Re: [PATCH 1/1] t9117: prefer test_path_* helper functions

On Thu, Feb 29, 2024 at 10:46 PM shejialuo <shejialuo@gmail.com> wrote:

Thanks for rerolling. t9117 is indeed a better choice[1] than t3070 for the exercise of replacing `test -blah` with `test_path_foo`.

[1]: https://lore.kernel.org/git/CAPig+cR2-6qONkosu7=qEQSJa_fvYuVQ0to47D5qx904zW08Eg@mail.gmail.com/

Generally speaking, you don't want to use `!` to negate the result of a `path_is_foo` assertion function. To understand why, take a look at the definition of `test_path_is_dir`:

    test_path_is_dir () {
        if ! test -d "$1"
        then
            echo "Directory $1 doesn't exist"
            false
        fi
    }

The test in question (t9117: "basic clone") is using `test ! -d` to assert that the directory `trunk` does not yet exist when the test begins; indeed, under normal circumstances, this directory should not yet be present. However, the call to test_path_is_dir() asserts that the directory _does_ exist, which is the opposite of `test ! -d`, and complains ("Directory trunk doesn't exist") when it doesn't exist. So, in the normal and typical case for all the tests in this script, `test_path_is_dir` is going to be complaining even though the non-existence of that directory is an expected condition.

Although you make the test pass by using `!` to invert the result of `test_path_is_dir`, the complaint will nevertheless get lodged, and may very well be confusing for anyone scrutinizing the output of the tests when running the script with `-v` or `-x`.

So, `test_path_is_dir` is not a good fit for this case which wants to assert that the path `trunk` does not yet exist. A better choice for this particular case would be `test_path_is_missing`.

These two changes make sense and the intent directly corresponds to the original code.

What follows is probably beyond the scope of your GSoC microproject, but there is a bit more of interest to note about these tests.

Rather than asserting some initial condition at the start of the test, it is more common and more robust simply to _ensure_ that the desired initial condition holds. So, for instance, instead of asserting `test ! -d target`, modern practice is to ensure that `target` doesn't exist. Thus:

    test_expect_success 'clone to target directory' '
        rm -rf target &&
        git svn clone "$svnrepo"/project/trunk target &&
        ...

is a more robust implementation. This also addresses the problem that the `rm -rf target` at the very end of each test won't be executed if any command earlier in the test fails (due to the short-circuiting behavior of the &&-operator).

As noted, this type of cleanup is probably overkill for your GSoC microproject so you need not tackle it. I mention it only for completeness. Also, if someone does tackle such a cleanup, it should be done as multiple patches, each making one distinct change (i.e. one patch dropping `test !-d` and moving `rm -rf` to the start of the test, and one which employs `test_path_foo` for the remaining `test -blah` invocations).