I have read through many questions on various stack exchange sites and unix help forums on how to modify shell options and then restore them - the most comprehensive one I found on here is at How to "undo" a `set -x`?
The received wisdom seems to be to either save off the result of
set +o or
shopt -po and then
eval it later to restore the previous settings.
However, in my own testing with bash 3.x and 4.x, the
errexit option does not get saved correctly when doing command substitution.
Here is an example script to show the issue:
set -o errexit set -o nounset echo "LOCAL SETTINGS:" set +o OLDOPTS=$(set +o) echo echo "SAVED SETTINGS:" echo "$OLDOPTS"
And output (I trimmed out some of the irrelevant variables):
LOCAL SETTINGS: set -o errexit set -o nounset SAVED SETTINGS: set +o errexit set -o nounset
This seems extremely dangerous. Most scripts I write depend on
errexit to halt execution if any commands fail. I just located a bug in one of my scripts caused by this, where the function that was supposed to restore
errexit at the end wound up overriding it, setting it back to the default of off for the duration of the script.
What I'd like to be able to do is write functions that can set options as needed and then restore all the options properly before exiting. But it seems as if in the subshell invoked by the command substitution,
errexit is not inherited.
I'm at a loss for how to save off the result of
set +o without using command substitution or jumping through FIFO hoops. I can read from
$SHELLOPTS but it is not writable or in
I know one alternative is to use a subshell function, but that introduces a lot of headaches for being able to log output as well as pass back multiple variables.
Probably related: https://stackoverflow.com/questions/29532904/bash-subshell-errexit-semantics (seems there is a workaround for bash 4.4 and up but I'd rather have a portable solution)