6.8 Writing Tests
As we mentioned earlier, contributions to the Parrot project are welcome. Contributing tests is a good place to start. You don't have to understand the code behind a PASM opcode to test it, you only have to understand what it's supposed to do. If you're working on some code and it doesn't do what the documentation advertises, you can isolate the problem in a test or series of tests and send them to the bug tracking system. There's a good chance the problem will be fixed before the next release. Writing tests makes it a lot easier for the developer to know when they've solved your problem—it's solved when your tests pass. It also prevents that problem from appearing again, because it's checked every time anyone runs make test. As you move along, you'll want to write tests for every bug you fix or new feature you add.
The Perl 5 testing framework is at the core of Parrot tests, particularly Test::Builder. Parrot's Parrot::Test module is an interface to Test::Builder and implements the extra features needed for testing Parrot, like the fact that PASM code has to be compiled to bytecode before it runs.
The main Parrot tests are in the top-level t/ directory of the Parrot source tree. t/op contains tests for basic opcodes and t/pmc has tests for PMCs. The names of the test files indicate the functionality tested, like integer.t, number.t, and string.t. Part of the make test target is the command perl t/harness, which runs all the .t files in the subdirectories under /t. You can run individual test files by passing their names to the harness script:
$ perl t/harness t/op/string.t t/op/integer.t
output_is(<<CODE, <<OUTPUT, "set_i"); set I0, 42 set I1, I0 print I1 print "\\n" end CODE 42 OUTPUT
The code here sets integer register I0 to the value 42, sets I1 to the value of I0, and then prints the value in I1. The test passes if the value printed was 42, and fails otherwise.
The output_is subroutine takes three strings: the code to run, the expected output, and a description of the test. The first two strings can be quite long, so the convention is to use Perl 5 here-documents. If you look into the code section, you'll see that the literal \n has to be escaped as \\n. Many tests use the non-interpolating (<<'CODE') form of here-document to avoid that problem. The description can be any text. In this case, it's the fully qualified name of the set opcode for integer registers, but it could have been "set a native integer register."
If you look up at the top of integer.t, you'll see the line:
use Parrot::Test tests => 38;
The use line for the Parrot::Test module imports a set of subroutines into the test file, including output_is. The end of the line gives the number of tests contained in the file.
The output_is subroutine looks for an exact match between the expected result and the actual output of the code. When the test result can't be compared exactly, you want output_like instead. It takes a Perl 5 regular expression for the expected output:
output_like(<<'CODE', <<'OUTPUT', "testing for text match"); ... CODE /^Output is some \d+ number\n$/ OUTPUT
Parrot::Test also exports output_isnt, which tests that the actual output of the code doesn't match a particular value.
There are a few guidelines to follow when you're writing a test for a new opcode or checking that an existing opcode has full test coverage. Tests should cover the opcode's standard operation, corner cases, and illegal input. The first tests should always cover the basic functionality of an opcode. Further tests can cover more complex operations and the interactions between opcodes. If the test program is complex or obscure, it helps to add comments. Tests should be self-contained to make it easy to identify where and why a test is failing.