perl/t/harness

#!./perl

# We suppose that perl _mostly_ works at this moment, so may use
# sophisticated testing.

BEGIN {
    chdir 't' if -d 't';
    @INC = '../lib';              # pick up only this build's lib
}

##############################################################################
# Test files which cannot be executed at the same time.
#
# List all files which might fail when executed at the same time as another
# test file from the same test directory. Being listed here does not mean
# the test will be run by itself, it just means it won't be run at the same
# time as any other file in the same test directory, it might be run at the
# same time as a file from a different test directory.
#
# Ideally this is always empty.
#
# Example: ../cpan/IO-Zlib/t/basic.t
#
my @_must_be_executed_serially = (
    # These two both create temporary subdirectories which they delete
    # at the end. If one deletes while the other is running a recursive
    # find in that subdir, bad things can happen. This was showing as
    # random crashes in find.t and taint.t in smokes, with errors like:
    # "Can't cd to .. from ./FF_find_t_RKdkBE/for_find/fb: Stale file handle"
    '../ext/File-Find/t/taint.t',
    '../ext/File-Find/t/find.t',
);

my %must_be_executed_serially = map { $_ => 1 } @_must_be_executed_serially;
##############################################################################

##############################################################################
# Test files which must be executed alone.
#
# List files which cannot be run at the same time as any other test. Typically
# this is used to handle tests which are sensitive to load and which might
# fail if they were run at the same time as something load intensive.
#
# Example: ../dist/threads-shared/t/waithires.t
#
my @_must_be_executed_alone = qw();
my %must_be_executed_alone = map { $_ => 1 } @_must_be_executed_alone;

my $OS = $ENV{FAKE_OS} || $^O;
my $is_linux = $OS eq "linux";
my $is_win32 = $OS eq "MSWin32";

if (!$is_linux) {
    $must_be_executed_alone{"../dist/threads-shared/t/waithires.t"} = 1;
}
##############################################################################

my $torture; # torture testing?

use TAP::Harness 3.13;
use strict;
use Config;

$::do_nothing = $::do_nothing = 1;
require './TEST';
our $Valgrind_Log;

my $Verbose = 0;
$Verbose++ while @ARGV && $ARGV[0] eq '-v' && shift;

# For valgrind summary output
my $htoolnm;
my $hgrind_ct;

my $dump_tests = 0;
if ($ARGV[0] && $ARGV[0] =~ /^-?-dumptests$/) {
    shift;
    $dump_tests = 1;
}

if ($ARGV[0] && $ARGV[0] =~ /^-?-torture$/) {
    shift;
    $torture = 1;
}

# Let tests know they're running in the perl core.  Useful for modules
# which live dual lives on CPAN.
$ENV{PERL_CORE} = 1;

my (@tests, @re, @anti_re);

# [.VMS]TEST.COM calls harness with empty arguments, so clean-up @ARGV
@ARGV = grep $_ && length( $_ ) => @ARGV;

while ($ARGV[0] && $ARGV[0]=~/^-?-(n?)re/) {
    my $ary= $1 ? \@anti_re : \@re;

    if ( $ARGV[0] !~ /=/ ) {
        shift @ARGV;
        while (@ARGV and $ARGV[0] !~ /^-/) {
            push @$ary, shift @ARGV;
        }
    } else {
        push @$ary, (split/=/,shift @ARGV)[1];
    }
}

my $jobs = $ENV{TEST_JOBS};
my ($rules, $state, $color);

if ($ENV{HARNESS_OPTIONS}) {
    for my $opt ( split /:/, $ENV{HARNESS_OPTIONS} ) {
        if ( $opt =~ /^j(\d*)$/ ) {
            $jobs ||= $1 || 9;
        }
        elsif ( $opt eq 'c' ) {
            $color = 1;
        }
        else {
            die "Unknown HARNESS_OPTIONS item: $opt\n";
        }
    }
}

$jobs ||= 1;

my %total_time;
sub _compute_tests_and_ordering($) {
    my @tests = $_[0]->@*;

    my %dir;
    my %all_dirs;
    my %map_file_to_dir;

    if (!$dump_tests) {
        require App::Prove::State;
        if (!$state) {
            # silence unhelpful warnings from App::Prove::State about not having
            # a save state, unless we actually set the PERL_TEST_STATE we don't care
            # and we don't need to know if its fresh or not.
            local $SIG{__WARN__} = $ENV{PERL_TEST_STATE} ? $SIG{__WARN__} : sub {
                return if $_[0] and $_[0]=~/No saved state/;
                warn $_[0];
            };
            my $state_file = $ENV{PERL_TEST_STATE_FILE} // 'test_state';
            if ($state_file) { # set PERL_TEST_STATE_FILE to 0 to skip this
                $state = App::Prove::State->new({ store => $state_file });
                $state->apply_switch('save');
                $state->apply_switch('slow') if $jobs > 1;
            }
        }
        # For some reason get_tests returns *all* the tests previously run,
        # (in the right order), not simply the selection in @tests
        # (in the right order). Not sure if this is a bug or a feature.
        # Whatever, *we* are only interested in the ones that are in @tests
        my %seen;
        @seen{@tests} = ();
        @tests = grep {exists $seen{$_} } $state->get_tests(0, @tests);
    }

    my %times;
    if ($state) {
        # Where known, collate the elapsed times by test name
        foreach ($state->results->tests()) {
            $times{$_->name} = $_->elapsed();
        }
    }

    my %partial_serials;
    # Preprocess the list of tests
    for my $file (@tests) {
        if ($is_win32) {
            $file =~ s,\\,/,g; # canonicalize path
        };

        # Keep a list of the distinct directory names, and another list of
        if ($file =~ m! \A ( (?: \.\. / )?
                                .*?
                            )             # $1 is the directory path name
                            /
                            ( [^/]* \. (?: t | pl ) ) # $2 is the test name
                        \z !x)
        {
            my $path = $1;
            my $name = $2;

            $all_dirs{$path} = 1;
            $map_file_to_dir{$file} = $path;
            # is this a file that requires we do special processing
            # on the directory as a whole?
            if ($must_be_executed_serially{$file}) {
                $partial_serials{$path} = 1;
            }
        }
    }

    my %split_partial_serials;

    my @alone_files;
    # Ready to figure out the timings.
    for my $file (@tests) {
        my $file_dir = $map_file_to_dir{$file};

        # if this is a file which must be processed alone
        if ($must_be_executed_alone{$file}) {
            push @alone_files, $file;
            next;
        }

        # Special handling is needed for a directory that has some test files
        # to execute serially, and some to execute in parallel.  This loop
        # gathers information that a later loop will process.
        if (defined $partial_serials{$file_dir}) {
            if ($must_be_executed_serially{$file}) {
                # This is a file to execute serially.  Its time contributes
                # directly to the total time for this directory.
                $total_time{$file_dir} += $times{$file} || 0;

                # Save the sequence number with the file for now; below we
                # will come back to it.
                push $split_partial_serials{$file_dir}{seq}->@*, [ $1, $file ];
            }
            else {
                # This is a file to execute in parallel after all the
                # sequential ones are done.  Save its time in the hash to
                # later calculate its time contribution.
                push $split_partial_serials{$file_dir}{par}->@*, $file;
                $total_time{$file} = $times{$file} || 0;
            }
        }
        else {
            # Treat every file in each non-serial directory as its own
            # "directory", so that it can be executed in parallel
            $dir{$file} = { seq => $file };
            $total_time{$file} = $times{$file} || 0;
        }
    }

    undef %all_dirs;

    # Here, everything is complete except for the directories that have both
    # serial components and parallel components.  The loop just above gathered
    # the information required to finish setting those up, which we now do.
    for my $partial_serial_dir (keys %split_partial_serials) {

        # Look at just the serial portion for now.
        my @seq_list = $split_partial_serials{$partial_serial_dir}{seq}->@*;

        # The 0th element contains the sequence number; the 1th element the
        # file name.  Get the name, sorted first by the number, then by the
        # name.  Doing it this way allows sequence numbers to be varying
        # length, and still get a numeric sort
        my @sorted_seq_list = map { $_->[1] }
                                sort {    $a->[0] <=>    $b->[0]
                                    or lc $a->[1] cmp lc $b->[1] } @seq_list;

        # Now look at the tests to run in parallel.  Sort in descending order
        # of execution time.
        my @par_list = sort sort_by_execution_order
                        $split_partial_serials{$partial_serial_dir}{par}->@*;

        # The total time to execute this directory is the serial time (already
        # calculated in the previous loop) plus the parallel time.  To
        # calculate an approximate parallel time, note that the minimum
        # parallel time is the maximum of each of the test files run in
        # parallel.  If the number of parallel jobs J is more than the number
        # of such files, N, it could be that all N get executed in parallel,
        # so that maximum is the actual value.  But if N > J, a second, or
        # third, ...  round will be required.  The code below just takes the
        # longest-running time for each round and adds that to the previous
        # total.  It is an imperfect estimate, but not unreasonable.
        my $par_time = 0;
        for (my $i = 0; $i < @par_list; $i += $jobs) {
            $par_time += $times{$par_list[$i]} || 0;
        }
        $total_time{$partial_serial_dir} += $par_time;

        # Now construct the rules.  Each of the parallel tests is made into a
        # single element 'seq' structure, like is done for all the other
        # parallel tests.
        @par_list = map { { seq => $_ } } @par_list;

        # Then the directory is ordered to have the sequential tests executed
        # first (serially), then the parallel tests (in parallel)

        $dir{$partial_serial_dir} =
                                { 'seq' => [ { seq => \@sorted_seq_list },
                                             { par => \@par_list        },
                                           ],
                                };
    }

    #print STDERR __LINE__, join "\n", sort sort_by_execution_order keys %dir

    # Generate T::H schedule rules that run the contents of each directory
    # sequentially.
    my @seq = { par => [ map { $dir{$_} } sort sort_by_execution_order
                                                                    keys %dir
                        ]
               };

    # and lastly add in the files which must be run by themselves without
    # any other tests /at all/ running at the same time.
    push @seq, map { +{ seq => $_ } } sort @alone_files if @alone_files;

    return \@seq;
}

sub sort_by_execution_order {
    # Directories, ordered by total time descending then name ascending
    return $total_time{$b} <=> $total_time{$a} || lc $a cmp lc $b;
}

if (@ARGV) {
    # If you want these run in speed order, just use prove

    # Note: we use glob even on *nix and not just on Windows
    # because arguments might be passed in via the TEST_ARGS
    # env var where they wont be expanded by the shell.
    @tests = map(glob($_),@ARGV);
    # This is a hack to force config_heavy.pl to be loaded, before the
    # prep work for running a test changes directory.
    1 if $Config{d_fork};
} else {
    # Ideally we'd get somewhere close to Tux's Oslo rules
    # my $rules = {
    #     par => [
    #         { seq => '../ext/DB_File/t/*' },
    #         { seq => '../ext/IO_Compress_Zlib/t/*' },
    #         { seq => '../lib/ExtUtils/t/*' },
    #         '*'
    #     ]
    # };

    # but for now, run all directories in sequence.

    unless (@tests) {
        my @seq = <base/*.t>;
        push @tests, @seq;

        my (@next, @last);

        # The remaining core tests are either intermixed with the non-core for
        # more parallelism (if PERL_TEST_HARNESS_ASAP is set non-zero) or done
        # after the above basic sanity tests, before any non-core ones.
        my $which = $ENV{PERL_TEST_HARNESS_ASAP} ? \@last : \@next;

        push @$which, qw(comp run cmd);
        push @$which, qw(io re opbasic op op/hook uni mro lib class porting perf test_pl);
        push @$which, 'japh' if $torture or $ENV{PERL_TORTURE_TEST};
        push @$which, 'win32' if $is_win32;
        push @$which, 'benchmark' if $ENV{PERL_BENCHMARK};
        push @$which, 'bigmem' if $ENV{PERL_TEST_MEMORY};

        if (@next) {
            @next = map { glob ("$_/*.t") } @next;
            push @tests, @next;
            push @seq, _compute_tests_and_ordering(\@next)->@*;
        }

        @last = map { glob ("$_/*.t") } @last;

        my ($non_ext, @ext_from_manifest)=
            _tests_from_manifest($Config{extensions}, $Config{known_extensions}, "all");
        push @last, @ext_from_manifest;

        push @seq, _compute_tests_and_ordering(\@last)->@*;
        push @tests, @last;

        $rules = { seq => \@seq };

        foreach my $test (@tests) {
            delete $non_ext->{$test};
        }

        my @in_manifest_but_not_found = sort keys %$non_ext;
        if (@in_manifest_but_not_found) {
            die "There are test files which are in MANIFEST but are not found by the t/harness\n",
                 "directory scanning rules. You should update t/harness line 339 or so.\n",
                 "Files:\n", map { "    $_\n" } @in_manifest_but_not_found;
        }
    }
}
if ($is_win32) {
    s,\\,/,g for @tests;
}
if (@re or @anti_re) {
    my @keepers;
    foreach my $test (@tests) {
        my $keep = 0;
        if (@re) {
            foreach my $re (@re) {
                $keep = 1 if $test=~/$re/;
            }
        } else {
            $keep = 1;
        }
        if (@anti_re) {
            foreach my $anti_re (@anti_re) {
                $keep = 0 if $test=~/$anti_re/;
            }
        }
        if ($keep) {
            push @keepers, $test;
        }
    }
    @tests= @keepers;
}

# Allow e.g., ./perl t/harness t/op/lc.t
for (@tests) {
    if (! -f $_ && !/^\.\./ && -f "../$_") {
        $_ = "../$_";
        s{^\.\./t/}{};
    }
}

dump_tests(\@tests) if $dump_tests;

filter_taint_tests(\@tests);

my %options;

my $type = 'perl';

# Load TAP::Parser now as otherwise it could be required in the short time span
# in which the harness process chdirs into ext/Dist
require TAP::Parser;

my $h = TAP::Harness->new({
    rules       => $rules,
    color       => $color,
    jobs        => $jobs,
    verbosity   => $Verbose,
    timer       => $ENV{HARNESS_TIMER},
    exec        => sub {
        my ($harness, $test) = @_;

        my $options = $options{$test};
        if (!defined $options) {
            $options = $options{$test} = _scan_test($test, $type);
        }

        (local $Valgrind_Log = "$test.valgrind-current") =~ s/^.*\///;

        return [ split ' ', _cmd($options, $type) ];
    },
});

# Print valgrind output after test completes
if ($ENV{PERL_VALGRIND}) {
    $h->callback(
                 after_test => sub {
                     my ($job) = @_;
                     my $test = $job->[0];
                     my $vfile = "$test.valgrind-current";
                     $vfile =~ s/^.*\///;

                     if ( (! -z $vfile) && open(my $voutput, '<', $vfile)) {
                        print "$test: Valgrind output:\n";
                        print "$test: $_" for <$voutput>;
                        close($voutput);
                     }

                     (local $Valgrind_Log = "$test.valgrind-current") =~ s/^.*\///;

                     _check_valgrind(\$htoolnm, \$hgrind_ct, \$test);
                 }
                 );
}

if ($state) {
    $h->callback(
                 after_test => sub {
                     $state->observe_test(@_);
                 }
                 );
    $h->callback(
                 after_runtests => sub {
                     $state->commit(@_);
                 }
                 );
}

$h->callback(
             parser_args => sub {
                 my ($args, $job) = @_;
                 my $test = $job->[0];
                 _before_fork($options{$test});
                 push @{ $args->{switches} }, "-I../../lib";
             }
             );

$h->callback(
             made_parser => sub {
                 my ($parser, $job) = @_;
                 my $test = $job->[0];
                 my $options = delete $options{$test};
                 _after_fork($options);
             }
             );

my $agg = $h->runtests(@tests);
_cleanup_valgrind(\$htoolnm, \$hgrind_ct);
printf "Finished test run at %s.\n", scalar(localtime);
exit($agg->has_errors ? 1 : 0);