Add benchmark test for keys() on empty hashes (RT26188)

1 files changed, 91 insertions, 0 deletions

diff --git a/t/benchmark/rt26188-speed-up-keys-on-empty-hash.t b/t/benchmark/rt26188-speed-up-keys-on-empty-hash.t
new file mode 100644
index 0000000000..43401adfe1
--- /dev/null
+++ b/t/benchmark/rt26188-speed-up-keys-on-empty-hash.t
@@ -0,0 +1,91 @@
+#!/usr/bin/perl -w
+use strict;
+use Benchmark;
+require './test.pl';
+skip_all("Benchmark tests not enabled") unless $ENV{PERL_BENCHMARK};
+plan(tests => 6);
+
+=head1 NAME
+
+rt26188 - benchmark speed for keys() on empty hashes
+
+=head1 DESCRIPTION
+
+If you have an empty hash, the speed of keys() depends
+on how many keys the hash previously held.
+
+For global hashes, getting the count for previously
+big hashes was substantially slower than for lexical hashes.
+
+This test checks that the speed difference for getting
+the number or list of keys from an empty hash is about the same
+(< 25%) for lexical and global hashes, both previously big and small.
+
+=head1 REFERENCE
+
+This test tests against RT ticket #26188
+
+L<http://rt.perl.org/rt3/Public/Bug/Display.html?id=26188>
+
+=cut
+
+use vars qw(%h_big %h_small);
+my %l_big = (1..50000);
+my %l_small = (1..10);
+
+%h_big = (1..50000);
+%h_small = (1..10);
+
+delete @h_big{keys %h_big};
+delete @h_small{keys %h_small};
+delete @l_big{keys %l_big};
+delete @l_small{keys %l_small};
+
+my $res = timethese shift || -3, {
+    big => '1 for keys %h_big',
+    small => '1 for keys %h_small',
+    scalar_big => '$a = keys %h_big',
+    scalar_small => '$a = keys %h_small',
+
+    lex_big => '1 for keys %l_big',
+    lex_small => '1 for keys %l_small',
+    lex_scalar_big => '$a = keys %l_big',
+    lex_scalar_small => '$a = keys %l_small',
+}, 'none';
+
+sub iters_per_second {
+    $_[0]->iters / $_[0]->cpu_p
+}
+
+sub about_as_fast_ok {
+    my ($res, $key1, $key2, $name) = @_;
+    $name ||= "Speed difference between $key1 and $key2 is less than 25%";
+    my %iters_per_second = map { $_ => iters_per_second( $res->{ $_ }) } ($key1, $key2);
+
+    my $ratio = abs(1 - $iters_per_second{ $key1 } / ($iters_per_second{ $key2 } || 1 ));
+    if (! cmp_ok( $ratio, '<', 0.25, $name )) {
+        diag( sprintf "%20s: %12.2f/s\n", $key1, $iters_per_second{ $key1 } );
+        diag( sprintf "%20s: %12.2f/s\n", $key2, $iters_per_second{ $key2 } );
+    };
+};
+
+about_as_fast_ok( $res, 'scalar_big', 'scalar_small',"Checking the count of hash keys in an empty hash (global)");
+
+about_as_fast_ok( $res, 'big', 'small', "Checking the list of hash keys in an empty hash (global)");
+
+about_as_fast_ok( $res, 'lex_scalar_big', 'lex_scalar_small',"Checking the count of hash keys in an empty hash (lexical)");
+
+about_as_fast_ok( $res, 'lex_big', 'lex_small', "Checking the list of hash keys in an empty hash (lexical)");
+
+about_as_fast_ok( $res, 'lex_scalar_big', 'scalar_big',"Checking the count of hash keys in an empty hash, global vs. lexical");
+
+about_as_fast_ok( $res, 'lex_big', 'big', "Checking the list of hash keys in an empty hash, global vs. lexical");
+
+__END__
+
+# code written
+    /* quick bailout if the hash is empty anyway.
+       I don't know if placeholders are included in the KEYS count, so a defensive check
+    */
+    if (! HvKEYS(hv) && !(flags & HV_ITERNEXT_WANTPLACEHOLDERS) )
+        return NULL;