Consider 'set -e' in lib/test-lib.sh

[kevina]

Without it simple typos may be hard to catch. For example, this trivial test script should fail:

#!/bin/sh
test_description="A trival test that should fail"
. lib/test-lib.sh
#set -e
test_expect_successs "false" 'false'
test_expect_success "true" 'true'
test_done

But it doesn't because I misspelled successs. Instead here is the output:

./t9999-example.sh: 8: ./t9999-example.sh: test_expect_successs: not found
ok 1 - true
# passed all 1 test(s)
1..1

It shows the error, but in a large script with lots of tests it may be very easy to miss.

If the set -e in uncommented than the script will abort and make the typo more obvious:

./t9999-example.sh: 5: ./t9999-example.sh: test_expect_successs: not found
FATAL: Unexpected exit with code 127

I am not sure what consequences using set -e globally will have though.

[kevina]

Note that set -e does not cause the script to abort if a test fails. For example if the script was (with both set -e and the typo fixed)

#!/bin/sh
test_description="A trival test that should fail"
. lib/test-lib.sh
set -e
test_expect_success "false" 'false'
test_expect_success "true" 'true'
test_done

the output will be:

not ok 1 - false
#       false
ok 2 - true
# failed 1 among 2 test(s)
1..2

[whyrusleeping]

mmm, i see. set -e doesnt apply like i thought it would because the test_expect_success commands are essentially running a subshell by calling eval and actually not returning a non-zero value even on test failure

[whyrusleeping]

cc @chriscool