Open CyberShadow opened 5 years ago
@CyberShadow -- looks like this is due to the fact that Bashcov exports SHELLOPTS=xtrace
into Bash's environment:
cat > child.sh <<'EOF'
#!/usr/bin/env bash
printf 1>&2 '%s: SHELLOPTS=%s\n' "${BASH_SOURCE[0]}" "$SHELLOPTS"
echo $unset
echo "Survived!"
EOF
cat > parent.sh <<'EOF'
#!/usr/bin/env bash
set -eu
printf 1>&2 '%s: SHELLOPTS=%s\n' "${BASH_SOURCE[0]}" "$SHELLOPTS"
./child.sh
EOF
chmod +x parent.sh child.sh
Works as expected when run without touching SHELLOPTS
:
$ ./parent.sh
./parent.sh: SHELLOPTS=braceexpand:errexit:hashall:interactive-comments:nounset
./child.sh: SHELLOPTS=braceexpand:hashall:interactive-comments
Survived!
Also works as expected when run with bash -x
:
$ bash -x ./parent.sh
+ set -eu
+ printf '%s: SHELLOPTS=%s\n' ./parent.sh braceexpand:errexit:hashall:interactive-comments:nounset:xtrace
./parent.sh: SHELLOPTS=braceexpand:errexit:hashall:interactive-comments:nounset:xtrace
+ ./child.sh
./child.sh: SHELLOPTS=braceexpand:hashall:interactive-comments
Survived!
But, when called with called with env SHELLOPTS=xtrace
:
$ env SHELLOPTS=xtrace ./parent.sh
+ set -eu
+ printf '%s: SHELLOPTS=%s\n' ./parent.sh braceexpand:errexit:hashall:interactive-comments:nounset:xtrace
./parent.sh: SHELLOPTS=braceexpand:errexit:hashall:interactive-comments:nounset:xtrace
+ ./child.sh
+ printf '%s: SHELLOPTS=%s\n' ./child.sh braceexpand:errexit:hashall:interactive-comments:nounset:xtrace
./child.sh: SHELLOPTS=braceexpand:errexit:hashall:interactive-comments:nounset:xtrace
./child.sh: line 3: unset: unbound variable
Interestingly, it looks like the issue crops up whenever the parent explicitly defines SHELLOPTS
, not just when it enables xtrace
:
$ env SHELLOPTS=emacs ./parent.sh
./parent.sh: SHELLOPTS=braceexpand:emacs:errexit:hashall:interactive-comments:nounset
./child.sh: SHELLOPTS=braceexpand:emacs:errexit:hashall:interactive-comments:nounset
./child.sh: line 3: unset: unbound variable
This behavior goes back a while:
matt@ratmaster /tmp/tmp.LXRydMzQPe :( $ for bv in 3.2 4.0 4.1 4.2 4.3 4.4 5.0; do printf 1>&2 -- '# %s\n' "$bv"; docker run -it --rm -v "${PWD}:/shellopts" --workdir /shellopts -u "$(id -u):$(id -g)" "bash:${bv}" env SHELLOPTS=emacs bash /shellopts/parent.sh; done
# 3.2
/shellopts/parent.sh: SHELLOPTS=braceexpand:emacs:errexit:hashall:interactive-comments:nounset
./child.sh: SHELLOPTS=braceexpand:emacs:errexit:hashall:interactive-comments:nounset
./child.sh: line 3: unset: unbound variable
# 4.0
/shellopts/parent.sh: SHELLOPTS=braceexpand:emacs:errexit:hashall:interactive-comments:nounset
./child.sh: SHELLOPTS=braceexpand:emacs:errexit:hashall:interactive-comments:nounset
./child.sh: line 3: unset: unbound variable
# 4.1
/shellopts/parent.sh: SHELLOPTS=braceexpand:emacs:errexit:hashall:interactive-comments:nounset
./child.sh: SHELLOPTS=braceexpand:emacs:errexit:hashall:interactive-comments:nounset
./child.sh: line 3: unset: unbound variable
# 4.2
/shellopts/parent.sh: SHELLOPTS=braceexpand:emacs:errexit:hashall:interactive-comments:nounset
./child.sh: SHELLOPTS=braceexpand:emacs:errexit:hashall:interactive-comments:nounset
./child.sh: line 3: unset: unbound variable
# 4.3
/shellopts/parent.sh: SHELLOPTS=braceexpand:emacs:errexit:hashall:interactive-comments:nounset
./child.sh: SHELLOPTS=braceexpand:emacs:errexit:hashall:interactive-comments:nounset
./child.sh: line 3: unset: unbound variable
# 4.4
/shellopts/parent.sh: SHELLOPTS=braceexpand:emacs:errexit:hashall:interactive-comments:nounset
./child.sh: SHELLOPTS=braceexpand:emacs:errexit:hashall:interactive-comments:nounset
./child.sh: line 3: unset: unbound variable
# 5.0
/shellopts/parent.sh: SHELLOPTS=braceexpand:emacs:errexit:hashall:interactive-comments:nounset
./child.sh: SHELLOPTS=braceexpand:emacs:errexit:hashall:interactive-comments:nounset
./child.sh: line 3: unset: unbound variable
Hypothesis: using env SHELLOPTS=xtrace
results in the equivalent of set -x; export SHELLOPTS
, causing child processes of Bash to inherit SHELLOPTS
. Further adjustments of SHELLOPTS
(e.g., set -eu
) are now inherited by child processes, too. So, revising ./parent.sh
to run export SHELLOPTS
:
cat > child.sh <<'EOF'
#!/usr/bin/env bash
printf 1>&2 '%s: SHELLOPTS=%s\n' "${BASH_SOURCE[0]}" "$SHELLOPTS"
echo $unset
echo "Survived!"
EOF
cat > parent.sh <<'EOF'
#!/usr/bin/env bash
set -eu
printf 1>&2 '%s: SHELLOPTS=%s\n' "${BASH_SOURCE[0]}" "$SHELLOPTS"
export SHELLOPTS
./child.sh
EOF
chmod +x parent.sh child.sh
Results in:
$ ./parent.sh
./parent.sh: SHELLOPTS=braceexpand:errexit:hashall:interactive-comments:nounset
./child.sh: SHELLOPTS=braceexpand:errexit:hashall:interactive-comments:nounset
./child.sh: line 3: unset: unbound variable
At present, there is not much Bashcov can do about this issue -- setting SHELLOPTS=xtrace
is essential to how it gathers coverage data for all shell scripts under test. A while back, I started working on a branch that globally enables executing tracing by setting BASH_ENV
to the path of a script that runs (more or less) set -x
, rather than by setting SHELLOPTS=xtrace
. I've left this effort on the back burner for the last several months, but may be able to find some time in the coming weeks to complete it.
Thanks. Would it make sense to make bashcov do the equivalent of bash -c "set -x ; source $script"
instead of env SHELLOPTS=xtrace $script
?
Would it make sense to make bashcov do the equivalent of bash -c "set -x ; source $script" instead of env SHELLOPTS=xtrace $script ?
I think this could be implemented as an optional execution mode, with a conspicuous note that using this mode may limit the number of scripts Bashcov is able to gather coverage data for, since (unlike env SHELLOPTS=xtrace
) set -x
doesn't globally enable execution tracing.
Consider these scripts:
The behavior is different depending on whether the script is run under bashcov or not: