/ /${c7}
|
||
${c4}/ ${c7}(_ _)${c4}_/${c7}
|
||
${c4}/ /${c7} | |
|
||
${c4}\\ \\${c7}__/ |
|
||
${c4}\\${c7}(_____/
|
||
EOF
|
||
;;
|
||
|
||
[Ff]ree[Bb][Ss][Dd]*)
|
||
read_ascii 1 <<-EOF
|
||
${c1}/\\,-'''''-,/\\
|
||
${c1}\\_) (_/
|
||
${c1}| |
|
||
${c1}| |
|
||
${c1}; ;
|
||
${c1}'-_____-'
|
||
EOF
|
||
;;
|
||
|
||
[Gg]entoo*)
|
||
read_ascii 5 <<-EOF
|
||
${c5} _-----_
|
||
${c5}( \\
|
||
${c5}\\ 0 \\
|
||
${c7} \\ )
|
||
${c7} / _/
|
||
${c7}( _-
|
||
${c7}\\____-
|
||
EOF
|
||
;;
|
||
|
||
[Gg]uix[Ss][Dd]*|[Gg]uix*)
|
||
read_ascii 3 <<-EOF
|
||
${c3}|.__ __.|
|
||
${c3}|__ \\ / __|
|
||
${c3}\\ \\ / /
|
||
${c3}\\ \\ / /
|
||
${c3}\\ \\ / /
|
||
${c3}\\ \\/ /
|
||
${c3}\\__/
|
||
EOF
|
||
;;
|
||
|
||
[Hh]aiku*)
|
||
read_ascii 3 <<-EOF
|
||
${c3} ,^,
|
||
${c3} / \\
|
||
${c3}*--_ ; ; _--*
|
||
${c3}\\ '" "' /
|
||
${c3}'. .'
|
||
${c3}.-'" "'-.
|
||
${c3}'-.__. .__.-'
|
||
${c3}|_|
|
||
EOF
|
||
;;
|
||
|
||
[Hh]yperbola*)
|
||
read_ascii <<-EOF
|
||
${c7} |\`__.\`/
|
||
${c7} \____/
|
||
${c7} .--.
|
||
${c7} / \\
|
||
${c7} / ___ \\
|
||
${c7}/ .\` \`.\\
|
||
${c7}/.\` \`.\\
|
||
EOF
|
||
;;
|
||
|
||
[Ii][Rr][Ii][Xx]*)
|
||
read_ascii 4 <<-EOF
|
||
TODO
|
||
TODO
|
||
TODO
|
||
TODO
|
||
TODO
|
||
TODO
|
||
TODO
|
||
EOF
|
||
;;
|
||
|
||
[Ll]inux*[Ll]ite*|[Ll]ite*)
|
||
read_ascii 3 <<-EOF
|
||
${c3} /\\
|
||
${c3} / \\
|
||
${c3} / ${c7}/ ${c3}/
|
||
${c3}> ${c7}/ ${c3}/
|
||
${c3}\\ ${c7}\\ ${c3}\\
|
||
${c3}\\_${c7}\\${c3}_\\
|
||
${c7} \\
|
||
EOF
|
||
;;
|
||
|
||
[Ll]inux*[Mm]int*|[Mm]int)
|
||
read_ascii 2 <<-EOF
|
||
${c2} ___________
|
||
${c2}|_ \\
|
||
${c2}| ${c7}| _____ ${c2}|
|
||
${c2}| ${c7}| | | | ${c2}|
|
||
${c2}| ${c7}| | | | ${c2}|
|
||
${c2}| ${c7}\\__${c7}___/ ${c2}|
|
||
${c2}\\_________/
|
||
EOF
|
||
;;
|
||
|
||
|
||
[Ll]inux*)
|
||
read_ascii 4 <<-EOF
|
||
${c4} ___
|
||
${c4}(${c7}.. ${c4}|
|
||
${c4}(${c5}<> ${c4}|
|
||
${c4}/ ${c7}__ ${c4}\\
|
||
${c4}( ${c7}/ \\ ${c4}/|
|
||
${c5}_${c4}/\\ ${c7}__)${c4}/${c5}_${c4})
|
||
${c5}\/${c4}-____${c5}\/
|
||
EOF
|
||
;;
|
||
|
||
[Mm]ac[Oo][Ss]*|[Dd]arwin*)
|
||
read_ascii 1 <<-EOF
|
||
${c1} .:'
|
||
${c1} _ :'_
|
||
${c2} .'\`_\`-'_\`\`.
|
||
${c2}:________.-'
|
||
${c3}:_______:
|
||
${c4} :_______\`-;
|
||
${c5} \`._.-._.'
|
||
EOF
|
||
;;
|
||
|
||
[Mm]ageia*)
|
||
read_ascii 2 <<-EOF
|
||
${c6} *
|
||
${c6} *
|
||
${c6} **
|
||
${c7} /\\__/\\
|
||
${c7}/ \\
|
||
${c7}\\ /
|
||
${c7} \\____/
|
||
EOF
|
||
;;
|
||
|
||
[Mm]anjaro*)
|
||
read_ascii 2 <<-EOF
|
||
${c2}||||||||| ||||
|
||
${c2}||||||||| ||||
|
||
${c2}|||| ||||
|
||
${c2}|||| |||| ||||
|
||
${c2}|||| |||| ||||
|
||
${c2}|||| |||| ||||
|
||
${c2}|||| |||| ||||
|
||
EOF
|
||
;;
|
||
|
||
[Mm]inix*)
|
||
read_ascii 4 <<-EOF
|
||
${c4} ,, ,,
|
||
${c4};${c7},${c4} ', ,' ${c7},${c4};
|
||
${c4}; ${c7}',${c4} ',,' ${c7},'${c4} ;
|
||
${c4}; ${c7}',${c4} ${c7},'${c4} ;
|
||
${c4}; ${c7};, '' ,;${c4} ;
|
||
${c4}; ${c7};${c4};${c7}',,'${c4};${c7};${c4} ;
|
||
${c4}', ${c7};${c4};; ;;${c7};${c4} ,'
|
||
${c4} '${c7};${c4}' '${c7};${c4}'
|
||
EOF
|
||
;;
|
||
|
||
[Mm][Xx]*)
|
||
read_ascii <<-EOF
|
||
${c7} \\\\ /
|
||
${c7} \\\\/
|
||
${c7} \\\\
|
||
${c7} /\\/ \\\\
|
||
${c7} / \\ /\\
|
||
${c7} / \\/ \\
|
||
${c7}/__________\\
|
||
EOF
|
||
;;
|
||
|
||
[Nn]et[Bb][Ss][Dd]*)
|
||
read_ascii 3 <<-EOF
|
||
${c7}\\\\${c3}\`-______,----__
|
||
${c7} \\\\ ${c3}__,---\`_
|
||
${c7} \\\\ ${c3}\`.____
|
||
${c7} \\\\${c3}-______,----\`-
|
||
${c7} \\\\
|
||
${c7} \\\\
|
||
${c7} \\\\
|
||
EOF
|
||
;;
|
||
|
||
[Nn]ix[Oo][Ss]*)
|
||
read_ascii 4 <<-EOF
|
||
${c4} \\\\ \\\\ //
|
||
${c4} ==\\\\__\\\\/ //
|
||
${c4} // \\\\//
|
||
${c4}==// //==
|
||
${c4} //\\\\___//
|
||
${c4}// /\\\\ \\\\==
|
||
${c4} // \\\\ \\\\
|
||
EOF
|
||
;;
|
||
|
||
[Oo]pen[Bb][Ss][Dd]*)
|
||
read_ascii 3 <<-EOF
|
||
${c3} _____
|
||
${c3} \\- -/
|
||
${c3} \\_/ \\
|
||
${c3} | ${c7}O O${c3} |
|
||
${c3} |_ < ) 3 )
|
||
${c3} / \\ /
|
||
${c3} /-_____-\\
|
||
EOF
|
||
;;
|
||
|
||
[Oo]pen[Ss][Uu][Ss][Ee]*|[Oo]pen*SUSE*|SUSE*|suse*)
|
||
read_ascii 2 <<-EOF
|
||
${c2} _______
|
||
${c2}__| __ \\
|
||
${c2} / .\\ \\
|
||
${c2} \\__/ |
|
||
${c2} _______|
|
||
${c2} \\_______
|
||
${c2}__________/
|
||
EOF
|
||
;;
|
||
|
||
[Oo]pen[Ww]rt*)
|
||
read_ascii 1 <<-EOF
|
||
${c1} _______
|
||
${c1}| |.-----.-----.-----.
|
||
${c1}| - || _ | -__| |
|
||
${c1}|_______|| __|_____|__|__|
|
||
${c1} ________|__| __
|
||
${c1}| | | |.----.| |_
|
||
${c1}| | | || _|| _|
|
||
${c1}|________||__| |____|
|
||
EOF
|
||
;;
|
||
|
||
[Pp]arabola*)
|
||
read_ascii 5 <<-EOF
|
||
${c5} __ __ __ _
|
||
${c5}.\`_//_//_/ / \`.
|
||
${c5} / .\`
|
||
${c5} / .\`
|
||
${c5} /.\`
|
||
${c5} /\`
|
||
EOF
|
||
;;
|
||
|
||
[Pp]op!_[Oo][Ss]*)
|
||
read_ascii 6 <<-EOF
|
||
${c6}______
|
||
${c6}\\ _ \\ __
|
||
${c6}\\ \\ \\ \\ / /
|
||
${c6}\\ \\_\\ \\ / /
|
||
${c6}\\ ___\\ /_/
|
||
${c6} \\ \\ _
|
||
${c6} __\\_\\__(_)_
|
||
${c6}(___________)
|
||
EOF
|
||
;;
|
||
|
||
[Pp]ure[Oo][Ss]*)
|
||
read_ascii <<-EOF
|
||
${c7} _____________
|
||
${c7}| _________ |
|
||
${c7}| | | |
|
||
${c7}| | | |
|
||
${c7}| |_________| |
|
||
${c7}|_____________|
|
||
EOF
|
||
;;
|
||
|
||
[Ss]lackware*)
|
||
read_ascii 4 <<-EOF
|
||
${c4} ________
|
||
${c4} / ______|
|
||
${c4} | |______
|
||
${c4} \\______ \\
|
||
${c4} ______| |
|
||
${c4}| |________/
|
||
${c4}|____________
|
||
EOF
|
||
;;
|
||
|
||
[Ss]un[Oo][Ss]|[Ss]olaris*)
|
||
read_ascii 3 <<-EOF
|
||
${c3} . .; .
|
||
${c3} . :; :: ;: .
|
||
${c3} .;. .. .. .;.
|
||
${c3}.. .. .. ..
|
||
${c3} .;, ,;.
|
||
EOF
|
||
;;
|
||
|
||
[Uu]buntu*)
|
||
read_ascii 3 <<-EOF
|
||
${c3} _
|
||
${c3} ---(_)
|
||
${c3} _/ --- \\
|
||
${c3}(_) | |
|
||
${c3} \\ --- _/
|
||
${c3} ---(_)
|
||
EOF
|
||
;;
|
||
|
||
[Vv]oid*)
|
||
read_ascii 2 <<-EOF
|
||
${c2} _______
|
||
${c2} _ \\______ -
|
||
${c2}| \\ ___ \\ |
|
||
${c2}| | / \ | |
|
||
${c2}| | \___/ | |
|
||
${c2}| \\______ \\_|
|
||
${c2} -_______\\
|
||
EOF
|
||
;;
|
||
|
||
*)
|
||
# On no match of a distribution ascii art, this function calls
|
||
# itself again, this time to look for a more generic OS related
|
||
# ascii art (KISS Linux -> Linux).
|
||
[ "$1" ] || {
|
||
get_ascii "$os"
|
||
return
|
||
}
|
||
|
||
printf 'error: %s is not currently supported.\n' "$os" >&6
|
||
printf 'error: Open an issue for support to be added.\n' >&6
|
||
exit 1
|
||
;;
|
||
esac
|
||
|
||
# Store the "width" (longest line) and "height" (number of lines)
|
||
# of the ascii art for positioning. This script prints to the screen
|
||
# *almost* like a TUI does. It uses escape sequences to allow dynamic
|
||
# printing of the information through user configuration.
|
||
#
|
||
# Iterate over each line of the ascii art to retrieve the above
|
||
# information. The 'sed' is used to strip '[3Xm' color codes from
|
||
# the ascii art so they don't affect the width variable.
|
||
while read -r line; do
|
||
# This was a ternary operation but they aren't supported in
|
||
# Minix's shell.
|
||
[ "${#line}" -gt "${ascii_width:-0}" ] &&
|
||
ascii_width=${#line}
|
||
|
||
# Using '<<-EOF' is the only way to loop over a command's
|
||
# output without the use of a pipe ('|').
|
||
# This ensures that any variables defined in the while loop
|
||
# are still accessible in the script.
|
||
done <<-EOF
|
||
$(printf %s "$ascii" | sed 's/\[3.m//g')
|
||
EOF
|
||
|
||
# Add a gap between the ascii art and the information.
|
||
ascii_width=$((ascii_width + 4))
|
||
}
|
||
|
||
main() {
|
||
# Hide 'stderr' unless the first argument is '-v'. This saves
|
||
# polluting the script with '2>/dev/null'.
|
||
[ "$1" = -v ] || exec 2>/dev/null
|
||
|
||
# Hide 'stdout' and selectively print to it using '>&6'.
|
||
# This gives full control over what it displayed on the screen.
|
||
exec 6>&1 >/dev/null
|
||
|
||
# Allow the user to execute their own script and modify or
|
||
# extend pfetch's behavior.
|
||
# shellcheck source=/dev/null
|
||
. "${PF_SOURCE:-/dev/null}" ||:
|
||
|
||
# Ensure that the 'TMPDIR' is writable as heredocs use it and
|
||
# fail without the write permission. This was found to be the
|
||
# case on Android where the temporary directory requires root.
|
||
[ -w "${TMPDIR:-/tmp}" ] || export TMPDIR=~
|
||
|
||
# Generic color list.
|
||
# Disable warning about unused variables.
|
||
# shellcheck disable=2034
|
||
{
|
||
c1='[31m'; c2='[32m'
|
||
c3='[33m'; c4='[34m'
|
||
c5='[35m'; c6='[36m'
|
||
c7='[37m'; c8='[38m'
|
||
}
|
||
|
||
# Avoid text-wrapping from wrecking the program output.
|
||
#
|
||
# Some terminals don't support these sequences, nor do they
|
||
# silently conceal them if they're printed resulting in
|
||
# partial sequences being printed to the terminal!
|
||
[ "$TERM" = dumb ] ||
|
||
[ "$TERM" = minix ] ||
|
||
[ "$TERM" = cons25 ] || {
|
||
# Disable line-wrapping.
|
||
printf '[?7l' >&6
|
||
|
||
# Enable line-wrapping again on exit.
|
||
trap 'printf [?7h >&6' EXIT
|
||
}
|
||
|
||
# Store the output of 'uname' to avoid calling it multiple times
|
||
# throughout the script. 'read <<EOF' is the simplest way of reading
|
||
# a command into a list of variables.
|
||
read -r os kernel arch <<-EOF
|
||
$(uname -srm)
|
||
EOF
|
||
|
||
# Always run 'get_os' for the purposes of detecting which ascii
|
||
# art to display.
|
||
get_os
|
||
|
||
# Allow the user to specify the order and inclusion of information
|
||
# functions through the 'PF_INFO' environment variable.
|
||
# shellcheck disable=2086
|
||
{
|
||
# Disable globbing and set the positional parameters to the
|
||
# contents of 'PF_INFO'.
|
||
set -f
|
||
set +f ${PF_INFO-ascii title os host kernel uptime pkgs memory}
|
||
|
||
# Grab the ascii art if the user has enabled it. This simply
|
||
# populates the '$ascii' variable.
|
||
case $@ in *ascii*) get_ascii; shift; esac
|
||
|
||
# Iterate over the info functions to determine the lengths of the
|
||
# "info names" for output alignment. The option names and subtitles
|
||
# match 1:1 so this is thankfully simple.
|
||
for info; do
|
||
command -v "get_$info" >/dev/null || continue
|
||
|
||
# This was a ternary operation but they aren't supported in
|
||
# Minix's shell.
|
||
[ "${#info}" -gt "${info_length:-0}" ] &&
|
||
info_length=${#info}
|
||
done
|
||
|
||
# Add an additional space of length to act as a gap.
|
||
info_length=$((info_length + 1))
|
||
|
||
while IFS= read -r line || [ "$1" ]; do
|
||
"get_$1"
|
||
shift "$(($# ? 1 : 0))"
|
||
printf '\033[%sC%s\r%s\n' "$ascii_width" "$out" "$line" >&6
|
||
out=
|
||
done <<-EOF
|
||
$ascii[m
|
||
EOF
|
||
}
|
||
}
|
||
|
||
main "$@"
|