Writing APR tests

All APR tests should be executable in 2 ways, as an individual program, or
as a part of the full test suite.  The full test suite is controlled with
the testall program.  At the beginning of the testall.c file, there is an
array of functions called tests.  The testall program loops through this 
array calling each function.  Each function returns a CuSuite variable, which
is then added to the SuiteList.  Once all Suites have been added, the SuiteList
is executed, and the output is printed to the screen.  All functions in the
array should follow the same basic format:

The Full Suite
--------------

/* The driver function.  This must return a CuSuite variable, which will
 * then be used to actually run the tests.  Essentially, all Suites are a
 * collection of tests.  The driver will take each Suite, and put it in a
 * SuiteList, which is a collection of Suites.
 */
CuSuite *testtime(void)
{
    /* The actual suite, this must be created for each test program.  Please
     * give it a useful name, that will inform the user of the feature being
     * tested.
     */
    CuSuite *suite = CuSuiteNew("Test Time");

    /* Each function must be added to the suite.  Each function represents
     * a single test.  It is possible to test multiple features in a single
     * function, although no tests currently do that.
     */
    SUITE_ADD_TEST(suite, test_now);
    SUITE_ADD_TEST(suite, test_gmtstr);
    SUITE_ADD_TEST(suite, test_localstr);
    SUITE_ADD_TEST(suite, test_exp_get_gmt);
    SUITE_ADD_TEST(suite, test_exp_get_lt);
    SUITE_ADD_TEST(suite, test_imp_gmt);
    SUITE_ADD_TEST(suite, test_rfcstr);
    SUITE_ADD_TEST(suite, test_ctime);
    SUITE_ADD_TEST(suite, test_strftime);
    SUITE_ADD_TEST(suite, test_strftimesmall);
    SUITE_ADD_TEST(suite, test_exp_tz);
    SUITE_ADD_TEST(suite, test_strftimeoffset);

    /* You must return the suite so that the driver knows which suites to
     * run.
     */
    return suite;
}

Building the full driver
------------------------

All you need to do to build the full driver is run:

    make

To run it, run:

    ./testall

Running individual tests
------------------------

It is not possible to build individual tests, however it is possible to
run individual tests.  When running the test suite, specify the name of the
tests that you want to run on the command line.  For example:

	./testall teststr testrand

Will run the Strings and Random generator tests.

Reading the test suite output
-----------------------------

Once you run the test suite, you will get output like:

All APR Tests:
    Test Strings:       ....
    Test Time:  ............

16 tests run:  16 passed, 0 failed, 0 not implemented.

Known test failures are documented in ../STATUS.

There are a couple of things to look at with this.  First, if you look at the
first function in this document, you should notice that the string passed to
the CuSuiteNew function is in the output.  That is why the string should
explain the feature you are testing.

Second, this test passed completely.  This is obvious in two ways.  First, and
most obvious, the summary line tells you that 16 tests were run and 16 tests
passed.  However, the results can also be found in the lines above.  Every
'.' in the output represents a passed test.

If a test fails, the output will look like:

All APR Tests:
    Test Strings:       ....
    Test Time:  ..F.........

16 tests run:  15 passed, 1 failed, 0 not implemented.

This is not very useful, because you don't know which test failed.  However,
once you know that a test failed, you can run the suite again, with the
-v option.  If you do this, you will get something like:

All APR Tests:
    Test Strings:       ....
    Test Time:  ..F.........

16 tests run:  15 passed, 1 failed, 0 not implemented.
Failed tests:
1) test_localstr: assert failed

In this case, we know the test_localstr function failed, and there is an
Assert in this that failed (I modified the test to fail for this document).
Now, you can look at what that test does, and why it would have failed.

There is one other possible output for the test suite (run with -v):

All APR Tests:
    Test Strings:       ....
    Test Time:  ..N.........

16 tests run:  15 passed, 0 failed, 1 not implemented.

Not Implemented tests:

Not Implemented tests:
1) test_localstr: apr_time_exp_lt not implemented on this platform

The 'N' means that a function has returned APR_ENOTIMPL.  This should be 
treated as an error, and the function should be implemented as soon as
possible.

Adding New test Suites to the full driver
-------------------------------------------

To add a new Suite to the full driver, you must make a couple of modifications.

1)  Edit test_apr.h, and add the prototype for the function.
2)  Edit testall.c, and add the function and name to the tests array.
3)  Edit Makefile.in, and add the .lo file to the testall target.

Once those four things are done, your tests will automatically be added
to the suite.

Writting an ABTS unit test
--------------------------

The aim of this quick and dirty Howto is to give a short introduction
to APR (Apache Portable Runtime) unit tests, and how to write
one. During my Google's Summer of Code 2005 project, I discovered a
small bug in the APR-Util's date parsing routines, and I needed to
write a unit test for the fixed code. I decided to write this
documentation because I did not find any. Thanks to Garrett Rooney for
his help on writing the unit test !

The APR and APR-Util libraries provide a platform independent API for
software developers. They contain a lot of modules, including network
programming, threads, string and memory management, etc. All these
functions need to be heavily tested so that developers can be sure the
library is reliable.

The ABTS give APR developers the ability to build a complete test
suite for the bunch of tests they wrote, which can then be ran under
various platforms. In this Howto, I will try teach you how to write an
ABTS unit test.

As you may probably know, a unit test is a simple routine which tests
a very specific feature of the tested software or library. To build a
unit test, you need three different things :

 * the to-be-tested function,
 * the input data that will be given to the function,
 * the expected output data.

The principle of a unit test is very simple : for each entry in your
set of input data, we pass it to our function, fetch what the function
returned and compare it to the corresponding expected output data. Of
course, the more edge cases you can test, the better your input data
set is.

The ABTS aims to quicken the write of unit test, and make them
available to the whole test suite by providing a set of preprocessor
macros. Adding a unit test to a test suite can be easily done by the
following piece of code :

abts_suite *testdaterfc(abts_suite *suite)
{
    suite = ADD_SUITE(suite);
    abts_run_test(suite, test_date_rfc, NULL);

    return suite;
}

Where test_date_rfc is the name of the function performing the
test. Writing such a function is, in the light of the explanation I
just gave, pretty much easy too. As I said, we need to check every
entry of our input data set. That gives us a loop. For each loop
iteration, we call our to-be-tested function, grab its result and
compare the returned value with the expected one.

Test functions must have the following prototype :

static void my_test_function(abts_case *tc, void *data);

The comparison step is performed by the ABTS, thus giving the
whole test suite the correct behavior if your unit test fails. Here
comes a list of the available test methods :

ABTS_INT_EQUAL(tc, a, b)
ABTS_INT_NEQUAL(tc, a, b)
ABTS_STR_EQUAL(tc, a, b)
ABTS_STR_NEQUAL(tc, a, b, c)
ABTS_PTR_NOTNULL(tc, b)
ABTS_PTR_EQUAL(tc, a, b)
ABTS_TRUE(tc, b)
ABTS_FAIL(tc, b)
ABTS_NOT_IMPL(tc, b)
ABTS_ASSERT(tc, a, b)

The first argument, tc is a reference to the unit test currently
processed by the test suite (passed to your test function). The other
parameters are the data to be tested. For example, the following line
will never make your unit test fail :

ABTS_INT_EQUAL(tc, 1, 1);

See, it's easy ! Let's take a look at the complete example :
testdaterfc. We want to test our date string parser. For this, we will
use some chosen date strings (from mail headers for example) written
in various formats but that should all be handled by our function, and
their equivalents in correct RFC822 format.

The function we want to test returns an apr_time_t}, which will be
directly given as input to the apr_rfc822_date() function, thus
producing the corresponding RFC822 date string. All we need to do
after this is to call the correct test method from the ABTS macros !

You can take a look at the apr-util/test/testdaterfc.c file for the
complete source code of this unit test.

Although this Howto is very small and mostly dedicated to the
testdaterfc unit test, I hope you'll find it useful. Good luck !

Writing tests for CuTest (no longer used)
-----------------------------------------

There are a couple of rules for writing good tests for the test suite.

1)  All tests can determine for themselves if it passed or not.  This means
that there is no reason for the person running the test suite to interpret
the results of the tests.
2)  Never use printf to add to the output of the test suite.  The suite
library should be able to print all of the information required to debug
a problem.
3)  Functions should be tested with both positive and negative tests.  This
means that you should test things that should both succeed and fail.
4)  Just checking the return code does _NOT_ make a useful test.  You must
check to determine that the test actually did what you expected it to do.

An example test
---------------

Finally, we will look at a quick test:

/* All tests are passed a CuTest variable.  This is how the suite determines
 * if the test succeeded or failed.
 */
static void test_localstr(CuTest *tc)
{
    apr_status_t rv;
    apr_time_exp_t xt;
    time_t os_now;

    rv = apr_time_exp_lt(&xt, now);
    os_now = now / APR_USEC_PER_SEC;
   
    /* If the function can return APR_ENOTIMPL, then you should check for it.
     * This allows platform implementors to know if they have to implement
     * the function.
     */
    if (rv == APR_ENOTIMPL) {
        CuNotImpl(tc, "apr_time_exp_lt");
    }

    /* It often helps to ensure that the return code was APR_SUCESS.  If it
     * wasn't, then we know the test failed.
     */
    CuAssertTrue(tc, rv == APR_SUCCESS);

    /* Now that we know APR thinks it worked properly, we need to check the
     * output to ensure that we got what we expected.
     */
    CuAssertStrEquals(tc, "2002-08-14 12:05:36.186711 -25200 [257 Sat] DST",
                      print_time(p, &xt));
}

Notice, the same test can fail for any of a number of reasons.  The first 
test to fail ends the test.

CuTest
------

CuTest is an open source test suite written by Asim Jalis.  It has been 
released under the zlib/libpng license.  That license can be found in the
CuTest.c and CuTest.h files.

The version of CuTest that is included in the APR test suite has been modified
from the original distribution in the following ways:

1)  The original distribution does not have a -v flag, the details are always
printed.
2)  The NotImplemented result does not exist.
3)  SuiteLists do not exist.  In the original distribution, you can add suites
to suites, but it just adds the tests in the first suite to the list of tests
in the original suite.  The output wasn't as detailed as I wanted, so I created
SuiteLists.

The first two modifications have been sent to the original author of CuTest,
but they have not been integrated into the base distribution.  The SuiteList
changes will be sent to the original author soon.

The modified version of CuTest is not currently in any CVS or Subversion
server.  In time, it will be hosted at rkbloom.net.

There are currently no docs for how to write tests, but the teststr and 
testtime programs should give an idea of how it is done.  In time, a document
should be written to define how tests are written.