Test Writing Standards/Guidelines

Test Writing Standards/Guidelines

The following is a list of ways to improve existing tests as well as a guideline for creating new ones.

Always specify a plan

It is okay to use no_plan during test development, but once you are done, or even when you are sending patches/commiting to repositories, specify your plan.

Check your use statements with use_ok().

Wrap your use statements in a use_ok() inside a BEGIN block.

    BEGIN {
        use_ok( "MARC::Batch" ) or die;

The or die; modifier is not required but it can help avoid comfusion if a module only partially compiles and some tests fail even though the code is correct. Besides, if the module won't even compile, there's no point in testing it, right?

All object creation is checked with isa_ok().

If you instantiate an object, check it with isa_ok().

    my $batch = MARC::Batch->new( 'USMARC', @files );
    isa_ok( $batch, "MARC::Batch" );

Constructors are the most obvious, but any function that returns an object should be checked that the object is of the expected type.

    while ( my $marc = $batch->next ) {
        isa_ok( $marc, "MARC::Record" );
        # more tests...

In this case, the next() method should return a MARC::Record every time.

All temporary files are cleaned up.

Check the return code from unlink(), and then check that the file is actually gone.

    is( unlink( $filename ), 1, "Remove $filename" );

    ok( !-e $filename, "Actually gone" );

(This idiom is to be used in ideal cases. On VMS you might need to unlink() several times the same file until it's completely deleted; and the is() test above will succeed only if $filename was actually present, which might not be what you wanted to test.)

All tests should have a label

The last argument in all the Test::More functions is for the test's label, or description. If you write a test you should know what it is doing, so label it.

The only exception to this can be isa_ok and can_ok as they are pretty self explanatory and provide default messages for you.

Object Equivalence testing

All object equality tests are done with is:

    is( $object, $object2, '... our objects are equal' );

Object inequality tests can then be done with isnt:

    isnt( $object, $object2, '... our objects are not equal' );

The reasoning behind this is that we are really comparing the stringified instances, and is and isnt compare strings. If the objects' stringifications don't provide enough information to meaningfully compare the objects, which might happen if overloading the equality or stringification operators (either directly or indirectly), find something that does (e.g. use the function overload::StrVal()).

Use the right function for the job

We should pay attention to types with comparisons we are doing. For comparing strings, use is and isnt and use cmp_ok for any numeric comparisons or more complex comparisons not for equality. It only makes sense to use the right operator in the right situation.

This also means that tests that look like this:

    ok( $cost == $expected_cost );

should be rewritten as:

    is( $cost, $expected_cost );

or, even better, since these values are numeric:

    cmp_ok( $cost, '==', $expected_cost, 'Cost calculated correctly.' );

It makes the test much more explicit and takes advantage of is, isnt and cmp_ok's error messages.

Use is_deeply to test things other than scalars

When it is possible, we should check the contents of a data structure explicity, rather than the implicit testing of just element/key counts or other such techniques.

With is_deeply you can do things like:

    my @expected = qw( Larry Moe Curly );

    my @stooges = get_stooges();

    is_deeply( \@stooges, \@expected, "Got proper stooges" );

and if they differ, you'll get a message saying where they differ.

At times when is_deeply is not appropriate, Test::More also includes three utility functions eq_array, eq_hash and eq_set which return a boolean when called. These can be utilized as such:

    my @expected = qw( Larry Moe Curly );

    my @stooges = get_stooges();

    # use eq_set to test arrays regardless of their order

    ok(eq_set( \@stooges, \@expected), "Got proper stooges" );

is_deeply is not suitable for comparing objects that overload dereferences to their implementation type.

Test all your assumptions.

Assumptions that are tested after the fact, should also be tested before the fact. By this I mean that if you are assuming that after an operation completes a variable will be true, but prior to that it was false. Confirm the initial false-ness first, then run your code, and test the truth after. This may seem like overkill to some, but in fact the first test only serves to strengthen the second test.

Test code is still code

Avoid the (heavy) use of globals. You wouldn't do it in regular code, so why do it in tests?

Along this same line, constructs like this should be avoided:

    my $item = Object->new();
    ... test $item here

    $item = Object->new(@args);
    ... different tests for $item here

Re-use of variables like this should generally be avoided. It only serves to confuse the flow of the test. Ideally instances that are tested should be thought of as single assignment variables and never be re-used.

So now the above becomes:

    my $item = Object->new();
    ... test $item here

    my $item_w_args = Object->new(@args);
    ... test $item_w_args here

Or even better, if your instances don't interact with one another, put them in their own lexical scopes:

        my $obj = Object->new();

        my $obj = Object->new( foo => 'bar' );


Maintained by Andy Lester, with contributions from Stevan Little and Rafael Garcia-Suarez.