=head1 NAME

Test2::Transition - Transition notes when upgrading to Test2


This is where gotchas and breakages related to the Test2 upgrade are
documented. The upgrade causes Test::Builder to defer to Test2 under the hood.
This transition is mostly transparent, but there are a few cases that can trip
you up.


This is the list of scenarios that break with the new internals.

=head2 Test::Builder1.5/2 conditionals

=head3 The Problem

a few years back there were two attempts to upgrade/replace Test::Builder.
Confusingly these were called Test::Builder2 and Test::Builder1.5, in that
order. Many people put conditionals in their code to check the Test::Builder
version number and adapt their code accordingly.

The Test::Builder2/1.5 projects both died out. Now the conditional code people
added has become a mine field. A vast majority of modules broken by Test2 fall
into this category.

=head3 The Fix

The fix is to remove all Test::Builder1.5/2 related code. Either use the
legacy Test::Builder API, or use Test2 directly.

=head2 Replacing the Test::Builder singleton

=head3 The Problem

Some test modules would replace the Test::Builder singleton instance with their
own instance or subclass. This was usually done to intercept or modify results
as they happened.

The Test::Builder singleton is now a simple compatibility wrapper around
Test2. The Test::Builder singleton is no longer the central place for
results. Many results bypass the Test::Builder singleton completely, which
breaks and behavior intended when replacing the singleton.

=head3 The Fix

If you simply want to intercept all results instead of letting them go to TAP,
you should look at the L<Test2::API> docs and read about pushing a new hub onto
the hub stack. Replacing the hub temporarily is now the correct way to
intercept results.

If your goal is purely monitoring of events use the C<< Test2::Hub->listen() >>
method exported by Test::More to watch events as they are fired. If you wish to
modify results before they go to TAP look at the C<< Test2::Hub->filter() >>

=head2 Directly Accessing Hash Elements

=head3 The Problem

Some modules look directly at hash keys on the Test::Builder singleton. The
problem here is that the Test::Builder singleton no longer holds anything

=head3 The Fix

The fix is to use the API specified in L<Test2::API> to look at or modify state
as needed.

=head2 Subtest indentation

=head3 The Problem

An early change, in fact the change that made Test2 an idea, was a change to
the indentation of the subtest note. It was decided it would be more readable
to outdent the subtest note instead of having it inline with the subtest:

    # subtest foo
        ok 1 - blah
    ok 1 - subtest foo

The old style indented the note:

        # subtest foo
        ok 1 - blah
    ok 1 - subtest foo

This breaks tests that do string comparison of TAP output.

=head3 The Fix

    my $indent = $INC{'Test2/API.pm'} ? '' : '    ';

        "${indent}# subtest foo",
        "Got subtest note"

Check if C<$INC{'Test2/API.pm'}> is set, if it is then no indentation should be
expected. If it is not set, then the old Test::Builder is in use, indentation
should be expected.


This is a list of cpan modules that have been known to have been broken by the
upgrade at one point.


These modules still function correctly, but their test suites will not pass. If
you already have these modules installed then you can continue to use them. If
you are trying to install them after upgrading Test::Builder you will need to
force installation, or bypass the broken tests.

=over 4

=item Test::DBIx::Class::Schema

This module has a test that appears to work around a Test::Builder bug. The bug
appears to have been fixed by Test2, which means the workaround causes a
failure. This can be easily updated, but nobody has done so yet.

Known broken in versions: 1.0.9 and older

=item Device::Chip

Tests break due to subtest indentation.

Known broken in version 0.07. Apparently works fine in 0.06 though. Patch has
been submitted to fix the issue.



These are modules that did not break, but had broken test suites that have
since been fixed.

=over 4

=item Test::Exception

Old versions work fine, but have a minor test name behavior that breaks with
Test2. Old versions will no longer install because of this. The latest version
on CPAN will install just fine. Upgrading is not required, but is recommended.

Fixed in version: 0.43

=item Data::Peek

Some tests depended on C<$!> and C<$?> being modified in subtle ways. A patch
was applied to correct things that changed.

The module itself works fine, there is no need to upgrade.

Fixed in version: 0.45

=item circular::require

Some tests were fragile and required base.pm to be loaded at a late stage.
Test2 was loading base.pm too early. The tests were updated to fix this.

The module itself never broke, you do not need to upgrade.

Fixed in version: 0.12

=item Test::Module::Used

A test worked around a now-fixed planning bug. There is no need to upgrade if
you have an old version installed. New versions install fine if you want them.

Fixed in version: 0.2.5

=item Test::Moose::More

Some tests were fragile, but have been fixed. The actual breakage was from the
subtest comment indentation change.

No need to upgrade, old versions work fine. Only new versions will install.

Fixed in version: 0.025

=item Test::FITesque

This was broken by a bugfix to how planning is done. The test was updated after
the bugfix.

Fixed in version: 0.04

=item Test::Kit

Old versions work fine, but would not install because L<Test::Aggregate> was in
the dependency chain. An upgrade should not be needed.

Fixed in version: 2.15

=item autouse

A test broke because it depended on Scalar::Util not being loaded. Test2 loads
Scalar::Util. The test was updated to load Test2 after checking Scalar::Util's
load status.

There is no need to upgrade if you already have it installed.

Fixed in version: 1.11



=over 4

=item Test::SharedFork

Old versions need to directly access Test::Builder singleton hash elements. The
latest version on CPAN will still do this on old Test::Builder, but will defer
to L<Test2::IPC> on Test2.

Fixed in version: 0.35

=item Test::Builder::Clutch

This works by doing overriding methods on the singleton, and directly accessing
hash values on the singleton. A new version has been released that uses the
Test2 API to accomplish the same result in a saner way.

Fixed in version: 0.07

=item Test::Dist::VersionSync

This had Test::Builder2 conditionals. This was fixed by removing the

Fixed in version: 1.1.4

=item Test::Modern

This relied on C<< Test::Builder->_try() >> which was a private method,
documented as something nobody should use. This was fixed by using a different

Fixed in version: 0.012

=item Test::UseAllModules

Version 0.14 relied on C<< Test::Builder->history >> which was available in
Test::Builder 1.5. Versions 0.12 and 0.13 relied on other Test::Builder

Fixed in version: 0.15

=item Test::More::Prefix

Worked by applying a role that wrapped C<< Test::Builder->_print_comment >>.
Fixed by adding an event filter that modifies the message instead when running
under Test2.

Fixed in version: 0.007



=over 4

=item Test::Aggregate

This distribution directly accesses the hash keys in the L<Test::Builder>
singleton. It also approaches the problem from the wrong angle, please consider
using L<Test2::Aggregate> for similar functionality and L<Test2::Harness>
which allows module preloading at the harness level.

Still broken as of version: 0.373

=item Test::Wrapper

This module directly uses hash keys in the L<Test::Builder> singleton. This
module is also obsolete thanks to the benefits of L<Test2>. Use C<intercept()>
from L<Test2::API> to achieve a similar result.

Still broken as of version: 0.3.0

=item Test::ParallelSubtest

This module overrides C<Test::Builder::subtest()> and
C<Test::Builder::done_testing()>. It also directly accesses hash elements of
the singleton. It has not yet been fixed.

Alternatives: L<Test2::AsyncSubtest> and L<Test2::Workflow> (not stable).

Still broken as of version: 0.05

=item Test::Pretty

See https://github.com/tokuhirom/Test-Pretty/issues/25

The author admits the module is crazy, and he is awaiting a stable release of
something new (Test2) to completely rewrite it in a sane way.

Still broken as of version: 0.32

=item Net::BitTorrent

The tests for this module directly access L<Test::Builder> hash keys. Most, if
not all of these hash keys have public API methods that could be used instead
to avoid the problem.

Still broken in version: 0.052

=item Test::Group

It monkeypatches Test::Builder, and calls it "black magic" in the code.

Still broken as of version: 0.20

=item Test::Flatten

This modifies the Test::Builder internals in many ways. A better was to
accomplish the goal of this module is to write your own subtest function.

Still broken as of version: 0.11

=item Log::Dispatch::Config::TestLog

Modifies Test::Builder internals.

Still broken as of version: 0.02

=item Test::Able

Modifies Test::Builder internals.

Still broken as of version: 0.11



=head2 LEGACY

    use Test::Builder;

    # A majority of tools out there do this:
    # my $TB = Test::Builder->new;
    # This works, but has always been wrong, forcing Test::Builder to implement
    # subtests as a horrific hack. It also causes problems for tools that try
    # to replace the singleton (also discouraged).

    sub my_ok($;$) {
        my ($bool, $name) = @_;
        my $TB = Test::Builder->new;
        $TB->ok($bool, $name);

    sub my_diag($) {
        my ($msg) = @_;
        my $TB = Test::Builder->new;

=head2 TEST2

    use Test2::API qw/context/;

    sub my_ok($;$) {
        my ($bool, $name) = @_;
        my $ctx = context();
        $ctx->ok($bool, $name);

    sub my_diag($) {
        my ($msg) = @_;
        my $ctx = context();

The context object has API compatible implementations of the following methods:

=over 4

=item ok($bool, $name)

=item diag(@messages)

=item note(@messages)

=item subtest($name, $code)


If you are looking for helpers with C<is>, C<like>, and others, see


=head2 LEGACY

    use Test::More;

    sub exclusive_ok {
        my ($bool1, $bool2, $name) = @_;

        # Ensure errors are reported 1 level higher
        local $Test::Builder::Level = $Test::Builder::Level + 1;

        $ok = $bool1 || $bool2;
        $ok &&= !($bool1 && $bool2);
        ok($ok, $name);

        return $bool;

Every single tool in the chain from this, to C<ok>, to anything C<ok> calls
needs to increment the C<$Level> variable. When an error occurs Test::Builder
will do a trace to the stack frame determined by C<$Level>, and report that
file+line as the one where the error occurred. If you or any other tool you use
forgets to set C<$Level> then errors will be reported to the wrong place.

=head2 TEST2

    use Test::More;

    sub exclusive_ok {
        my ($bool1, $bool2, $name) = @_;

        # Grab and store the context, even if you do not need to use it
        # directly.
        my $ctx = context();

        $ok = $bool1 || $bool2;
        $ok &&= !($bool1 && $bool2);
        ok($ok, $name);

        return $bool;

Instead of using C<$Level> to perform a backtrace, Test2 uses a context
object. In this sample you create a context object and store it. This locks the
context (errors report 1 level up from here) for all wrapped tools to find. You
do not need to use the context object, but you do need to store it in a
variable. Once the sub ends the C<$ctx> variable is destroyed which lets future
tools find their own.

=head1 USING UTF8

=head2 LEGACY

    # Set the mode BEFORE anything loads Test::Builder
    use open ':std', ':encoding(utf8)';
    use Test::More;


    # Modify the filehandles
    my $builder = Test::More->builder;
    binmode $builder->output,         ":encoding(utf8)";
    binmode $builder->failure_output, ":encoding(utf8)";
    binmode $builder->todo_output,    ":encoding(utf8)";

=head2 TEST2

    use Test2::API qw/test2_stack/;


Though a much better way is to use the L<Test2::Plugin::UTF8> plugin, which is
part of L<Test2::Suite>.


The following people have all contributed to this document in some way, even if
only for review.

=over 4

=item Chad Granum (EXODIST) E<lt>exodist@cpan.orgE<gt>


=head1 SOURCE

The source code repository for Test2 can be found at


=over 4

=item Chad Granum E<lt>exodist@cpan.orgE<gt>



Copyright 2020 Chad Granum E<lt>exodist@cpan.orgE<gt>.

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

See F<http://www.perl.com/perl/misc/Artistic.html>