On 6/9/25 7:05 PM, Shinichiro Kawasaki wrote:
On Jun 09, 2025 / 10:40, Bart Van Assche wrote:
Has the following alternative been considered? Let test script authors add
"set -e" and "set +e" where appropriate but only in a subshell. If a failure
occurs, only the subshell will be exited and cleanup code will still be
executed if it occurs past the subshell. A disadvantage of this
approach is that global variables can't be set from inside the subshell
and that another mechanism is needed than local variables to pass output
from the subshell to the context outside it, e.g. a pipe.
This idea sounds something to consider, but I'm not sure if I fully understand
it. The word "subshell" is not clear for me. Do you mean subshells those test
case authors create in each test script? If so I'm not sure how to ensure that
"set -e" and "set +e" only happen in the subshells. Or do you mean to modify the
check script to create subshells dedicated for each test case run? If so, I will
need some work to understand its impact.
From https://www.gnu.org/software/bash/manual/bash.html:
<quote>
( list )
Placing a list of commands between parentheses forces the shell to
create a subshell (see Command Execution Environment), and each of the
commands in list is executed in that subshell environment. Since the
list is executed in a subshell, variable assignments do not remain in
effect after the subshell completes.
</quote>
Here is an example that shows how a subshell can be used to halt a test
with "set -e" if a failure occurs in such a way that error handling is
still executed:
$ bash -c '(set -e; false; echo "Skipped because the previous command
failed"); echo "Error handling commands outside the subshell are still
executed"'
Error handling commands outside the subshell are still executed
@@ -372,6 +380,7 @@ _call_test() {
fi
TIMEFORMAT="%Rs"
+ ((ERR_EXIT)) && set -e
pushd . >/dev/null || return
{ time "$test_func" >"${seqres}.out" 2>&1; } 2>"${seqres}.runtime"
TEST_RUN["exit_status"]=$?
This change makes it harder to write test code because it forces authors
to surround cleanup code with something like if cleanup_code; then :; fi.
I see, this point makes sense. I'm okay to not call "set -e" in the
_call_test(). If we take this approach, test cases can do "set -e" and "set +e"
wherever in the test scripts, but they must declare ERR_EXIT=1.
Do you agree that with the style of the above example ERR_EXIT is not
needed at all?
Thanks,
Bart.
