PHP: Quality Assurance
Creating new test files
phpt Test Basics
The first thing you need to know about tests is that we need more!!! Although PHP works just great
99.99% of the time, not having a very comprehensive test suite means that we take more risks every time
we add to or modify the PHP implementation. The second
thing you need to know is that if you can write PHP you can write tests. Thirdly - we are a friendly
and welcoming community, don't be scared about writing to
(php-qa@lists.php.net) - we won't bite!
So what are phpt tests?
A phpt test is a little script used by the php internal and quality
assurance teams to test PHP's functionality. It can be used with new releases to make
sure they can do all the things that previous releases can, or to help find bugs in current
releases. By writing phpt tests you are helping to make PHP more stable.
What skills are needed to write a phpt test?
All that is really needed to write a phpt test
is a basic understanding of the PHP language, a text editor, and a way to get the results
of your code. That is it. So if you have been writing and running PHP scripts already -
you have everything you need.
What do you write phpt tests on?
Basically you can write a phpt test on one of the various
php functions available. You can write a test on a basic language function (a string
function or an array function) , or a function provided by one of PHP's numerous extensions
(a mysql function or a image function or a mcrypt function).
You can find out what functions already have phpt tests by looking in the html
version of the git repository (ext/standard/tests/ is a good place to start looking - though not
all the tests currently written are in there). If you look at the gcov pages you
can see which functions have lots of tests and which need more, although these pages only
show which lines of code are covered by test cases so even if the coverage looks good there may
be more interesting tests to write - for example covering error cases. There is also a table
here that shows which functions and methods are called from PHPT tests.
If you want more guidance than that you can always ask
the PHP Quality Assurance Team on their mailing list
(php-qa@lists.php.net) where they
would like you to direct your attentions.
How is a phpt test is used?
When a test is called by the run-tests.php script it takes various
parts of the phpt file to name and create a .php file. That .php file is then executed. The
output of the .php file is then compared to a different section of the phpt file. If the output of
the script "matches" the output provided in the phpt script - it passes.
What should a phpt test do?
Basically - it should try and break the PHP function. It should check not
only the functions normal parameters, but it should also check edge cases. Intentionally generating
an error is allowed and encouraged.
Writing phpt Tests
Naming Conventions
Phpt tests follow a very strict naming convention. This is done to easily identify what each phpt
test is for. Tests should be named according to the following list:
Tests for bugs
bug<bugid>.phpt (bug17123.phpt)
Tests for a function's basic behaviour
<functionname>_basic.phpt (dba_open_basic.phpt)
Tests for a function's error behaviour
<functionname>_error.phpt (dba_open_error.phpt)
Tests for variations in a function's behaviour
<functionname>_variation.phpt (dba_open_variation.phpt)
General tests for extensions
<extname><no>.phpt (dba_003.phpt)
The convention of using _basic, _error and _variation was introduced when we
found that writing a single test case for each function resulted in unacceptably large
test cases. It's quite hard to debug problems when the test case generates 100s of lines of output.
The "basic" test case for a function should just address the single most simple
thing that the function is designed to do. For example, if writing a test for the sin() function
a basic test would just be to check that sin() returns the correct values for some known angles
- eg 30, 90, 180.
The "error" tests for a function are test cases which are designed to provoke errors, warnings or notices.
There can be more than one error case, if so the convention is to name the test cases mytest_error1.phpt,
mytest_error2.phpt and so on.
The "variation" tests are any tests that don't fit into "basic" or "error" tests. For example one might
use a variation tests to test boundary conditions.
How big is a test case?
Small. Really - the smaller the better, a good guide is no more than 10 lines of output. The reason
for this is that if we break something in PHP and it breaks your test case we need to be able to find
out quite quickly what we broke, going through 1000s of line of test case output is not easy. Having said that
it's sometimes just not practical to stay within the 10 line guideline, in this case you can help a lot
by commenting the output. You may find plenty of much longer tests in PHP - the small tests message is something
that we learnt over time, in fact we are slowly going through and splitting tests up when we need to.
Comments
Comments help. Not an essay - just a couple of lines on what the objective of the test is. It may seem completely
obvious to you as you write it, but it might not be to someone looking at it later on
Basic Format
A test must contain the sections TEST, FILE and either EXPECT or EXPECTF at a minimum. The example
below illustrates a minimal test.
ext/standard/tests/strings/strtr.phpt
--TEST--
strtr() function - basic test for strtr()
--FILE--
<?php
/* Do not change this test it is a README.TESTING example. */
$trans = array("hello"=>"hi", "hi"=>"hello", "a"=>"A", "world"=>"planet");
var_dump(strtr("# hi all, I said hello world! #", $trans));
?>
--EXPECT--
string(32) "# hello All, I sAid hi planet! #"
As you can see the file is divided into several sections. The TEST section holds a one line title
of the phpt test, this shoudl be a simple description and shouldn't ever excede one line, if you need to write more explanation
add comments in the body of the test case. The phpt files name is used when generating a .php file. The FILE section is used
as the body of the .php file, so don't forget to open and close your php tags. The EXPECT section is
the part used as a comparison to see if the test passes. It is a
good idea to generate output with var_dump() calls.
PHPT structure details
A phpt test can have many more parts than just the minimum. In fact some of the mandatory parts have
alternatives that may be used if the situation warrants it. The phpt sections are documented here.
There is also a very useful set of slides, written by Marcus Boerger here.
Look at the talk entitled "The need for speed, ERM testing".
Autogenerating test cases
It isn't possible (or even sensible) to try and generate complete test cases for PHP. However
there is a script in PHP5.3 which will help to generate the framework. It can save you
some typing and ensure that you get a good basic format.
See test case generation for instructions on how to use it.
Analyzing failing tests
While writing tests you will probably run into tests not passing while you think they should. The 'make test' command
provides you with debug information. Several files will be added per test in the same directory as the .phpt file itself.
Considering your test file is named foo.phpt, these files provide you with information that can help you find out what went wrong:
foo.diff
A diff file between the expected output (be it in EXPECT, EXPECTF or another option) and the actual output.
foo.exp
The expected output.
foo.log
A log containing expected output, actual output and results. Most likely very similar to info in the other files.
foo.out
The actual output of your .phpt test part.
foo.php
The php code that was executed for this test.
foo.sh
An executable file that executes the test for you as it was executed during failure.
Testing your test cases
Most people who write tests for PHP don't have access to a huge number of operating systems but the tests are run
on every system that runs PHP. It's good to test your test on as many platforms as you can - Linux and Windows
are the most important, it's increasingly important to make sure that tests run on 64 bit as well as 32 bit platforms. If you only
have access to one operating system - don't worry, if you have karma, commit the test but watch php-qa@lists.php.net for reports
of failures on other platforms. If you don't have karma to commit have a look at the next section.
When you are testing your test case it's really important to make sure that you
clean up any temporary resources (eg files) that you used in the test. There is a special --CLEAN-- section
to help you do this - see .
Another good check is to look at what lines of code in the PHP source your test case covers.
This is easy to do, there are some instructions on the PHP Wiki.
What should I do with my test case when I've written and tested it?
The next step is to get someone to review it. If it's short you can paste it into a note and
send it to php-qa@lists.php.net. If the test is a bit too
long for that then put it somewhere were people can download it (pastebin is
sometimes used). Appending tests to notes as files doesn't work well - so please don't do that. Your
note to php-qa@lists.php.net should say
what level of PHP you have tested it on and what platform(s) you've run it on. Someone from
the PHP QA group will review your test and reply to you. They may ask for some changes
or suggest better ways to do things, or they may commit it to PHP.
Writing Portable PHP Tests
Writing portable tests can be hard if you don't have access to all
the many platforms that PHP can run on. Do your best. If in doubt,
don't disable a test. It is better that the test runs in as many
environments as possible.
If you know a new test won't run in a specific environment, try to
write the complementary test for that environment.
Make sure sets of data are consistently ordered. SQL queries are
not guaranteed to return results in the same order unless an ORDER BY
clause is used. Directory listings are another example that can vary:
use an appropriate PHP function to sort them befor printing. Both of
these examples have affected PHP tests in the past.
Make sure that any test touching parsing or display of dates uses a
hard-defined timezone - preferable 'UTC'. It is important tha this is defined in
the file section using:
date_default_timezone_set('UTC');
and not in the INI section. This is because of the order in which settings are checked which is:
date_default_timezone_set() -> TZ environmental -> INI setting -> System Setting
If a TZ environmental varaibale is found the INI setting will be ignored.
Tests that run, or only have have matching EXPECT output, on 32bit
platforms can use a SKIPIF section like:
--SKIPIF--
<?php
if (PHP_INT_SIZE != 4) die("skip this test is for 32bit platforms only");
?>
Tests for 64bit platforms can use:
--SKIPIF--
<?php
if (PHP_INT_SIZE != 8) die("skip this test is for 64bit platforms only");
?>
To run a test only on Windows
--SKIPIF--
<?php
if (substr(PHP_OS, 0, 3) != 'WIN') die("skip this test is for Windows platforms only");
?>
To run a test only on Linux
--SKIPIF--
<?php
if (!stristr(PHP_OS, "Linux")) die("skip this test is Linux platforms only");
?>
To skip a test on Mac OS X Darwin
--SKIPIF--
<?php
if (!stristr(PHP_OS, "Darwin")) die("skip this test is for Mac OS X platforms only");
?>
Examples
EXPECTF
/ext/standard/tests/strings/str_shuffle.phpt is a good example for using
EXPECTF instead of EXPECT. From time to time the algorithm used for shuffle
changed and sometimes the machine used to execute the code has influence
on the result of shuffle. But it always returns a three character string
detectable by %s (that matches any string until the end of the line). Other scan-able
forms are %a for any amount of chars (at least one), %i for integers, %d for numbers
only, %f for floating point values, %c for single characters, %x for
hexadecimal values, %w for any number of whitespace characters and %e for
DIRECTORY_SEPARATOR ('\' or '/').
See also EXPECTF details
/ext/standard/tests/strings/str_shuffle.phpt
--TEST--
Testing str_shuffle.
--FILE--
<?php
/* Do not change this test it is a README.TESTING example. */
$s = '123';
var_dump(str_shuffle($s));
var_dump($s);
?>
--EXPECTF--
string(3) "%s"
string(3) "123"
EXPECTREGEX
/ext/standard/tests/strings/strings001.phpt is a good example for using
EXPECTREGEX instead of EXPECT. This test also shows that in EXPECTREGEX
some characters need to be escaped since otherwise they would be
interpreted as a regular expression.
/ext/standard/tests/strings/strings001.phpt
--TEST--
Test whether strstr() and strrchr() are binary safe.
--FILE--
<?php
/* Do not change this test it is a README.TESTING example. */
$s = "alabala nica".chr(0)."turska panica";
var_dump(strstr($s, "nic"));
var_dump(strrchr($s," nic"));
?>
--EXPECTREGEX--
string\(18\) \"nica\x00turska panica\"
string\(19\) \" nica\x00turska panica\"
SKIPIF
Some tests depend on modules or functions available only in certain versions
or they even require minimum version of php or zend. These tests should be
skipped when the requirement cannot be fulfilled. To achieve this you can
use the SKIPIF section. To tell run-tests.php that your test should be skipped
the SKIPIF section must print out the word "skip" followed by a reason why
the test should skip.
/ext/exif/tests/exif005.phpt
--TEST--
Check for exif_read_data, unusual IFD start
--SKIPIF--
<?php
if (!extension_loaded('exif')) print 'skip exif extension not available';
?>
--FILE--
<?php
/* Do not change this test it is a README.TESTING example.
* test5.jpg is a 1*1 image that contains an Exif section with ifd = 00000009h
*/
$image = exif_read_data('./ext/exif/tests/test5.jpg','',true,false);
var_dump($image['IFD0']);
?>
--EXPECT--
array(2) {
["ImageDescription"]=>
string(11) "Ifd00000009"
["DateTime"]=>
string(19) "2002:10:18 20:06:00"
}
Test script and SKIPIF code should be directly written into *.phpt. However,
it is recommended to use include files when more test scripts depend on the
same SKIPIF code or when certain test files need the same values for some
input.
Note: no file used by any test should have one of the following extensions:
".php", ".log", ".mem", ".exp", ".out" or ".diff". When you use an include file for the
SKIPIF section it should be named "skipif.inc" and an include file used in the
FILE section of many tests should be named "test.inc".
Final Notes
Cleaning up after running a test
Sometimes test cases create files or directories as part of the test case and it's important to remove these after the test ends, the --CLEAN--
section is provided to help with this.
The PHP code in the --CLEAN-- section is executed separately from the code in the --FILE-- section. For example, this code:
--TEST--
Will fail to clean up
--FILE--
<?php
$temp_filename = "fred.tmp";
$fp = fopen($temp_filename, "w");
fwrite($fp, "Hello Boys!");
fclose($fp);
?>
--CLEAN--
<?php
unlink($temp_filename);
?>
--EXPECT--
will not remove the temporary file because the variable $filename is not defined in the --CLEAN-- section.
Here is a better way to write the code:
--TEST--
This will remove temporary files
--FILE--
<?php
$temp_filename = __DIR__."/fred.tmp";
$fp = fopen($temp_filename, "w");
fwrite ($fp, "Hello Boys!\n");
fclose($fp);
?>
--CLEAN--
<?php
$temp_filename = __DIR__."/fred.tmp";
unlink($temp_filename);
?>
--EXPECT--
Note the use of the __DIR__ construct which will ensure that the temporary file is created in the same directory as
the phpt test script.
When creating temporary files it is a good idea to use an extension that indicates the use of the file, eg .tmp. It's also a good
idea to avoid using extensions that are already used for other purposes, eg .inc, .php. Similarly, it is helpful to give the temporary file a name
that is clearly related to the test case. For example, mytest.phpt should create mytest.tmp (or mytestN.tmp, N=1, 2,3,...) then if by any
chance the temporary file isnt't removed properly it will be obvious which test case created it.
When writing and debugging a test case with a --CLEAN-- section it is helpful to remember that the php code in the --CLEAN-- section
is executed separately from the code in the --FILE-- section. For example, in a test case called mytest.phpt, code from the --FILE--
section is run from a file called mytest.php and code from the --CLEAN-- section is run from a file called mytest.clean.php. If the test passes,
both the .php and .clean.php files are removed by run-tests.php. You can prevent the removal by using the --keep option of run-tests.php,
this is a very useful option if you need to check that the --CLEAN-- section code is working as you intended.
Finally - if you are using CVS it's helpful to add the extension that you use for test-related temporary files to the .cvsignore file -
this will help to prevent you from accidentally checking temporary files into CVS.
Redirecting tests
Using --REDIRECTTEST-- it is possible to redirect from one test to a bunch
of other tests. That way multiple extensions can refer to the same set of
test scripts probably using it with a different configuration.
The block is eval'd and supposed to return an array describing how to
redirect. The resulting array must contain the key 'TEST' that stores the
redirect target as a string. This string usually is the directory where the
test scripts are located and should be relative. Optionally you can use
the 'ENV' as an array configuring the environment to be set when executing
the tests. This way you can pass configuration to the executed tests.
Redirect tests may especially contain --SKIPIF--, --ENV-- and --ARGS--
sections but they no not use any --EXPECT-- section.
The redirected tests themselves are just normal tests.
Error reporting in tests
All tests should run correctly with error_reporting(E_ALL) and
display_errors=1. This is the default when called from run-tests.php. If you
have a good reason for lowering the error reporting, use --INI-- section and
comment this in your testcode.
If your test intentionally generates a PHP warning message use
$php_errormsg variable, which you can then output. This will result in a
consistent error message output across all platforms and PHP configurations,
preventing your test from failing due inconsistencies in the error message
content. Alternatively you can use --EXPECTF-- and check for the message by
replacing the path of the source of the message with "%s" and the line number
with "%d". The end of a message in a test file "example.phpt" then looks like
"in %sexample.php on line %d". We explicitly dropped the last path devider as
that is a system dependent character '/' or '\'.
Last bit
Often you want to run test scripts without run-tests.php by
simply executing them on command line like any other php script. But sometimes
it disturbs having a long --EXPECT-- block, so that you don't see the actual
output as it scrolls away overwritten by the blocks following the actual file
block. The workaround is to use terminate the --FILE-- section with the two
lines "===DONE===" and "<?php exit(0); ?>.
When doing so run-tests.php does not execute the line containing the exit call
as that would suppress leak messages. Actually run-tests.php ignores any part
after a line consisting only of "===DONE===".
Here is an example:
--TEST--
Test hypot() - dealing with mixed number/character input
--INI--
precision=14
--FILE--
<?php
$a="23abc";
$b=-33;
echo "$a :$b ";
$res = hypot($a, $b);
var_dump($res);
?>
===DONE===
<?php exit(0); ?>
--EXPECTF--
23abc :-33 float(40.224370722238)
===DONE===
If executed as PHP script the output will stop after the code on the --FILE-- section
has been run.
