On 03/09/2017 02:48 AM, Ville Skyttä wrote:
> I don't know what the reason for making the variable read only is. But
> I think you could work around it by setting BASH_COMPLETION_COMPAT_DIR
> to a fake value, e.g. /prevent/sourcing in your rc files before the
> profile.d snippet is sourced (thus preventing it from loading
> bash_completion), and then unsetting it before "manually" loading the
> profile.d snippet where you want, and let it do its job unmodified.
Ville,
Problem here is that bash invokes /etc/profile prior to the end-user having
control via .bashrc. All that stuff in /etc/profile.d/bash_completion.sh is
performed outside my control. (i prefer not to mangle up too many
system-provided files, or affect my user's expectations of the environment)
It might be possible to set it via 'pam_env' or something (which *should* take
effect prior to the shell invoking /etc/profile) That's a little messy, and i'd
prefer not to have to bring yet another component into the picture.
But, let's do a proof-of-concept:
# cat ~/.pam_environment
BASH_COMPLETION_COMPAT_DIR="BOGUS"
# cat /root/.bashrc
.....
\unset -f unalias ; \unalias -a ; \unset -f command
\unset LD_LIBRARY_PATH LD_PRELOAD SHELLOPTS CDPATH HISTFILE HOSTFILE
INPUTRC 2>/dev/null
confpath="$(command -p getconf PATH 2>/dev/null)"
# Establish baseline path with no NFS components
export PATH="/opt/sbin:/opt/bin:/sbin:/usr/sbin:${confpath:-/bin:/usr/bin}"
\type typeset >/dev/null && \unset -f $(\typeset -f | \grep '()' | \cut -d'
' -f1)
.....
is_shell_interactive() { [ "${-%i*}" != "${-}" ] ;}
.....
if is_shell_interactive; then
# RELOAD bash-completion
echo "DEBUG:
BASH_COMPLETION_COMPAT_DIR[before]=${BASH_COMPLETION_COMPAT_DIR}" 1>&2
unset BASH_COMPLETION_COMPAT_DIR
. /etc/profile.d/bash_completion.sh
echo "DEBUG:
BASH_COMPLETION_COMPAT_DIR[after]=${BASH_COMPLETION_COMPAT_DIR}" 1>&2
Results in, when i remote 'ssh' login:
DEBUG: BASH_COMPLETION_COMPAT_DIR[before]="BOGUS"
DEBUG: BASH_COMPLETION_COMPAT_DIR[after]=/etc/bash_completion.d
# set | grep longopt\ \(
_longopt ()
_split_longopt ()
So, indeed, that *will* work -- presuming pam_env PAM enabled services (my
ssh+su+login config at least uses it).
Thanks for the idea. That gets me a drop on doing this now (other than what i
was doing, which was just invoking the two command lines that
/etc/profile.d/bash_completion.sh script was doing inside the conditional
block). I think it still would be conceptually better to not have to rely on
pam_env for this to work, and have bash-completion implement some re-sourcable
invocation mechanism.
--stephen
