package Test2::API::Context;
use strict;
use warnings;

our $VERSION = '1.302183';


use Carp qw/confess croak/;
use Scalar::Util qw/weaken blessed/;
use Test2::Util qw/get_tid try pkg_to_file get_tid/;

use Test2::EventFacet::Trace();
use Test2::API();

# Preload some key event types
my %LOADED = (
    map {
        my $pkg  = "Test2::Event::$_";
        my $file = "Test2/Event/$_.pm";
        require $file unless $INC{$file};
        ( $pkg => $pkg, $_ => $pkg )
    } qw/Ok Diag Note Plan Bail Exception Waiting Skip Subtest Pass Fail V2/
);

use Test2::Util::ExternalMeta qw/meta get_meta set_meta delete_meta/;
use Test2::Util::HashBase qw{
    stack hub trace _on_release _depth _is_canon _is_spawn _aborted
    errno eval_error child_error thrown
};

# Private, not package vars
# It is safe to cache these.
my $ON_RELEASE = Test2::API::_context_release_callbacks_ref();
my $CONTEXTS   = Test2::API::_contexts_ref();

sub init {
    my $self = shift;

    confess "The 'trace' attribute is required"
        unless $self->{+TRACE};

    confess "The 'hub' attribute is required"
        unless $self->{+HUB};

    $self->{+_DEPTH} = 0 unless defined $self->{+_DEPTH};

    $self->{+ERRNO}       = $! unless exists $self->{+ERRNO};
    $self->{+EVAL_ERROR}  = $@ unless exists $self->{+EVAL_ERROR};
    $self->{+CHILD_ERROR} = $? unless exists $self->{+CHILD_ERROR};
}

sub snapshot { bless {%{$_[0]}, _is_canon => undef, _is_spawn => undef, _aborted => undef}, __PACKAGE__ }

sub restore_error_vars {
    my $self = shift;
    ($!, $@, $?) = @$self{+ERRNO, +EVAL_ERROR, +CHILD_ERROR};
}

sub DESTROY {
    return unless $_[0]->{+_IS_CANON} || $_[0]->{+_IS_SPAWN};
    return if $_[0]->{+_ABORTED} && ${$_[0]->{+_ABORTED}};
    my ($self) = @_;

    my $hub = $self->{+HUB};
    my $hid = $hub->{hid};

    # Do not show the warning if it looks like an exception has been thrown, or
    # if the context is not local to this process or thread.
    {
        # Sometimes $@ is uninitialized, not a problem in this case so do not
        # show the warning about using eq.
        no warnings 'uninitialized';
        if($self->{+EVAL_ERROR} eq $@ && $hub->is_local) {
            require Carp;
            my $mess = Carp::longmess("Context destroyed");
            my $frame = $self->{+_IS_SPAWN} || $self->{+TRACE}->frame;
            warn <<"            EOT";
A context appears to have been destroyed without first calling release().
Based on \$@ it does not look like an exception was thrown (this is not always
a reliable test)

This is a problem because the global error variables (\$!, \$@, and \$?) will
not be restored. In addition some release callbacks will not work properly from
inside a DESTROY method.

Here are the context creation details, just in case a tool forgot to call
release():
  File: $frame->[1]
  Line: $frame->[2]
  Tool: $frame->[3]

Here is a trace to the code that caused the context to be destroyed, this could
be an exit(), a goto, or simply the end of a scope:
$mess

Cleaning up the CONTEXT stack...
            EOT
        }
    }

    return if $self->{+_IS_SPAWN};

    # Remove the key itself to avoid a slow memory leak
    delete $CONTEXTS->{$hid};
    $self->{+_IS_CANON} = undef;

    if (my $cbk = $self->{+_ON_RELEASE}) {
        $_->($self) for reverse @$cbk;
    }
    if (my $hcbk = $hub->{_context_release}) {
        $_->($self) for reverse @$hcbk;
    }
    $_->($self) for reverse @$ON_RELEASE;
}

# release exists to implement behaviors like die-on-fail. In die-on-fail you
# want to die after a failure, but only after diagnostics have been reported.
# The ideal time for the die to happen is when the context is released.
# Unfortunately die does not work in a DESTROY block.
sub release {
    my ($self) = @_;

    ($!, $@, $?) = @$self{+ERRNO, +EVAL_ERROR, +CHILD_ERROR} and return if $self->{+THROWN};

    ($!, $@, $?) = @$self{+ERRNO, +EVAL_ERROR, +CHILD_ERROR} and return $self->{+_IS_SPAWN} = undef
        if $self->{+_IS_SPAWN};

    croak "release() should not be called on context that is neither canon nor a child"
        unless $self->{+_IS_CANON};

    my $hub = $self->{+HUB};
    my $hid = $hub->{hid};

    croak "context thinks it is canon, but it is not"
        unless $CONTEXTS->{$hid} && $CONTEXTS->{$hid} == $self;

    # Remove the key itself to avoid a slow memory leak
    $self->{+_IS_CANON} = undef;
    delete $CONTEXTS->{$hid};

    if (my $cbk = $self->{+_ON_RELEASE}) {
        $_->($self) for reverse @$cbk;
    }
    if (my $hcbk = $hub->{_context_release}) {
        $_->($self) for reverse @$hcbk;
    }
    $_->($self) for reverse @$ON_RELEASE;

    # Do this last so that nothing else changes them.
    # If one of the hooks dies then these do not get restored, this is
    # intentional
    ($!, $@, $?) = @$self{+ERRNO, +EVAL_ERROR, +CHILD_ERROR};

    return;
}

sub do_in_context {
    my $self = shift;
    my ($sub, @args) = @_;

    # We need to update the pid/tid and error vars.
    my $clone = $self->snapshot;
    @$clone{+ERRNO, +EVAL_ERROR, +CHILD_ERROR} = ($!, $@, $?);
    $clone->{+TRACE} = $clone->{+TRACE}->snapshot(pid => $$, tid => get_tid());

    my $hub = $clone->{+HUB};
    my $hid = $hub->hid;

    my $old = $CONTEXTS->{$hid};

    $clone->{+_IS_CANON} = 1;
    $CONTEXTS->{$hid} = $clone;
    weaken($CONTEXTS->{$hid});
    my ($ok, $err) = &try($sub, @args);
    my ($rok, $rerr) = try { $clone->release };
    delete $clone->{+_IS_CANON};

    if ($old) {
        $CONTEXTS->{$hid} = $old;
        weaken($CONTEXTS->{$hid});
    }
    else {
        delete $CONTEXTS->{$hid};
    }

    die $err  unless $ok;
    die $rerr unless $rok;
}

sub done_testing {
    my $self = shift;
    $self->hub->finalize($self->trace, 1);
    return;
}

sub throw {
    my ($self, $msg) = @_;
    $self->{+THROWN} = 1;
    ${$self->{+_ABORTED}}++ if $self->{+_ABORTED};
    $self->release if $self->{+_IS_CANON} || $self->{+_IS_SPAWN};
    $self->trace->throw($msg);
}

sub alert {
    my ($self, $msg) = @_;
    $self->trace->alert($msg);
}

sub send_ev2_and_release {
    my $self = shift;
    my $out  = $self->send_ev2(@_);
    $self->release;
    return $out;
}

sub send_ev2 {
    my $self = shift;

    my $e;
    {
        local $Carp::CarpLevel = $Carp::CarpLevel + 1;
        $e = Test2::Event::V2->new(
            trace => $self->{+TRACE}->snapshot,
            @_,
        );
    }

    if ($self->{+_ABORTED}) {
        my $f = $e->facet_data;
        ${$self->{+_ABORTED}}++ if $f->{control}->{halt} || defined($f->{control}->{terminate}) || defined($e->terminate);
    }
    $self->{+HUB}->send($e);
}

sub build_ev2 {
    my $self = shift;

    local $Carp::CarpLevel = $Carp::CarpLevel + 1;
    Test2::Event::V2->new(
        trace => $self->{+TRACE}->snapshot,
        @_,
    );
}

sub send_event_and_release {
    my $self = shift;
    my $out = $self->send_event(@_);
    $self->release;
    return $out;
}

sub send_event {
    my $self  = shift;
    my $event = shift;
    my %args  = @_;

    my $pkg = $LOADED{$event} || $self->_parse_event($event);

    my $e;
    {
        local $Carp::CarpLevel = $Carp::CarpLevel + 1;
        $e = $pkg->new(
            trace => $self->{+TRACE}->snapshot,
            %args,
        );
    }

    if ($self->{+_ABORTED}) {
        my $f = $e->facet_data;
        ${$self->{+_ABORTED}}++ if $f->{control}->{halt} || defined($f->{control}->{terminate}) || defined($e->terminate);
    }
    $self->{+HUB}->send($e);
}

sub build_event {
    my $self  = shift;
    my $event = shift;
    my %args  = @_;

    my $pkg = $LOADED{$event} || $self->_parse_event($event);

    local $Carp::CarpLevel = $Carp::CarpLevel + 1;
    $pkg->new(
        trace => $self->{+TRACE}->snapshot,
        %args,
    );
}

sub pass {
    my $self = shift;
    my ($name) = @_;

    my $e = bless(
        {
            trace => bless({%{$self->{+TRACE}}}, 'Test2::EventFacet::Trace'),
            name  => $name,
        },
        "Test2::Event::Pass"
    );

    $self->{+HUB}->send($e);
    return $e;
}

sub pass_and_release {
    my $self = shift;
    my ($name) = @_;

    my $e = bless(
        {
            trace => bless({%{$self->{+TRACE}}}, 'Test2::EventFacet::Trace'),
            name  => $name,
        },
        "Test2::Event::Pass"
    );

    $self->{+HUB}->send($e);
    $self->release;
    return 1;
}

sub fail {
    my $self = shift;
    my ($name, @diag) = @_;

    my $e = bless(
        {
            trace => bless({%{$self->{+TRACE}}}, 'Test2::EventFacet::Trace'),
            name  => $name,
        },
        "Test2::Event::Fail"
    );

    for my $msg (@diag) {
        if (ref($msg) eq 'Test2::EventFacet::Info::Table') {
            $e->add_info({tag => 'DIAG', debug => 1, $msg->info_args});
        }
        else {
            $e->add_info({tag => 'DIAG', debug => 1, details => $msg});
        }
    }

    $self->{+HUB}->send($e);
    return $e;
}

sub fail_and_release {
    my $self = shift;
    my ($name, @diag) = @_;

    my $e = bless(
        {
            trace => bless({%{$self->{+TRACE}}}, 'Test2::EventFacet::Trace'),
            name  => $name,
        },
        "Test2::Event::Fail"
    );

    for my $msg (@diag) {
        if (ref($msg) eq 'Test2::EventFacet::Info::Table') {
            $e->add_info({tag => 'DIAG', debug => 1, $msg->info_args});
        }
        else {
            $e->add_info({tag => 'DIAG', debug => 1, details => $msg});
        }
    }

    $self->{+HUB}->send($e);
    $self->release;
    return 0;
}

sub ok {
    my $self = shift;
    my ($pass, $name, $on_fail) = @_;

    my $hub = $self->{+HUB};

    my $e = bless {
        trace => bless( {%{$self->{+TRACE}}}, 'Test2::EventFacet::Trace'),
        pass  => $pass,
        name  => $name,
    }, 'Test2::Event::Ok';
    $e->init;

    $hub->send($e);
    return $e if $pass;

    $self->failure_diag($e);

    if ($on_fail && @$on_fail) {
        $self->diag($_) for @$on_fail;
    }

    return $e;
}

sub failure_diag {
    my $self = shift;
    my ($e) = @_;

    # Figure out the debug info, this is typically the file name and line
    # number, but can also be a custom message. If no trace object is provided
    # then we have nothing useful to display.
    my $name  = $e->name;
    my $trace = $e->trace;
    my $debug = $trace ? $trace->debug : "[No trace info available]";

    # Create the initial diagnostics. If the test has a name we put the debug
    # info on a second line, this behavior is inherited from Test::Builder.
    my $msg = defined($name)
        ? qq[Failed test '$name'\n$debug.\n]
        : qq[Failed test $debug.\n];

    $self->diag($msg);
}

sub skip {
    my $self = shift;
    my ($name, $reason, @extra) = @_;
    $self->send_event(
        'Skip',
        name => $name,
        reason => $reason,
        pass => 1,
        @extra,
    );
}

sub note {
    my $self = shift;
    my ($message) = @_;
    $self->send_event('Note', message => $message);
}

sub diag {
    my $self = shift;
    my ($message) = @_;
    my $hub = $self->{+HUB};
    $self->send_event(
        'Diag',
        message => $message,
    );
}

sub plan {
    my ($self, $max, $directive, $reason) = @_;
    $self->send_event('Plan', max => $max, directive => $directive, reason => $reason);
}

sub bail {
    my ($self, $reason) = @_;
    $self->send_event('Bail', reason => $reason);
}

sub _parse_event {
    my $self = shift;
    my $event = shift;

    my $pkg;
    if ($event =~ m/^\+(.*)/) {
        $pkg = $1;
    }
    else {
        $pkg = "Test2::Event::$event";
    }

    unless ($LOADED{$pkg}) {
        my $file = pkg_to_file($pkg);
        my ($ok, $err) = try { require $file };
        $self->throw("Could not load event module '$pkg': $err")
            unless $ok;

        $LOADED{$pkg} = $pkg;
    }

    confess "'$pkg' is not a subclass of 'Test2::Event'"
        unless $pkg->isa('Test2::Event');

    $LOADED{$event} = $pkg;

    return $pkg;
}

1;

__END__

=pod

=encoding UTF-8

=head1 NAME

Test2::API::Context - Object to represent a testing context.

=head1 DESCRIPTION

The context object is the primary interface for authors of testing tools
written with L<Test2>. The context object represents the context in
which a test takes place (File and Line Number), and provides a quick way to
generate events from that context. The context object also takes care of
sending events to the correct L<Test2::Hub> instance.

=head1 SYNOPSIS

In general you will not be creating contexts directly. To obtain a context you
should always use C<context()> which is exported by the L<Test2::API> module.

    use Test2::API qw/context/;

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

        if ($bool) {
            $ctx->pass($name);
        }
        else {
            $ctx->fail($name);
        }

        $ctx->release; # You MUST do this!
        return $bool;
    }

Context objects make it easy to wrap other tools that also use context. Once
you grab a context, any tool you call before releasing your context will
inherit it:

    sub wrapper {
        my ($bool, $name) = @_;
        my $ctx = context();
        $ctx->diag("wrapping my_ok");

        my $out = my_ok($bool, $name);
        $ctx->release; # You MUST do this!
        return $out;
    }

=head1 CRITICAL DETAILS

=over 4

=item you MUST always use the context() sub from Test2::API

Creating your own context via C<< Test2::API::Context->new() >> will almost never
produce a desirable result. Use C<context()> which is exported by L<Test2::API>.

There are a handful of cases where a tool author may want to create a new
context by hand, which is why the C<new> method exists. Unless you really know
what you are doing you should avoid this.

=item You MUST always release the context when done with it

Releasing the context tells the system you are done with it. This gives it a
chance to run any necessary callbacks or cleanup tasks. If you forget to
release the context it will try to detect the problem and warn you about it.

=item You MUST NOT pass context objects around

When you obtain a context object it is made specifically for your tool and any
tools nested within. If you pass a context around you run the risk of polluting
other tools with incorrect context information.

If you are certain that you want a different tool to use the same context you
may pass it a snapshot. C<< $ctx->snapshot >> will give you a shallow clone of
the context that is safe to pass around or store.

=item You MUST NOT store or cache a context for later

As long as a context exists for a given hub, all tools that try to get a
context will get the existing instance. If you try to store the context you
will pollute other tools with incorrect context information.

If you are certain that you want to save the context for later, you can use a
snapshot. C<< $ctx->snapshot >> will give you a shallow clone of the context
that is safe to pass around or store.

C<context()> has some mechanisms to protect you if you do cause a context to
persist beyond the scope in which it was obtained. In practice you should not
rely on these protections, and they are fairly noisy with warnings.

=item You SHOULD obtain your context as soon as possible in a given tool

You never know what tools you call from within your own tool will need a
context. Obtaining the context early ensures that nested tools can find the
context you want them to find.

=back

=head1 METHODS

=over 4

=item $ctx->done_testing;

Note that testing is finished. If no plan has been set this will generate a
Plan event.

=item $clone = $ctx->snapshot()

This will return a shallow clone of the context. The shallow clone is safe to
store for later.

=item $ctx->release()

This will release the context. This runs cleanup tasks, and several important
hooks. It will also restore C<$!>, C<$?>, and C<$@> to what they were when the
context was created.

B<Note:> If a context is acquired more than once an internal refcount is kept.
C<release()> decrements the ref count, none of the other actions of
C<release()> will occur unless the refcount hits 0. This means only the last
call to C<release()> will reset C<$?>, C<$!>, C<$@>,and run the cleanup tasks.

=item $ctx->throw($message)

This will throw an exception reporting to the file and line number of the
context. This will also release the context for you.

=item $ctx->alert($message)

This will issue a warning from the file and line number of the context.

=item $stack = $ctx->stack()

This will return the L<Test2::API::Stack> instance the context used to find
the current hub.

=item $hub = $ctx->hub()

This will return the L<Test2::Hub> instance the context recognizes as the
current one to which all events should be sent.

=item $dbg = $ctx->trace()

This will return the L<Test2::EventFacet::Trace> instance used by the context.

=item $ctx->do_in_context(\&code, @args);

Sometimes you have a context that is not current, and you want things to use it
as the current one. In these cases you can call
C<< $ctx->do_in_context(sub { ... }) >>. The codeblock will be run, and
anything inside of it that looks for a context will find the one on which the
method was called.

This B<DOES NOT> affect context on other hubs, only the hub used by the context
will be affected.

    my $ctx = ...;
    $ctx->do_in_context(sub {
        my $ctx = context(); # returns the $ctx the sub is called on
    });

B<Note:> The context will actually be cloned, the clone will be used instead of
the original. This allows the thread id, process id, and error variables to be correct without
modifying the original context.

=item $ctx->restore_error_vars()

This will set C<$!>, C<$?>, and C<$@> to what they were when the context was
created. There is no localization or anything done here, calling this method
will actually set these vars.

=item $! = $ctx->errno()

The (numeric) value of C<$!> when the context was created.

=item $? = $ctx->child_error()

The value of C<$?> when the context was created.

=item $@ = $ctx->eval_error()

The value of C<$@> when the context was created.

=back

=head2 EVENT PRODUCTION METHODS

B<Which one do I use?>

The C<pass*> and C<fail*> are optimal if they meet your situation, using one of
them will always be the most optimal. That said they are optimal by eliminating
many features.

Method such as C<ok>, and C<note> are shortcuts for generating common 1-task
events based on the old API, however they are forward compatible, and easy to
use. If these meet your needs then go ahead and use them, but please check back
often for alternatives that may be added.

If you want to generate new style events, events that do many things at once,
then you want the C<*ev2*> methods. These let you directly specify which facets
you wish to use.

=over 4

=item $event = $ctx->pass()

=item $event = $ctx->pass($name)

This will send and return an L<Test2::Event::Pass> event. You may optionally
provide a C<$name> for the assertion.

The L<Test2::Event::Pass> is a specially crafted and optimized event, using
this will help the performance of passing tests.

=item $true = $ctx->pass_and_release()

=item $true = $ctx->pass_and_release($name)

This is a combination of C<pass()> and C<release()>. You can use this if you do
not plan to do anything with the context after sending the event. This helps
write more clear and compact code.

    sub shorthand {
        my ($bool, $name) = @_;
        my $ctx = context();
        return $ctx->pass_and_release($name) if $bool;

        ... Handle a failure ...
    }

    sub longform {
        my ($bool, $name) = @_;
        my $ctx = context();

        if ($bool) {
            $ctx->pass($name);
            $ctx->release;
            return 1;
        }

        ... Handle a failure ...
    }

=item my $event = $ctx->fail()

=item my $event = $ctx->fail($name)

=item my $event = $ctx->fail($name, @diagnostics)

This lets you send an L<Test2::Event::Fail> event. You may optionally provide a
C<$name> and C<@diagnostics> messages.

Diagnostics messages can be simple strings, data structures, or instances of
L<Test2::EventFacet::Info::Table> (which are converted inline into the
L<Test2::EventFacet::Info> structure).

=item my $false = $ctx->fail_and_release()

=item my $false = $ctx->fail_and_release($name)

=item my $false = $ctx->fail_and_release($name, @diagnostics)

This is a combination of C<fail()> and C<release()>. This can be used to write
clearer and shorter code.

    sub shorthand {
        my ($bool, $name) = @_;
        my $ctx = context();
        return $ctx->fail_and_release($name) unless $bool;

        ... Handle a success ...
    }

    sub longform {
        my ($bool, $name) = @_;
        my $ctx = context();

        unless ($bool) {
            $ctx->pass($name);
            $ctx->release;
            return 1;
        }

        ... Handle a success ...
    }


=item $event = $ctx->ok($bool, $name)

=item $event = $ctx->ok($bool, $name, \@on_fail)

B<NOTE:> Use of this method is discouraged in favor of C<pass()> and C<fail()>
which produce L<Test2::Event::Pass> and L<Test2::Event::Fail> events. These
newer event types are faster and less crufty.

This will create an L<Test2::Event::Ok> object for you. If C<$bool> is false
then an L<Test2::Event::Diag> event will be sent as well with details about the
failure. If you do not want automatic diagnostics you should use the
C<send_event()> method directly.

The third argument C<\@on_fail>) is an optional set of diagnostics to be sent in
the event of a test failure. Unlike with C<fail()> these diagnostics must be
plain strings, data structures are not supported.

=item $event = $ctx->note($message)

Send an L<Test2::Event::Note>. This event prints a message to STDOUT.

=item $event = $ctx->diag($message)

Send an L<Test2::Event::Diag>. This event prints a message to STDERR.

=item $event = $ctx->plan($max)

=item $event = $ctx->plan(0, 'SKIP', $reason)

This can be used to send an L<Test2::Event::Plan> event. This event
usually takes either a number of tests you expect to run. Optionally you can
set the expected count to 0 and give the 'SKIP' directive with a reason to
cause all tests to be skipped.

=item $event = $ctx->skip($name, $reason);

Send an L<Test2::Event::Skip> event.

=item $event = $ctx->bail($reason)

This sends an L<Test2::Event::Bail> event. This event will completely
terminate all testing.

=item $event = $ctx->send_ev2(%facets)

This lets you build and send a V2 event directly from facets. The event is
returned after it is sent.

This example sends a single assertion, a note (comment for stdout in
Test::Builder talk) and sets the plan to 1.

    my $event = $ctx->send_event(
        plan   => {count => 1},
        assert => {pass  => 1, details => "A passing assert"},
        info => [{tag => 'NOTE', details => "This is a note"}],
    );

=item $event = $ctx->build_e2(%facets)

This is the same as C<send_ev2()>, except it builds and returns the event
without sending it.

=item $event = $ctx->send_ev2_and_release($Type, %parameters)

This is a combination of C<send_ev2()> and C<release()>.

    sub shorthand {
        my $ctx = context();
        return $ctx->send_ev2_and_release(assert => {pass => 1, details => 'foo'});
    }

    sub longform {
        my $ctx = context();
        my $event = $ctx->send_ev2(assert => {pass => 1, details => 'foo'});
        $ctx->release;
        return $event;
    }

=item $event = $ctx->send_event($Type, %parameters)

B<It is better to use send_ev2() in new code.>

This lets you build and send an event of any type. The C<$Type> argument should
be the event package name with C<Test2::Event::> left off, or a fully
qualified package name prefixed with a '+'. The event is returned after it is
sent.

    my $event = $ctx->send_event('Ok', ...);

or

    my $event = $ctx->send_event('+Test2::Event::Ok', ...);

=item $event = $ctx->build_event($Type, %parameters)

B<It is better to use build_ev2() in new code.>

This is the same as C<send_event()>, except it builds and returns the event
without sending it.

=item $event = $ctx->send_event_and_release($Type, %parameters)

B<It is better to use send_ev2_and_release() in new code.>

This is a combination of C<send_event()> and C<release()>.

    sub shorthand {
        my $ctx = context();
        return $ctx->send_event_and_release(Pass => { name => 'foo' });
    }

    sub longform {
        my $ctx = context();
        my $event = $ctx->send_event(Pass => { name => 'foo' });
        $ctx->release;
        return $event;
    }

=back

=head1 HOOKS

There are 2 types of hooks, init hooks, and release hooks. As the names
suggest, these hooks are triggered when contexts are created or released.

=head2 INIT HOOKS

These are called whenever a context is initialized. That means when a new
instance is created. These hooks are B<NOT> called every time something
requests a context, just when a new one is created.

=head3 GLOBAL

This is how you add a global init callback. Global callbacks happen for every
context for any hub or stack.

    Test2::API::test2_add_callback_context_init(sub {
        my $ctx = shift;
        ...
    });

=head3 PER HUB

This is how you add an init callback for all contexts created for a given hub.
These callbacks will not run for other hubs.

    $hub->add_context_init(sub {
        my $ctx = shift;
        ...
    });

=head3 PER CONTEXT

This is how you specify an init hook that will only run if your call to
C<context()> generates a new context. The callback will be ignored if
C<context()> is returning an existing context.

    my $ctx = context(on_init => sub {
        my $ctx = shift;
        ...
    });

=head2 RELEASE HOOKS

These are called whenever a context is released. That means when the last
reference to the instance is about to be destroyed. These hooks are B<NOT>
called every time C<< $ctx->release >> is called.

=head3 GLOBAL

This is how you add a global release callback. Global callbacks happen for every
context for any hub or stack.

    Test2::API::test2_add_callback_context_release(sub {
        my $ctx = shift;
        ...
    });

=head3 PER HUB

This is how you add a release callback for all contexts created for a given
hub. These callbacks will not run for other hubs.

    $hub->add_context_release(sub {
        my $ctx = shift;
        ...
    });

=head3 PER CONTEXT

This is how you add release callbacks directly to a context. The callback will
B<ALWAYS> be added to the context that gets returned, it does not matter if a
new one is generated, or if an existing one is returned.

    my $ctx = context(on_release => sub {
        my $ctx = shift;
        ...
    });

=head1 THIRD PARTY META-DATA

This object consumes L<Test2::Util::ExternalMeta> which provides a consistent
way for you to attach meta-data to instances of this class. This is useful for
tools, plugins, and other extensions.

=head1 SOURCE

The source code repository for Test2 can be found at
F<http://github.com/Test-More/test-more/>.

=head1 MAINTAINERS

=over 4

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

=back

=head1 AUTHORS

=over 4

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

=item Kent Fredric E<lt>kentnl@cpan.orgE<gt>

=back

=head1 COPYRIGHT

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://dev.perl.org/licenses/>

=cut