Test/Harness/Straps.pm

# -*- Mode: cperl; cperl-indent-level: 4 -*-
package Test::Harness::Straps;

use strict;
use vars qw($VERSION);
$VERSION = '0.26';

use Config;
use Test::Harness::Assert;
use Test::Harness::Iterator;
use Test::Harness::Point;

# Flags used as return values from our methods.  Just for internal
# clarification.
my $YES   = (1==1);
my $NO    = !$YES;

=head1 NAME

Test::Harness::Straps - detailed analysis of test results

=head1 SYNOPSIS

  use Test::Harness::Straps;

  my $strap = Test::Harness::Straps->new;

  # Various ways to interpret a test
  my %results = $strap->analyze($name, \@test_output);
  my %results = $strap->analyze_fh($name, $test_filehandle);
  my %results = $strap->analyze_file($test_file);

  # UNIMPLEMENTED
  my %total = $strap->total_results;

  # Altering the behavior of the strap  UNIMPLEMENTED
  my $verbose_output = $strap->dump_verbose();
  $strap->dump_verbose_fh($output_filehandle);


=head1 DESCRIPTION

B<THIS IS ALPHA SOFTWARE> in that the interface is subject to change
in incompatible ways.  It is otherwise stable.

Test::Harness is limited to printing out its results.  This makes
analysis of the test results difficult for anything but a human.  To
make it easier for programs to work with test results, we provide
Test::Harness::Straps.  Instead of printing the results, straps
provide them as raw data.  You can also configure how the tests are to
be run.

The interface is currently incomplete.  I<Please> contact the author
if you'd like a feature added or something change or just have
comments.

=head1 CONSTRUCTION

=head2 new()

  my $strap = Test::Harness::Straps->new;

Initialize a new strap.

=cut

sub new {
    my $class = shift;
    my $self  = bless {}, $class;

    $self->_init;

    return $self;
}

=for private $strap->_init

  $strap->_init;

Initialize the internal state of a strap to make it ready for parsing.

=cut

sub _init {
    my($self) = shift;

    $self->{_is_vms}   = ( $^O eq 'VMS' );
    $self->{_is_win32} = ( $^O =~ /^(MS)?Win32$/ );
    $self->{_is_macos} = ( $^O eq 'MacOS' );
}

=head1 ANALYSIS

=head2 $strap->analyze( $name, \@output_lines )

    my %results = $strap->analyze($name, \@test_output);

Analyzes the output of a single test, assigning it the given C<$name>
for use in the total report.  Returns the C<%results> of the test.
See L<Results>.

C<@test_output> should be the raw output from the test, including
newlines.

=cut

sub analyze {
    my($self, $name, $test_output) = @_;

    my $it = Test::Harness::Iterator->new($test_output);
    return $self->_analyze_iterator($name, $it);
}


sub _analyze_iterator {
    my($self, $name, $it) = @_;

    $self->_reset_file_state;
    $self->{file} = $name;
    my %totals  = (
                   max      => 0,
                   seen     => 0,

                   ok       => 0,
                   todo     => 0,
                   skip     => 0,
                   bonus    => 0,

                   details  => []
                  );

    # Set them up here so callbacks can have them.
    $self->{totals}{$name}         = \%totals;
    while( defined(my $line = $it->next) ) {
        $self->_analyze_line($line, \%totals);
        last if $self->{saw_bailout};
    }

    $totals{skip_all} = $self->{skip_all} if defined $self->{skip_all};

    my $passed = ($totals{max} == 0 && defined $totals{skip_all}) ||
                 ($totals{max} && $totals{seen} &&
                  $totals{max} == $totals{seen} &&
                  $totals{max} == $totals{ok});
    $totals{passing} = $passed ? 1 : 0;

    return %totals;
}


sub _analyze_line {
    my $self = shift;
    my $line = shift;
    my $totals = shift;

    $self->{line}++;

    my $linetype;
    my $point = Test::Harness::Point->from_test_line( $line );
    if ( $point ) {
        $linetype = 'test';

        $totals->{seen}++;
        $point->set_number( $self->{'next'} ) unless $point->number;

        # sometimes the 'not ' and the 'ok' are on different lines,
        # happens often on VMS if you do:
        #   print "not " unless $test;
        #   print "ok $num\n";
        if ( $self->{lone_not_line} && ($self->{lone_not_line} == $self->{line} - 1) ) {
            $point->set_ok( 0 );
        }

        if ( $self->{todo}{$point->number} ) {
            $point->set_directive_type( 'todo' );
        }

        if ( $point->is_todo ) {
            $totals->{todo}++;
            $totals->{bonus}++ if $point->ok;
        }
        elsif ( $point->is_skip ) {
            $totals->{skip}++;
        }

        $totals->{ok}++ if $point->pass;

        if ( ($point->number > 100_000) && ($point->number > ($self->{max}||100_000)) ) {
            if ( !$self->{too_many_tests}++ ) {
                warn "Enormous test number seen [test ", $point->number, "]\n";
                warn "Can't detailize, too big.\n";
            }
        }
        else {
            my $details = {
                ok          => $point->pass,
                actual_ok   => $point->ok,
                name        => _def_or_blank( $point->description ),
                type        => _def_or_blank( $point->directive_type ),
                reason      => _def_or_blank( $point->directive_reason ),
            };

            assert( defined( $details->{ok} ) && defined( $details->{actual_ok} ) );
            $totals->{details}[$point->number - 1] = $details;
        }
    } # test point
    elsif ( $line =~ /^not\s+$/ ) {
        $linetype = 'other';
        # Sometimes the "not " and "ok" will be on separate lines on VMS.
        # We catch this and remember we saw it.
        $self->{lone_not_line} = $self->{line};
    }
    elsif ( $self->_is_header($line) ) {
        $linetype = 'header';

        $self->{saw_header}++;

        $totals->{max} += $self->{max};
    }
    elsif ( $self->_is_bail_out($line, \$self->{bailout_reason}) ) {
        $linetype = 'bailout';
        $self->{saw_bailout} = 1;
    }
    elsif (my $diagnostics = $self->_is_diagnostic_line( $line )) {
        $linetype = 'other';
        my $test = $totals->{details}[-1];
        $test->{diagnostics} ||=  '';
        $test->{diagnostics}  .= $diagnostics;
    }
    else {
        $linetype = 'other';
    }

    $self->{callback}->($self, $line, $linetype, $totals) if $self->{callback};

    $self->{'next'} = $point->number + 1 if $point;
} # _analyze_line


sub _is_diagnostic_line {
    my ($self, $line) = @_;
    return if index( $line, '# Looks like you failed' ) == 0;
    $line =~ s/^#\s//;
    return $line;
}

=for private $strap->analyze_fh( $name, $test_filehandle )

    my %results = $strap->analyze_fh($name, $test_filehandle);

Like C<analyze>, but it reads from the given filehandle.

=cut

sub analyze_fh {
    my($self, $name, $fh) = @_;

    my $it = Test::Harness::Iterator->new($fh);
    return $self->_analyze_iterator($name, $it);
}

=head2 $strap->analyze_file( $test_file )

    my %results = $strap->analyze_file($test_file);

Like C<analyze>, but it runs the given C<$test_file> and parses its
results.  It will also use that name for the total report.

=cut

sub analyze_file {
    my($self, $file) = @_;

    unless( -e $file ) {
        $self->{error} = "$file does not exist";
        return;
    }

    unless( -r $file ) {
        $self->{error} = "$file is not readable";
        return;
    }

    local $ENV{PERL5LIB} = $self->_INC2PERL5LIB;
    if ( $Test::Harness::Debug ) {
        local $^W=0; # ignore undef warnings
        print "# PERL5LIB=$ENV{PERL5LIB}\n";
    }

    # *sigh* this breaks under taint, but open -| is unportable.
    my $line = $self->_command_line($file);

    unless ( open(FILE, "$line|" )) {
        print "can't run $file. $!\n";
        return;
    }

    my %results = $self->analyze_fh($file, \*FILE);
    my $exit    = close FILE;
    $results{'wait'} = $?;
    if( $? && $self->{_is_vms} ) {
        eval q{use vmsish "status"; $results{'exit'} = $?};
    }
    else {
        $results{'exit'} = _wait2exit($?);
    }
    $results{passing} = 0 unless $? == 0;

    $self->_restore_PERL5LIB();

    return %results;
}


eval { require POSIX; &POSIX::WEXITSTATUS(0) };
if( $@ ) {
    *_wait2exit = sub { $_[0] >> 8 };
}
else {
    *_wait2exit = sub { POSIX::WEXITSTATUS($_[0]) }
}

=for private $strap->_command_line( $file )

Returns the full command line that will be run to test I<$file>.

=cut

sub _command_line {
    my $self = shift;
    my $file = shift;

    my $command =  $self->_command();
    my $switches = $self->_switches($file);

    $file = qq["$file"] if ($file =~ /\s/) && ($file !~ /^".*"$/);
    my $line = "$command $switches $file";

    return $line;
}


=for private $strap->_command()

Returns the command that runs the test.  Combine this with C<_switches()>
to build a command line.

Typically this is C<$^X>, but you can set C<$ENV{HARNESS_PERL}>
to use a different Perl than what you're running the harness under.
This might be to run a threaded Perl, for example.

You can also overload this method if you've built your own strap subclass,
such as a PHP interpreter for a PHP-based strap.

=cut

sub _command {
    my $self = shift;

    return $ENV{HARNESS_PERL}   if defined $ENV{HARNESS_PERL};
    return qq["$^X"]            if $self->{_is_win32} && ($^X =~ /[^\w\.\/\\]/);
    return $^X;
}


=for private $strap->_switches( $file )

Formats and returns the switches necessary to run the test.

=cut

sub _switches {
    my($self, $file) = @_;

    my @existing_switches = $self->_cleaned_switches( $Test::Harness::Switches, $ENV{HARNESS_PERL_SWITCHES} );
    my @derived_switches;

    local *TEST;
    open(TEST, $file) or print "can't open $file. $!\n";
    my $shebang = <TEST>;
    close(TEST) or print "can't close $file. $!\n";

    my $taint = ( $shebang =~ /^#!.*\bperl.*\s-\w*([Tt]+)/ );
    push( @derived_switches, "-$1" ) if $taint;

    # When taint mode is on, PERL5LIB is ignored.  So we need to put
    # all that on the command line as -Is.
    # MacPerl's putenv is broken, so it will not see PERL5LIB, tainted or not.
    if ( $taint || $self->{_is_macos} ) {
	my @inc = $self->_filtered_INC;
	push @derived_switches, map { "-I$_" } @inc;
    }

    # Quote the argument if there's any whitespace in it, or if
    # we're VMS, since VMS requires all parms quoted.  Also, don't quote
    # it if it's already quoted.
    for ( @derived_switches ) {
	$_ = qq["$_"] if ((/\s/ || $self->{_is_vms}) && !/^".*"$/ );
    }
    return join( " ", @existing_switches, @derived_switches );
}

=for private $strap->_cleaned_switches( @switches_from_user )

Returns only defined, non-blank, trimmed switches from the parms passed.

=cut

sub _cleaned_switches {
    my $self = shift;

    local $_;

    my @switches;
    for ( @_ ) {
	my $switch = $_;
	next unless defined $switch;
	$switch =~ s/^\s+//;
	$switch =~ s/\s+$//;
	push( @switches, $switch ) if $switch ne "";
    }

    return @switches;
}

=for private $strap->_INC2PERL5LIB

  local $ENV{PERL5LIB} = $self->_INC2PERL5LIB;

Takes the current value of C<@INC> and turns it into something suitable
for putting onto C<PERL5LIB>.

=cut

sub _INC2PERL5LIB {
    my($self) = shift;

    $self->{_old5lib} = $ENV{PERL5LIB};

    return join $Config{path_sep}, $self->_filtered_INC;
}

=for private $strap->_filtered_INC()

  my @filtered_inc = $self->_filtered_INC;

Shortens C<@INC> by removing redundant and unnecessary entries.
Necessary for OSes with limited command line lengths, like VMS.

=cut

sub _filtered_INC {
    my($self, @inc) = @_;
    @inc = @INC unless @inc;

    if( $self->{_is_vms} ) {
	# VMS has a 255-byte limit on the length of %ENV entries, so
	# toss the ones that involve perl_root, the install location
        @inc = grep !/perl_root/i, @inc;

    }
    elsif ( $self->{_is_win32} ) {
	# Lose any trailing backslashes in the Win32 paths
	s/[\\\/+]$// foreach @inc;
    }

    my %seen;
    $seen{$_}++ foreach $self->_default_inc();
    @inc = grep !$seen{$_}++, @inc;

    return @inc;
}


{ # Without caching, _default_inc() takes a huge amount of time
    my %cache;
    sub _default_inc {
        my $self = shift;
        my $perl = $self->_command;
        $cache{$perl} ||= [do {
            local $ENV{PERL5LIB};
            my @inc =`$perl -le "print join qq[\\n], \@INC"`;
            chomp @inc;
        }];
        return @{$cache{$perl}};
    }
}


=for private $strap->_restore_PERL5LIB()

  $self->_restore_PERL5LIB;

This restores the original value of the C<PERL5LIB> environment variable.
Necessary on VMS, otherwise a no-op.

=cut

sub _restore_PERL5LIB {
    my($self) = shift;

    return unless $self->{_is_vms};

    if (defined $self->{_old5lib}) {
        $ENV{PERL5LIB} = $self->{_old5lib};
    }
}

=head1 Parsing

Methods for identifying what sort of line you're looking at.

=for private _is_diagnostic

    my $is_diagnostic = $strap->_is_diagnostic($line, \$comment);

Checks if the given line is a comment.  If so, it will place it into
C<$comment> (sans #).

=cut

sub _is_diagnostic {
    my($self, $line, $comment) = @_;

    if( $line =~ /^\s*\#(.*)/ ) {
        $$comment = $1;
        return $YES;
    }
    else {
        return $NO;
    }
}

=for private _is_header

  my $is_header = $strap->_is_header($line);

Checks if the given line is a header (1..M) line.  If so, it places how
many tests there will be in C<< $strap->{max} >>, a list of which tests
are todo in C<< $strap->{todo} >> and if the whole test was skipped
C<< $strap->{skip_all} >> contains the reason.

=cut

# Regex for parsing a header.  Will be run with /x
my $Extra_Header_Re = <<'REGEX';
                       ^
                        (?: \s+ todo \s+ ([\d \t]+) )?      # optional todo set
                        (?: \s* \# \s* ([\w:]+\s?) (.*) )?     # optional skip with optional reason
REGEX

sub _is_header {
    my($self, $line) = @_;

    if( my($max, $extra) = $line =~ /^1\.\.(\d+)(.*)/ ) {
        $self->{max}  = $max;
        assert( $self->{max} >= 0,  'Max # of tests looks right' );

        if( defined $extra ) {
            my($todo, $skip, $reason) = $extra =~ /$Extra_Header_Re/xo;

            $self->{todo} = { map { $_ => 1 } split /\s+/, $todo } if $todo;

            if( $self->{max} == 0 ) {
                $reason = '' unless defined $skip and $skip =~ /^Skip/i;
            }

            $self->{skip_all} = $reason;
        }

        return $YES;
    }
    else {
        return $NO;
    }
}

=for private _is_bail_out

  my $is_bail_out = $strap->_is_bail_out($line, \$reason);

Checks if the line is a "Bail out!".  Places the reason for bailing
(if any) in $reason.

=cut

sub _is_bail_out {
    my($self, $line, $reason) = @_;

    if( $line =~ /^Bail out!\s*(.*)/i ) {
        $$reason = $1 if $1;
        return $YES;
    }
    else {
        return $NO;
    }
}

=for private _reset_file_state

  $strap->_reset_file_state;

Resets things like C<< $strap->{max} >> , C<< $strap->{skip_all} >>,
etc. so it's ready to parse the next file.

=cut

sub _reset_file_state {
    my($self) = shift;

    delete @{$self}{qw(max skip_all todo too_many_tests)};
    $self->{line}       = 0;
    $self->{saw_header} = 0;
    $self->{saw_bailout}= 0;
    $self->{lone_not_line} = 0;
    $self->{bailout_reason} = '';
    $self->{'next'}       = 1;
}

=head1 Results

The C<%results> returned from C<analyze()> contain the following
information:

  passing           true if the whole test is considered a pass
                    (or skipped), false if its a failure

  exit              the exit code of the test run, if from a file
  wait              the wait code of the test run, if from a file

  max               total tests which should have been run
  seen              total tests actually seen
  skip_all          if the whole test was skipped, this will
                      contain the reason.

  ok                number of tests which passed
                      (including todo and skips)

  todo              number of todo tests seen
  bonus             number of todo tests which
                      unexpectedly passed

  skip              number of tests skipped

So a successful test should have max == seen == ok.


There is one final item, the details.

  details           an array ref reporting the result of
                    each test looks like this:

    $results{details}[$test_num - 1] =
            { ok          => is the test considered ok?
              actual_ok   => did it literally say 'ok'?
              name        => name of the test (if any)
              diagnostics => test diagnostics (if any)
              type        => 'skip' or 'todo' (if any)
              reason      => reason for the above (if any)
            };

Element 0 of the details is test #1.  I tried it with element 1 being
#1 and 0 being empty, this is less awkward.

=head1 EXAMPLES

See F<examples/mini_harness.plx> for an example of use.

=head1 AUTHOR

Michael G Schwern C<< <schwern at pobox.com> >>, currently maintained by
Andy Lester C<< <andy at petdance.com> >>.

=head1 SEE ALSO

L<Test::Harness>

=cut

sub _def_or_blank {
    return $_[0] if defined $_[0];
    return "";
}

1;