As stated in other answers, when you run a script, it creates a sub-shell.
When the script exits, all modifications to that shell are lost.
What we need is actually to run a new shell where the virtual environment is active, and not exit from it.
Be aware, this is a new shell, not the one in use before you run your script.
What this mean is, if you type exit in it, it will exit from the subshell, and return to the previous one (the one where you ran the script), it won't close your xterm or whatever, as you may have expected.
The trouble is, when we exec bash, it reads its rc files (/etc/bash.bashrc, ~/.bashrc), which will change the shell environment. The solution is to provide bash with a way to setup the shell as usual, while additionnally activate the virtual environment. To do this, we create a temporary file, recreating the original bash behavior, and adding a few things we need to enable our venv. We then ask bash to use it instead of its usual rc files.
A beneficial side-effect of having a new shell "dedicated" to our venv, is that to deactivate the virtual environment, the only thing needed is to exit the shell.
I use this in the script exposed below to provide a 'deactivate' option, which acts by sending a signal to the new shell (kill -SIGUSR1), this signal is intercepted (trap ...) and provoke the exit from the shell.
Note: i use SIGUSR1 as to not interfere with whatever could be set in the "normal" behavior.
The script i use:
#!/bin/bash
PYTHON=python3
myname=$(basename "$0")
mydir=$(cd $(dirname "$0") && pwd)
venv_dir="${mydir}/.venv/dev"
usage() {
printf "Usage: %s (activate|deactivate)\n" "$myname"
}
[ $# -eq 1 ] || { usage >&2; exit 1; }
in_venv() {
[ -n "$VIRTUAL_ENV" -a "$VIRTUAL_ENV" = "$venv_dir" -a -n "$VIRTUAL_ENV_SHELL_PID" ]
}
case $1 in
activate)
# check if already active
in_venv && {
printf "Virtual environment already active\n"
exit 0
}
# check if created
[ -e "$venv_dir" ] || {
$PYTHON -m venv --clear --prompt "venv: dev" "$venv_dir" || {
printf "Failed to initialize venv\n" >&2
exit 1
}
}
# activate
tmp_file=$(mktemp)
cat <<EOF >"$tmp_file"
# original bash behavior
if [ -f /etc/bash.bashrc ]; then
source /etc/bash.bashrc
fi
if [ -f ~/.bashrc ]; then
source ~/.bashrc
fi
# activating venv
source "${venv_dir}/bin/activate"
# remove deactivate function:
# we don't want to call it by mistake
# and forget we have an additional shell running
unset -f deactivate
# exit venv shell
venv_deactivate() {
printf "Exitting virtual env shell.\n" >&2
exit 0
}
trap "venv_deactivate" SIGUSR1
VIRTUAL_ENV_SHELL_PID=$$
export VIRTUAL_ENV_SHELL_PID
# remove ourself, don't let temporary files laying around
rm -f "${tmp_file}"
EOF
exec "/bin/bash" --rcfile "$tmp_file" -i || {
printf "Failed to execute virtual environment shell\n" >&2
exit 1
}
;;
deactivate)
# check if active
in_venv || {
printf "Virtual environment not found\n" >&2
exit 1
}
# exit venv shell
kill -SIGUSR1 $VIRTUAL_ENV_SHELL_PID || {
printf "Failed to kill virtual environment shell\n" >&2
exit 1
}
exit 0
;;
*)
usage >&2
exit 1
;;
esac
