apenwarr-redo/t/shelltest.od
Avery Pennarun 1f64cc4525 shelltest.od: add more "set -e" tests and add a 'skip' return code.
Based on the earlier t/000-set-minus-e bug in minimal/do on some
shells, let's add some extra tests that reveal the weirdness on those
shells.  Unfortunately because they are so popular (including bash and
zsh), we can't reject them outright for failing this one.

While we're here, add a new return code, "skip", which notes that a
test has failed but is not important enough to be considered a warning
or failure.  Previously we just had these commented out, which is not
quite obvious enough.

...and I updated a few comments while reviewing some of the older
tests.
2018-12-17 16:17:37 +00:00

479 lines
11 KiB
Text

#
# Most of these tests were inspired by:
# http://www.gnu.org/software/hello/manual/autoconf/Shell-Substitutions.html
#
# Note that this file isn't exactly a test for POSIX compliance. It's a test
# for usefulness-compliance, that is, interpreting POSIX in a particular way
# for consistency, so that users of redo can depend on all the following
# functionality.
#
# Where POSIX makes a clear statement about something working a particular
# way, POSIX wins. But where POSIX is unclear on a point, I don't mind
# picking a particular way and requiring it here, if that makes writing
# scripts easier for everyone in the future.
#
exec >&2
set +e
rm -f shelltest.failed shelltest.warned
fail()
{
echo " failed: $1"
: >shelltest.failed
}
warn()
{
echo " warning: $1"
: >shelltest.warned
}
# For things that should be a warning, but which are just too common
skip()
{
echo " skip: $1"
}
quiet_stderr()
{
if [ -n "$SHELLTEST_QUIET" ]; then
"$@" 2>/dev/null
else
"$@"
fi
}
name=foo.o.o
ext=.o
[ "${name#foo.o}" = ".o" ] || fail 3
spacey="this has * and spaces"
case $spacey in
*) spaceout=$name$spacey ;;
esac
[ "$spaceout" = "$name$spacey" ] || fail 4
n() { echo "$#$@"; }
f=" - "
out=$(n - ""$f"" -)
[ "$out" = "5- - -" ] || warn 5
n1() { echo $#; }
n2() { n1 "$@"; }
t1=$(n1)
t2=$(n2)
[ "$t1" = "0" ] || fail 6
[ "$t2" = "0" ] || fail 7
n1() { for i in "$@"; do echo $i; done; }
n2() { for i in ${1+"$@"}; do echo $i; done; }
t1=$(n1 "Hello World" "!")
t2=$(n2 "Hello World" "!")
WANT="Hello World
!"
[ "$t1" = "$WANT" ] || fail 8
[ "$t2" = "$WANT" ] || fail 9
n() { echo ${10}; }
t1=$(n 1 2 3 4 5 6 7 8 9 xx yy)
[ "$t1" = "xx" ] || fail 10
chicken1=`echo " $spacey" | sed s/a/./g`
chicken2="`quiet_stderr echo " $spacey" | sed s/a/./g`"
chicken3=$(echo " $spacey" | sed s/a/./g)
chicken4="$(echo " $spacey" | sed s/a/./g)"
[ "$chicken1" = " this h.s * .nd sp.ces" ] || fail 11
[ "$chicken2" = " this h.s * .nd sp.ces" ] || fail 12
[ "$chicken3" = " this h.s * .nd sp.ces" ] || fail 13
[ "$chicken4" = " this h.s * .nd sp.ces" ] || fail 14
f1=
f2=goo
g1=
g2=goo
out=$(echo ${f1:-foo} ${f2:-foo} ${g1:=foo} ${g2:=foo})
: ${f1:-roo} ${f2:-roo} ${g1:=roo} ${g2:=roo}
[ "$out" = "foo goo foo goo" ] || fail 16
[ "$f1$f2$g1$g2" = "gooroogoo" ] || fail 17
# This is really a test that ${x:=} can span multiple lines without requiring
# quoting, which is apparently required by POSIX but not supported in some
# shells. But to make ash/dash not abort the test script *entirely* when it
# fails, we use the $() + eval nonsense.
f3=$(quiet_stderr eval ': ${f3:=
a
b
}' && echo "$f3")
g3="
a
b"
# This is kind of a major problem, but rejecting dash and busybox sh because
# of this may cause more trouble than it's worth: people end up writing .do
# scripts with bashisms unnecessarily :(
[ "$f3" = "$g3" ] || warn 18
# Note: assignment of $@ in this context is unspecified (what do you even expect
# it to do?) so we don't test it. But we should expect $* to work.
set "a b" "c d" "*"
t1=$*
[ "$t1" = "a b c d *" ] || fail 19
unset a
t1=$(echo ${a-b c})
t2=$(echo ${a-'b c'})
t3=$(echo "${a-b c}")
t4=$(echo "${a-"b c"}")
t5=$(cat <<EOF
${a-b c}
EOF
)
[ "$t1" = "b c" ] || fail 21
[ "$t2" = "b c" ] || fail 22
[ "$t3" = "b c" ] || fail 23
[ "$t4" = "b c" ] || fail 24 # freebsd 11.2 sh fails this
[ "$t5" = "b c" ] || fail 25
unset a
t1=$(echo ${a=b c})
t2=$(echo ${a='b c'})
t3=$(echo "${a=b c}")
t4=$(echo "${a="b c"}")
t5=$(cat <<EOF
${a=b c}
EOF
)
[ "$t1" = "b c" ] || fail 31
[ "$t2" = "b c" ] || warn 32
[ "$t3" = "b c" ] || fail 33
[ "$t4" = "b c" ] || fail 34
[ "$t5" = "b c" ] || fail 35
unset foo
foo=${foo='}'} # unconfuse syntax highlighting: '
[ "$foo" = "}" ] || fail 41
foo=${foo='}'} # unconfuse syntax highlighting: '
[ "$foo" = "}" ] || fail 42
default="yu,yaa"
unset var
: ${var="$default"}
t1=$(cat <<EOF
$var
EOF
)
[ "$t1" = "yu,yaa" ] || fail 43
default="a b c"
unset list1 list2
: ${list1="$default"} ${list2=$default}
t1=$(for c in $list1; do echo $c; done)
t2=$(for c in $list2; do echo $c; done)
WANT="a
b
c"
[ "$t1" = "$WANT" ] || fail 44
[ "$t2" = "$WANT" ] || fail 45
# Arguably, 'export' and 'local' shouldn't change variable assignment
# quoting rules, but in many shells they do, and POSIX doesn't say anything
# about it. This leads to really confusing whitespace bugs in some shells,
# where
# local x; x=$a
# and
# local x=$a
# mean two different things when $a contains whitespace.
bob="a b *"
bob=$(quiet_stderr eval 'export bob=$bob:hello'; echo "$bob")
[ "$bob" = "a b *:hello" ] || skip 46
bob="a b *"
nob=$(eval 'f() { local nob=$bob:hello; echo "$nob"; }'; quiet_stderr f)
[ "$nob" = "a b *:hello" ] || skip 47
# Someone pointed out that temporary variable assignments aren't
# temporary anymore, if the thing you're calling is a function or builtin.
f() { ls >/dev/null; }
g=1 h=1 i=1
g=2 f
h=2 :
i=2 ls >/dev/null
[ "$g" = "2" ] || warn 48 # freebsd dash 0.5.10.2 fails this
[ "$h" = "2" ] || fail 49
[ "$i" = "1" ] || fail 50
var='a a b b'
var2='*a'
t1=${#var}
t2=${var#a* }
t3=${var##a* }
t4=${var%b*}
t5=${var%%b*}
t6="${var2#'*'}"
t7="${var2#"*"}"
[ "$t1" = "7" ] || fail 51
[ "$t2" = "a b b" ] || fail 52
[ "$t3" = "b" ] || fail 53
[ "$t4" = "a a b " ] || fail 54
[ "$t5" = "a a " ] || fail 55
# shells I tested work with #"*" but dash doesn't work with #'*' . The
# workaround is to simply use double quotes, so I'll make that one a warning.
[ "$t6" = "a" ] || warn 56
[ "$t7" = "a" ] || fail 56
in="a
b
"
t1=$(echo "$in" | tr a A)
t2="$(echo "$in" | tr a A)"
[ "$t1" = "A
b" ] || exit 57
[ "$t2" = "A
b" ] || exit 58
echo "`printf 'foo\r\n'` bar" >broken
echo "`printf 'foo\r\n'`"" bar" | diff -q - broken || fail 59
#
# This one is too obnoxious. dash and ash pass the test, but most shells
# don't, and this case is just too dumb to care about. Just don't do that!
#
t=`echo $(case x in x) echo hello;; esac)`
[ "$t" = "hello" ] || skip 60
#
# Note that with the little-known optional left-paren, this *does* work.
# Let's try it to make sure that remains true.
t=`echo $(case x in (x) echo hello;; esac)`
[ "$t" = "hello" ] || fail 60
x=5
t1=$(($x + 4))
t2=$(echo $(( 010 + 0x10 )))
[ "$t1" = "9" ] || fail 61
[ "$t2" = "24" ] || fail 62
t=$(echo hello ^ cat)
[ "$t" = "hello ^ cat" ] || fail 65
t1=$(for d in this-glob-does-*-not-exist; do echo "$d"; done)
t2=$(for d in this-glob-does-*-not-exist; do echo "$d"; done)
# http://www.gnu.org/software/hello/manual/autoconf/Assignments.html
false || foo=bar; [ "$?" = 0 ] || fail 71
foo=`exit 1`; [ "$?" != 0 ] || fail 72
# http://www.gnu.org/software/hello/manual/autoconf/Shell-Functions.html
f1() { echo 1; }
f2(){ echo 2;}
f3()(echo 3)
f4()if true; then echo 4; fi
f5() ( exit 5 )
[ "$(f1)" = 1 ] || fail 81
[ "$(f2)" = 2 ] || fail 82
[ "$(f3)" = 3 ] || fail 83
[ "$(f4)" = 4 ] || fail 84
f5 && fail 85
f6() (
f6b() { return 1; }
set -e
f6b
fail 86
)
f6
f7() { :; }; f7=; f7 || fail 87
a=
f8() { echo $a; };
t8=$(a=1 f8)
[ "$t8" = "1" ] || fail 88
# Some versions of zsh (eg. 5.0.8 on MacOS, but not 5.3.1
# on Debian) fail this one. It seems to be very specific:
# "set -e" aborts the subshell, if a
# function-within-a-function returns an error code, even
# though there's an "|| true" there, but only if in the
# else clause, and only if the outer () does *not* have an
# "|| something" clause.
#
# "x || true" should not abort when set -e, even if x is
# a function.
(
set -e
f() { false; }
g() { true | f; }
if false; then
:
else
! g || true
fi
)
rv=$?
[ "$rv" = 0 ] || fail 89
rm -f shellfail
echo "false; true" >shellfail
# 'false' should *not* abort because outer context has "||".
# Then 'true' gives success.
set +e
(
set -e
false
true
) || fail 89a
# 'false' still should not abort even in a subshell.
set +e
(
set -e
( false; true )
[ "$?" = 0 ] || fail 89b1
true
) || fail 89b
# Check whether "." scripts inherit "set -e" setting *and* the outer "||"
# context.
set +e
(
set -e
( . ./shellfail )
# There seems to be widespread disagreement on what should happen
# here. All shells seem to inherit "set -e" into a "." script, but
# they disagree about whether the outer "||" context (which
# effectively disables "set -e") should be inherited. ksh93, bash,
# and zsh inherit it, while ash, dash, and mksh (among others) throw
# it away, causing "set -e" to abort.
#
# Honestly, the latter behaviour seems a lot more like what I'd
# expect (a separate script doesn't want its "set -e" behaviour
# to depend on whether it's deep in a context with an "||" somewhere
# outside it), so let's warn on the former. But I'm willing to hear
# arguments either way. -- apenwarr
[ "$?" = 0 ] && warn 89c
# even if "." aborted and returned nonzero, our own context should
# not have aborted, so we expect to get here and avoid the outer
# fail below.
true
) || fail 89c1
# http://www.gnu.org/software/hello/manual/autoconf/Limitations-of-Builtins.html
. /dev/null || fail 90
(! : | :) && fail 91 || true
(! { :; }) && fail 92 || true
t3=none
case frog.c in
(*.c) t3=c ;;
(*) t3=all ;;
esac
[ "$t3" = "c" ] || fail 93
t4=$(echo '\n' | wc -l) # echo should *not* be parsing backslash-escapes
[ "$t4" -eq 1 ] || warn 94
f5() {
for arg; do
echo $arg
done
}
t5=$(f5 a=5 b c)
[ "$t5" = "a=5
b
c" ] || fail 95
t6=$(printf -- '%d %d' 5 6)
[ "$t6" = "5 6" ] || fail 96
echo 'word\ game stuff' >shellfile
read t7a t7b <shellfile
read -r t8a t8b <shellfile
[ "$t7a" = 'word game' ] || fail 97
[ "$t7b" = 'stuff' ] || fail 97
[ "$t8a" = 'word\' ] || fail 98
[ "$t8b" = 'game stuff' ] || fail 98
test -e shellfile || fail 99
[ "-a" = "-a" ] || fail 100
[ "-a" != "-b" ] || fail 101
[ "-a" -a "!" ] || fail 102
(unset foo && unset foo && unset foo) || fail 103
# http://www.gnu.org/software/hello/manual/autoconf/Limitations-of-Usual-Tools.html
rm -f || fail 110
# reported by singpolyma in busybox-w32's issue tracker
(
set -e
HOME="$PWD/space home dir"
rm -rf "$HOME"
mkdir "$HOME"
quiet_stderr cd ~ || exit 1
cd ..
rmdir 'space home dir'
) || warn 111
# reported by Alex Bradbury:
# http://www.debian.org/doc/debian-policy/ch-files.html#s-scripts
# Debian says that 'local' isn't POSIX, but I like it, so let's reject
# any shell where it doesn't exist.
x=5 y=6 z=7
lt() { x=1; local y=2 z=3; }
lt
[ "$x" = "1" ] || fail 112
[ "$y$z" = "67" ] || fail 113
# reported by Tim Allen:
# some shells don't return 0 from an empty '.' script; they return the
# previous command's exit code instead.
# See:
# http://pubs.opengroup.org/onlinepubs/009695399/utilities/dot.html
# http://austingroupbugs.net/view.php?id=114
false
. ./nothing.od || warn 114
# this is actually a bash/kshism, but is allowed by POSIX: the parameters to
# '.' should be passed to the sub-script.
#
# We used to require this, because it's so useful despite being optional in
# POSIX. But unfortunately, too many shells (including dash) can't do it,
# so we ended up always using bash, which leads people to write .do scripts
# with bashisms.
set x y z
# dotparams.od might warn 115
. ./dotparams.od a b || fail 117
[ "$1-$2-$3" = "x-y-z" ] || fail 116
# Warn about the way bash allows '==' in its test command.
# This is the #1 bashism I've observed, and totally unnecessary, so let's
# dock it some points so it doesn't always end up as the primary shell.
[ 1 == 1 ] 2>/dev/null && warn 118
# Some shells apparently don't support left padding in printf
x=$(printf "a%-5sc" "b")
[ "$x" = "ab c" ] || warn 119
[ -e shelltest.failed ] && exit 41
[ -e shelltest.warned ] && exit 42
exit 40