The Perl Toolchain Summit needs more sponsors. If your company depends on Perl, please support this very important event.
package PPIx::Regexp::Test;

use strict;
use warnings;

use base qw{ Exporter };

use PPIx::Regexp;
use PPIx::Regexp::Dumper;
use PPIx::Regexp::Element;
use PPIx::Regexp::Tokenizer;
use PPIx::Regexp::Util qw{ __instance };
use Scalar::Util qw{ looks_like_number refaddr };
use Test::More 0.88;

our $VERSION = '0.038';

our @EXPORT_OK = qw{
    cache_count
    choose
    class
    cmp_ok
    content
    count
    diag
    different
    done_testing
    dump_result
    false
    finis
    equals
    navigate
    note
    parse
    plan
    ppi
    result
    skip
    tokenize
    true
    value
};

our @EXPORT = @EXPORT_OK;	## no critic (ProhibitAutomaticExportation)

my (
    $initial_class,	# For static methods; set by parse() or tokenize()
    $kind,		# of thing; set by parse() or tokenize()
    $nav,		# Navigation used to get to current object, as a
			#    string.
    $obj,		# Current object:
    			#    PPIx::Regexp::Tokenizer if set by tokenize(),
			#    PPIx::Regexp if set by parse(), or
			#    PPIx::Regexp::Element is set by navigate().
    $parse,		# Result of parse:
    			#    array ref if set by tokenize(), or
			#    PPIx::Regexp object if set by parse()
    $result,		# Operation result.
);

sub cache_count {		## no critic (RequireArgUnpacking)
    my ( $expect ) = @_;
    defined $expect or $expect = 0;
    $obj = undef;
    $parse = undef;
    _pause();
    $result = PPIx::Regexp->_cache_size();
    @_ = ( $result, $expect,
	"Should be $expect leftover cache contents" );
    goto &is;
}

sub choose {
    my @args = @_;
    $obj = $parse;
    return navigate( @args );
}

sub class {		## no critic (RequireArgUnpacking)
    my ( $class ) = @_;
    $result = ref $obj || $obj;
    if ( defined $class ) {
	@_ = ( $obj, $class, "$kind $nav" );
	goto &isa_ok;
    } else {
	@_ = ( ref $obj || undef, $class, "Class of $kind $nav" );
	goto &is;
    }
}

sub content {		## no critic (RequireArgUnpacking)
    my @args = @_;
    my $expect = pop @args;
    $result = undef;
    defined $obj and $result = $obj->content();
    my $safe;
    if ( defined $result ) {
	($safe = $result) =~ s/([\\'])/\\$1/smxg;
    } else {
	$safe = 'undef';
    }
    @_ = ( $result, $expect, "$kind $nav content '$safe'" );
    goto &is;
}

sub count {		## no critic (RequireArgUnpacking)
    my ( @args ) = @_;
    my $expect = pop @args;
    if ( ref $parse eq 'ARRAY' ) {
	$result = @{ $parse };
	@_ = ( $result, $expect, "Expect $expect tokens" );
    } elsif ( ref $obj eq 'ARRAY' ) {
	$result = @{ $obj };
	@_ = ( $result, $expect, "Expect $expect tokens" );
    } elsif ( $obj->can( 'children' ) ) {
	$result = $obj->children();
	@_ = ( $result, $expect, "Expect $expect children" );
    } else {
	$result = $obj->can( 'children' );
	@_ = ( $result, ref( $obj ) . "->can( 'children')" );
	goto &ok;
    }
    goto &is;
}

sub different {		## no critic (RequireArgUnpacking)
    my @args = @_;
    @args < 3 and unshift @args, $obj;
    my ( $left, $right, $name ) = @args;
    if ( ! defined $left && ! defined $right ) {
	@_ = ( undef, $name );
    } elsif ( ! defined $left || ! defined $right ) {
	@_ = ( 1, $name );
    } elsif ( ref $left && ref $right ) {
	@_ = ( refaddr( $left ) != refaddr( $right ), $name );
    } elsif ( ref $left || ref $right ) {
	@_ = ( 1, $name );
    } elsif ( looks_like_number( $left ) && looks_like_number( $right ) ) {
	@_ = ( $left != $right, $name );
    } else {
	@_ = ( $left ne $right, $name );
    }
    goto &ok;
}

sub dump_result {
    my ( $opt, @args ) = _parse_constructor_args( { test => 1 }, @_ );
    if ( $opt->{test} ) {
	my ( $expect, $name ) = splice @args, -2;
	my $got = PPIx::Regexp::Dumper->new( $obj, @args )->string();
	@_ = ( $got, $expect, $name );
	goto &is;
    } elsif ( __instance( $result, 'PPIx::Regexp::Tokenizer' ) ||
	__instance( $result, 'PPIx::Regexp::Element' ) ) {
	diag( PPIx::Regexp::Dumper->new( $obj, @args )->string() );
    } elsif ( eval { require YAML; 1; } ) {
	diag( "Result dump:\n", YAML::Dump( $result ) );
    } elsif ( eval { require Data::Dumper; 1 } ) {
	diag( "Result dump:\n", Data::Dumper::Dumper( $result ) );
    } else {
	diag( "Result dump unavailable.\n" );
    }
    return;
}

sub equals {		## no critic (RequireArgUnpacking)
    my @args = @_;
    @args < 3 and unshift @args, $obj;
    my ( $left, $right, $name ) = @args;
    if ( ! defined $left && ! defined $right ) {
	@_ = ( 1, $name );
    } elsif ( ! defined $left || ! defined $right ) {
	@_ = ( undef, $name );
    } elsif ( ref $left && ref $right ) {
	@_ = ( refaddr( $left ) == refaddr( $right ), $name );
    } elsif ( ref $left || ref $right ) {
	@_ = ( undef, $name );
    } elsif ( looks_like_number( $left ) && looks_like_number( $right ) ) {
	@_ = ( $left == $right, $name );
    } else {
	@_ = ( $left eq $right, $name );
    }
    goto &ok;
}

sub false {		## no critic (RequireArgUnpacking)
    my ( $method, $args ) = @_;
    ref $args eq 'ARRAY' or $args = [ $args ];
    my $class = ref $obj;
    if ( $obj->can( $method ) ) {
	$result = $obj->$method( @{ $args } );
	my $fmtd = _format_args( $args );
	@_ = ( ! $result, "$class->$method$fmtd is false" );
    } else {
	$result = undef;
	@_ = ( undef, "$class->$method() exists" );
    }
    goto &ok;
}

sub finis {		## no critic (RequireArgUnpacking)
    $obj = $parse = $result = undef;
    _pause();
    $result = PPIx::Regexp::Element->_parent_keys();
    @_ = ( $result, 0, 'Should be no leftover objects' );
    goto &is;
}

{

    my %array = map { $_ => 1} qw{
	children delimiters finish schildren start tokens type
    };

    sub navigate {
	my @args = @_;
	my $scalar = 1;
	@args > 1
	    and ref $args[-1] eq 'ARRAY'
	    and @{ $args[-1] } == 0
	    and $array{$args[-2]}
	    and $scalar = 0;
	my @nav = ();
	while ( @args ) {
	    if ( __instance( $args[0], 'PPIx::Regexp::Element' ) ) {
		$obj = shift @args;
	    } elsif ( ref $obj eq 'ARRAY' ) {
		my $inx = shift @args;
		push @nav, $inx;
		$obj = $obj->[$inx];
	    } else {
		my $method = shift @args;
		my $args = shift @args;
		ref $args eq 'ARRAY' or $args = [ $args ];
		push @nav, $method, $args;
		$obj->can( $method ) or return;
		if ( @args || $scalar ) {
		    $obj = $obj->$method( @{ $args } ) or return;
		} else {
		    $obj = [ $obj->$method( @{ $args } ) ];
		}
	    }
	}
	$nav = _safe( @nav );
	$nav =~ s/ ' ( \w+ ) ' , /$1 =>/smxg;
	$nav =~ s/ \[ \s+ \] /[]/smxg;
	$result = $obj;
	return $obj;
    }

}

sub parse {		## no critic (RequireArgUnpacking)
    my ( $opt, $regexp, @args ) = _parse_constructor_args(
	{ test => 1 }, @_ );
    $initial_class = 'PPIx::Regexp';
    $kind = 'element';
    $result = $obj = $parse = PPIx::Regexp->new( $regexp, @args );
    $nav = '';
    $opt->{test} or return;
    @_ = ( $parse, 'PPIx::Regexp', $regexp );
    goto &isa_ok;
}

sub ppi {		## no critic (RequireArgUnpacking)
    my @args = @_;
    my $expect = pop @args;
    $result = undef;
    defined $obj and $result = $obj->ppi()->content();
    my $safe;
    if ( defined $result ) {
	($safe = $result) =~ s/([\\'])/\\$1/smxg;
    } else {
	$safe = 'undef';
    }
    @_ = ( $result, $expect, "$kind $nav ppi() content '$safe'" );
    goto &is;
}

sub result {
    return $result;
}

sub tokenize {		## no critic (RequireArgUnpacking)
    my ( $opt, $regexp, @args ) = _parse_constructor_args(
	{ test => 1, tokens => 1 }, @_ );
    $initial_class = 'PPIx::Regexp::Tokenizer';
    $kind = 'token';
    $obj = PPIx::Regexp::Tokenizer->new( $regexp, @args );
    if ( $obj && $opt->{tokens} ) {
	$parse = [ $obj->tokens() ];
    } else {
	$parse = [];
    }
    $result = $parse;
    $nav = '';
    $opt->{test} or return;
    @_ = ( $obj, 'PPIx::Regexp::Tokenizer', $regexp );
    goto &isa_ok;
}

sub true {		## no critic (RequireArgUnpacking)
    my ( $method, $args ) = @_;
    ref $args eq 'ARRAY' or $args = [ $args ];
    my $class = ref $obj;
    if ( $obj->can( $method ) ) {
	$result = $obj->$method( @{ $args } );
	my $fmtd = _format_args( $args );
	@_ = ( $result, "$class->$method$fmtd is true" );
    } else {
	$result = undef;
	@_ = ( undef, "$class->$method() exists" );
    }
    goto &ok;
}

sub value {		## no critic (RequireArgUnpacking)
    my ( $method, $args, $expect ) = @_;
    ref $args eq 'ARRAY' or $args = [ $args ];

    my $invocant = $obj || $initial_class;
    my $class = ref $obj || $obj || $initial_class;
    if ( ! $invocant->can( $method ) ) {
	@_ = ( undef, "$class->$method() exists" );
	goto &ok;
    }

    $result = ref $expect eq 'ARRAY' ?
	[ $invocant->$method( @{ $args } ) ] :
	$invocant->$method( @{ $args } );

    my $fmtd = _format_args( $args );
    my $answer = _format_args( [ $expect ], bare => 1 );
    @_ = ( $result, $expect, "${class}->$method$fmtd is $answer" );
    if ( ref $result ) {
	goto &is_deeply;
    } else {
	goto &is;
    }
}

sub _format_args {
    my ( $args, %opt ) = @_;
    my @rslt;
    foreach my $arg ( @{ $args } ) {
	if ( ! defined $arg ) {
	    push @rslt, 'undef';
	} elsif ( looks_like_number( $arg ) ) {
	    push @rslt, $arg;
	} else {
	    push @rslt, $arg;
	    $rslt[-1] =~ s/ ' /\\'/smxg;
	    $rslt[-1] = "'$rslt[-1]'";
	}
    }
    my $string = join ', ', @rslt;
    $opt{bare} and return $string;
    @rslt or return '()';
    return "( $string )";
}

sub _parse_constructor_args {
    my ( $opt, @args ) = @_;
    my @rslt = ( $opt );
    foreach my $arg ( @args ) {
	if ( $arg =~ m/ \A - -? (no)? (\w+) \z /smx &&
	    exists $opt->{$2} ) {
	    $opt->{$2} = !$1;
	} else {
	    push @rslt, $arg;
	}
    }
    return @rslt;
}

sub _pause {
    if ( eval { require Time::HiRes; 1 } ) {	# Cargo cult programming.
	Time::HiRes::sleep( 0.1 );		# Something like this is
    } else {					# in PPI's
	sleep 1;				# t/08_regression.t, and
    }						# who am I to argue?
    return;
}

# quote a string.
sub _safe {
    my @args = @_;
    my @rslt;
    foreach my $item ( @args ) {
	if ( __instance( $item, 'PPIx::Regexp::Element' ) ) {
	    $item = $item->content();
	}
	if ( ! defined $item ) {
	    push @rslt, 'undef';
	} elsif ( ref $item eq 'ARRAY' ) {
	    push @rslt, join( ' ', '[', _safe( @{ $item } ), ']' );
	} elsif ( looks_like_number( $item ) ) {
	    push @rslt, $item;
	} else {
	    $item =~ s/ ( [\\'] ) /\\$1/smxg;
	    push @rslt, "'$item'";
	}
    }
    return join( ', ', @rslt );
}

1;

__END__

=head1 NAME

PPIx::Regexp::Test - support for testing PPIx::Regexp

=head1 SYNOPSIS

 use lib qw{ inc };
 use PPIx::Regexp::Test;

 parse   ( '/foo/' );
 value   ( failures => [], 0 );
 class   ( 'PPIx::Regexp' );
 choose  ( child => 0 );
 class   ( 'PPIx::Regexp::Token::Structure' );
 content ( '' );
 # and so on

=head1 DETAILS

This package exports various subroutines in support of testing
C<PPIx::Regexp>. Most of these are tests, with C<Test::More> doing the
dirty work. A few simply set up data for tests.

The whole test rig works by parsing (or tokenizing) a regular
expression, followed by a series of unit tests on the results of the
parse. Each set of unit tests is performed by selecting an object to
test using the C<choose> or C<navigate> subroutine, followed by the
tests to be performed on that object. A few tests do not test parse
objects, but rather the state of the system as a whole.

The following subroutines are exported:

=head2 cache_count

 cache_count( 1 );

This test compares the number of objects in the C<new_from_cache> cache
to its argument, succeeding if they are equal. If no argument is passed,
the default is 0.

=head2 choose

 choose( 2 );  # For tokenizer
 choose( child => 1, child => 2, type => 0 ); # For full parse

This subroutine does not itself represent a test. It chooses an object
from the parse tree for further testing. If testing a tokenizer, the
argument is the token number (from 0) to select. If testing a full
parse, the arguments are the navigation methods used to reach the
object to be tested, starting from the C<PPIx::Regexp> object. The
arguments to the methods are passed in an array reference, but if there
is a single argument it can be passed as a scalar, as in the example.

=head2 class

 class( 'PPIx::Regexp::Token::Structure' );

This test checks to see if the current object is of the given class, and
succeeds if it is. If the current object is C<undef>, the test fails.

=head2 content

 content( '\N{LATIN SMALL LETTER A}' );

This test checks to see if the C<content> method of the current object
is equal to the given string. If the current object is C<undef>, the
test fails.

=head2 cmp_ok

This subroutine is exported from L<Test::More|Test::More>.

=head2 count

 count( 42 );

This test checks the number of objects returned by an operation that
returns more than one object. It succeeds if the number of objects
returned is equal to the given number.

This test is valid only after C<tokenize>, or a C<choose> or C<navigate>
whose argument list ends in one of

 children => []
 finish => []
 start => []
 type => []

=head2 different

 different( $o1, $o2, 'Test name' );

This test compares two things, succeeding if they are different.
References are compared by reference address and scalars by value
(numeric or string comparison as appropriate). If the first argument is
omitted it defaults to the current object.

=head2 dump_result

 dump_result( tokens => 1, <<'EOD', 'Test tokenization dump' );
 ... expected dump here ...
 EOD

This test performs the specified dump on the current object and succeeds
if the result matches the expectation. The name of the test is the last
argument, and the expected result is the next-to-last argument. All
other arguments are passed to
L<PPIx::Regexp::Dumper|PPIx::Regexp::Dumper>.

Well, almost all other arguments are passed to the dumper. You can
specify C<--notest> to skip the test. In this case the result of the
last operation is dumped. L<PPIx::Regexp::Dumper|PPIx::Regexp::Dumper>
is used if appropriate; otherwise you get a L<YAML|YAML> dump if that is
available, or a L<Data::Dumper|Data::Dumper> dump if not. If no dumper
class can be found, a diagnostic is produced. You can also specify
C<--test>, but this is the default. This option is removed from the
argument list before the test name (etc) is determined.

=head2 equals

 equals( $o1, $o2, 'Test name' );

This test compares two things, succeeding if they are equal. References
are compared by reference address and scalars by value (numeric or string
comparison as appropriate). If the first argument is omitted it defaults
to the current object.

=head2 false

 false( significant => [] );

This test succeeds if the given method, with the given arguments, called
on the current object, returns a false value.

=head2 finis

 finis();

This test should be last in a series, and no references to parse objects
should be held when it is run. It checks the number of objects in the
internal C<%parent> hash, and succeeds if it is zero.

=head2 navigate

 navigate( snext_sibling => [] );

Like C<choose>, this is not a test, but selects an object for testing.
Unlike C<choose>, selection starts from the current object, not the top
of the parse tree.

=head2 parse

 parse( 's/foo/bar/g' );

This test parses the given regular expression into a C<PPIx::Regexp>
object, and succeeds if a C<PPIx::Regexp> object was in fact generated.

If you specify argument C<--notest>, the parse is done but no test is
performed. You would do this if you expected the parse to fail (e.g. you
are testing error handling). You can also explicitly specify C<--test>,
but this is the default.

All other arguments are passed to the L<PPIx::Regexp|PPIx::Regexp>
constructor.

=head2 plan

This subroutine is exported from L<Test::More|Test::More>.

=head2 content

 ppi( '$foo' );

This test calls the current object's C<ppi()> method, and checks to see
if the content of the returned L<PPI::Document|PPI::Document> is equal
to the given string. If the current object is C<undef> or does not have
a C<ppi()> method, the test fails.

=head2 result

 my $val = result();

This subroutine returns the result of the most recent operation that
actually produces one. It should be called immediately after the
operation, mostly because I have not documented all the subroutines that
produce a result.

=head2 tokenize

 tokenize( 'm/foo/smx' );

This test tokenizes the given regular expression into a
C<PPIx::Regexp::Tokenizer> object, and succeeds if a
C<PPIx::Regexp::Tokenizer> object was in fact generated.

If you specify argument C<--notest>, the parse is done but no test is
performed. You would do this if you expected the parse to fail (e.g. you
are testing error handling). You can also explicitly specify C<--test>,
but this is the default.

If you specify argument C<--notokens>, the tokenizer is built, but the
tokens are not extracted. You would do this when you want a subsequent
operation to call C<tokens()>. You can also explicitly specify
C<--tokens>, but this is the default.

All other arguments are passed to the
L<PPIx::Regexp::Tokenizer|PPIx::Regexp::Tokenizer> constructor.

=head2 true

 true( significant => [] );

This test succeeds if the given method, with the given arguments, called
on the current object, returns a true value.

=head2 value

 value( max_capture_number => [], 3 );

This test succeeds if the given method, with the given arguments, called
on the current object, returns the given value.

If the current object is undefined, the given method is called on the
intended initial class, otherwise there would be no way to test the
errstr() method.

The result of the method call is accessable via the L<result()|/result>
subroutine.

=head1 SUPPORT

Support is by the author. Please file bug reports at
L<http://rt.cpan.org>, or in electronic mail to the author.

=head1 AUTHOR

Thomas R. Wyant, III F<wyant at cpan dot org>

=head1 COPYRIGHT AND LICENSE

Copyright (C) 2009-2015 by Thomas R. Wyant, III

This program is free software; you can redistribute it and/or modify it
under the same terms as Perl 5.10.0. For more details, see the full text
of the licenses in the directory LICENSES.

This program is distributed in the hope that it will be useful, but
without any warranty; without even the implied warranty of
merchantability or fitness for a particular purpose.

=cut

# ex: set textwidth=72 :