Re: [PATCHv4 02/15] t4017 (diff-retval): replace manual exit code check with test_expect_code

["Ævar Arnfjörð Bjarmason" <avarab@gmail.com>]

On Fri, Oct 1, 2010 at 10:23, Jonathan Nieder <jrnieder@gmail.com> wrote:
> Ævar Arnfjörð Bjarmason wrote:
>> On Wed, Sep 29, 2010 at 18:07, Junio C Hamano <gitster@pobox.com> wrote:
>>> Elijah Newren <newren@gmail.com> writes:
>
>>>> -test_expect_success 'git diff-tree HEAD^ HEAD' '
>>>> +test_expect_code 1 'git diff-tree HEAD^ HEAD' '
>>>>       git diff-tree --exit-code HEAD^ HEAD
>>>> -     test $? = 1
>>>>  '
>>
>> It also looks like this will pass for for all exit codes that *aren't*
>> 1, because if $? != 1 +test_expect_code will get the exit code of
>> 1.
>
> You probably missed the - indicating that the "test $? = 1" was being
> removed.

Correct. I misread that.

> +check_exit_status () {
> +       echo "$1" >expect.status
> +       shift
> +       "$@"
> +       echo "$?" >actual.status
> +       test_cmp expect.status actual.status
> +}

If we add this it should be in the test-lib.sh, it'll probably be
useful for other tests.

>  test_expect_success 'setup' '
>        echo "1 " >a &&
>        git add . &&
> @@ -21,110 +29,81 @@ test_expect_success 'git diff --quiet -w  HEAD^^ HEAD^' '
>  '
>
>  test_expect_success 'git diff --quiet HEAD^^ HEAD^' '
> -       test_must_fail git diff --quiet HEAD^^ HEAD^
> +       check_exit_status 1 git diff --quiet HEAD^^ HEAD^
>  '

In most uses of check_exit_status you're using it is the very last
command within a test_expect_success. Isn't it redundant to using just
"test_expect_code $code ..." there?

>  test_expect_success 'check detects leftover conflict markers' '
> @@ -133,10 +112,8 @@ test_expect_success 'check detects leftover conflict markers' '
>        echo binary >>b &&
>        git commit -m "side" b &&
>        test_must_fail git merge master &&
> -       git add b && (
> -               git --no-pager diff --cached --check >test.out
> -               test $? = 2
> -       ) &&
> +       git add b &&
> +       check_exit_status 2 git --no-pager diff --cached --check >test.out &&
>        test 3 = $(grep "conflict marker" test.out | wc -l) &&
>        git reset --hard
>  '

But of course in cases like these it's needed.

In any case, we only use test_expect_code for two tests in the whole
test suite now, and checking the exit status for individual commands
is more self-documenting and less prone to break if we add more tests
to a given test.

So IMO the best thing to do would be to re-appropriate
"test_expect_code" so that it runs inside a test (i.e. does what your
check_exit_status does), and not at the top-level.

Then do s/check_exit_status/test_expect_code/g on your patch, and
change the tests using test_expect_code in t1504-ceiling-dirs.sh and
t6020-merge-df.sh to use test_expect_success + test_expect_code.