It depended on specific error messages to detect failure but the
implementation changed and broke the test. This fixes the breakage
minimally.
Signed-off-by: Junio C Hamano <junkio@cox.net>
test_expect_success \
"fail to create $n" \
"touch .git/$n_dir
- git-update-ref $n $A >out 2>err
- test "'$? = 1 &&
- test "" = "$(cat out)" &&
- grep "error: unable to resolve reference" err &&
- grep '"$n err"
+ git-update-ref $n $A >out 2>err"'
+ test $? != 0'
rm -f .git/$n_dir out err
test_expect_success \