View on
Randy Stauner > Test-Aggregate > Test::Aggregate



Annotate this POD


New  2
Open  2
View/Report Bugs
Module Version: 0.375   Source  


Test::Aggregate - Aggregate *.t tests to make them run faster.


Version 0.375


    use Test::Aggregate;

    my $tests = Test::Aggregate->new( {
        dirs => $aggregate_test_dir,
    } );

    ok $some_data, 'Test::Aggregate also re-exports Test::More functions';


WARNING: this is ALPHA code. The interface is not guaranteed to be stable. Further, check out Test::Aggregate::Nested (included with this distribution). It's a more robust implementation which does not have the same limitations as Test::Aggregate.

A common problem with many test suites is that they can take a long time to run. The longer they run, the less likely you are to run the tests. This module borrows a trick from Apache::Registry to load up your tests at once, create a separate package for each test and wraps each package in a method named run_the_tests. This allows us to load perl only once and related modules only once. If you have modules which are expensive to load, this can dramatically speed up a test suite.


For a whole variety of reasons, tests run in BEGIN/CHECK/INIT/INIT blocks are now deprecated. They cause all sorts of test sequence headaches. Plus, they break the up-coming nested TAP work. You will have a problem if you use this common idiom:

     use_ok 'My::Module' or die;

Instead, just use the module and put the use_ok tests in a t/load.t file or something similar and don't aggregate it. See the following for more information:


Create a separate directory for your tests. This should not be a subdirectory of your regular test directory. Write a small driver program and put it in your regular test directory (t/ is the standard):

 use Test::Aggregate;
 my $other_test_dir = 'aggregate_tests';
 my $tests = Test::Aggregate->new( {
    dirs => $other_test_dir

 ok $some_data, 'Test::Aggregate also re-exports Test::More functions';

Take your simplest tests and move them, one by one, into the new test directory and keep running the Test::Aggregate program. You'll find some tests will not run in a shared environment like this. You can either fix the tests or simply leave them in your regular test directory. See how this distribution's tests are organized for an example.

Note that Test::Aggregate also exports all exported functions from Test::More, allowing you to run other tests after the aggregated tests have run.

 use Test::Aggregate;
 my $other_test_dir = 'aggregate_tests';
 my $tests = Test::Aggregate->new( {
    dirs => $other_test_dir
 ok !(-f 't/data/tmp.txt'), '... and our temp file should be deleted';

Some tests cannot run in an aggregate environment. These may include test for this with the $ENV{TEST_AGGREGATE} variable:

 package Some::Package;

     die __PACKAGE__ ." cannot run in aggregated tests"



 my $tests = Test::Aggregate->new(
         dirs            => 'aggtests',
         verbose         => 1,            # optional, but recommended
         dump            => 'dump.t',     # optional
         shuffle         => 1,            # optional
         matching        => qr/customer/, # optional
         set_filenames   => 0,            # optional and not recommended
         tidy            => 1,            # optional and experimental
         test_nowarnings => 0,            # optional and experimental

Creates a new Test::Aggregate instance. Accepts a hashref with the following keys:



Attempts to aggregate and run all tests listed in the directories specified in the constructor.


Since BEGIN and END blocks are for the entire aggregated tests and not for each test program (see CAVEATS), you might find that you need to have setup/teardown functions for tests. These are useful if you need to setup connections to test databases, clear out temp files, or any of a variety of tasks that your test suite might require. Here's a somewhat useless example, pulled from our tests:

 use strict;
 use warnings;
 use lib 'lib', 't/lib';
 use Test::Aggregate;
 use Test::More;
 my $dump = 'dump.t';
 my ( $startup, $shutdown ) = ( 0, 0 );
 my ( $setup,   $teardown ) = ( 0, 0 );
 my $tests = Test::Aggregate->new(
         dirs     => 'aggtests',
         dump     => $dump,
         startup  => sub { $startup++ },
         shutdown => sub { $shutdown++ },
         setup    => sub { $setup++ },
         teardown => sub { $teardown++ },
 is $startup,  1, 'Startup should be called once';
 is $shutdown, 1, '... as should shutdown';
 is $setup,    4, 'Setup should be called once for each test program';
 is $teardown, 4, '... as should teardown';

Note that you can still dump these to a dump file. This will only work if Data::Dump::Streamer 1.11 or later is installed.

There are four attributes which can be passed to the constructor, each of which expects a code reference:


You shouldn't be using global variables and a dependence on them can break your code. However, Perl provides quite a few handy global variables which, unfortunately, can easily break your tests if you change them in one test and another assumes an unchanged value. As a result, we localize many of Perl's most common global variables for you, using the following syntax:

    local %ENV = %ENV; 

The following global variables are localized for you. Any others must be localized manually per test.


Not all tests can be included with this technique. If you have Test::Class tests, there is no need to run them with this. Otherwise:


Before aggregating tests, make sure that you add tests one at a time to the aggregated test directory. Attempting to add many tests to the directory at once and then experiencing a failure means it will be much harder to track down which tests caused the failure.

Debugging aggregated tests which fail is a multi-step process. Let's say the following fails:

 my $tests = Test::Aggregate->new(
         dump    => 'dump.t',
         shuffle => 1,
         dirs    => 'aggtests',

Manually run the tests

The first step is to manually run all of the tests in the aggtests dir.

 prove -r aggtests/

If the failures appear the same, fix them just like you would fix any other test failure and then rerun the Test::Aggregate code.

Sometimes this means that a different number of tests run from what the aggregted tests run. Look for code which ends the program prematurely, such as an exception or an exit statement.

Run a dump file

If this does not fix your problem, create a dump file by passing dump => $dumpfile to the constructor (as in the above example). Then try running this dumpfile directly to attempt to replicate the error:

 prove -r $dumpfile

Tweaking the dump file

Assuming the error has been replicated, open up the dump file. The beginning of the dump file will have some code which overrides some Test::Builder internals. After that, you'll see the code which runs the tests. It will look similar to this:

 if ( __FILE__ eq 'dump.t' ) {
     Test::More::diag("******** running tests for aggtests/boilerplate.t ********")
        if $ENV{TEST_VERBOSE};

     Test::More::diag("******** running tests for aggtests/subs.t ********")
        if $ENV{TEST_VERBOSE};

     Test::More::diag("******** running tests for aggtests/00-load.t ********")
        if $ENV{TEST_VERBOSE};

     Test::More::diag("******** running tests for aggtests/slow_load.t ********")
        if $ENV{TEST_VERBOSE};

You can try to narrow down the problem by commenting out all of the run_the_tests lines and gradually reintroducing them until you can figure out which one is actually causing the failure.


My Tests Threw an Exception But Passed Anyway!

This really isn't a Test::Aggregate problem so much as a general Perl problem. For each test file, Test::Aggregate wraps the tests in an eval and checks my $error = $@. Unfortunately, we sometimes get code like this:


And internally, the 'Server' class throws an exception but uses its own evals in a DESTROY block (or something similar) to trap it. If the code you call uses an eval but fails to localize it, it wipes out your eval. Neat, eh? Thus, you never get a chance to see the error. For various reasons, this tends to impact Test::Aggregate when a DESTROY block is triggered and calls code which internally uses eval (e.g., DBIx::Class). You can often fix this with:

    local $@ = $@;  # localize but preserve the value
    my $self = shift;
    # do whatever you want

BEGIN and END blocks

Remember that since the tests are now being run at once, these blocks will no longer run on a per-test basis, but will run for the entire aggregated set of tests. You may need to examine these individually to determine the problem.

CHECK and INIT blocks.

Sorry, but you can't use these (just as in modperl). See perlmod for more information about them and why they won't work.


This is a great test module. When aggregating tests together, however, it can cause pain as you'll often discover warnings that you never new existed. For a quick fix, add this before you attempt to run your tests:

 $INC{'Test/'} = 1;

That will disable Test::NoWarnings, but you'll want to go in later to fix them.


Many tests make assumptions about the paths to files and moving them into a new test directory can break this.


Tests which use $0 can be problematic as the code is run in an eval through Test::Aggregate and $0 may not match expectations. This also means that it can behave differently if run directly from a dump file.

As it turns out, you can assign to $0! We do this by default and set the $0 to the correct filename. If you don't want this behavior, pass set_filenames => 0 to the constructor.

Minimal test case

If you cannot solve the problem, feel free to try and create a minimal test case and send it to me (assuming it's something I can run).


Curtis Poe, <ovid at>


Please report any bugs or feature requests to bug-test-aggregate at, or through the web interface at I will be notified, and then you'll automatically be notified of progress on your bug as I make changes.


You can find documentation for this module with the perldoc command.

    perldoc Test::Aggregate

You can also find information oneline:


Many thanks to mauzo ( for helping me find the 'skip_all' bug.

Thanks to Johan Lindström for pointing me to Apache::Registry.


Copyright 2007 Curtis "Ovid" Poe, all rights reserved.

This program is free software; you can redistribute it and/or modify it under the same terms as Perl itself.

syntax highlighting: