pp_sort.c: fix fencepost error in call to av_extend()

In [rt.cpan.org #39196] issue #17496 there is a report
that Tie::File produced spurious blank lines in the file
after

    @tied= sort @tied;

it turns out that this is because Tie::File treats
EXTEND similarly to STORESIZE (which is arguably not
entirely correct, but also not that weird) coupled
with an off by one error in the calls to av_extend()
in pp_sort.

This patch fixes the fencepost error, adds some comments
to av_extend() to make it clear what it is doing, and
adds a test that EXTEND is called by this code with
correct argument.

1 files changed, 21 insertions, 2 deletions

diff --git a/t/op/sort.t b/t/op/sort.t
index d201f00afd..f2e139dff0 100644
--- a/t/op/sort.t
+++ b/t/op/sort.t
@@ -7,7 +7,8 @@ BEGIN {
     set_up_inc('../lib');
 }
 use warnings;
-plan(tests => 199);
+plan(tests => 203);
+use Tie::Array; # we need to test sorting tied arrays
 
 # these shouldn't hang
 {
@@ -433,7 +434,6 @@ cmp_ok($x,'eq','123',q(optimized-away comparison block doesn't take any other ar
     @a = qw(b c a); $r1 = \$a[1]; @a = sort mysort @a; $r2 = \$a[0];
     is "$$r1-$$r2-@a", "c-c-c b a", "inplace sort with function of lexical";
 
-    use Tie::Array;
     my @t;
     tie @t, 'Tie::StdArray';
 
@@ -494,6 +494,25 @@ cmp_ok($x,'eq','123',q(optimized-away comparison block doesn't take any other ar
     is ("@a", "3 4 5", "RT #128340");
 
 }
+{
+    @Tied_Array_EXTEND_Test::ISA= 'Tie::StdArray';
+    my $extend_count;
+    sub Tied_Array_EXTEND_Test::EXTEND {
+        $extend_count= $_[1];
+        return;
+    }
+    my @t;
+    tie @t, "Tied_Array_EXTEND_Test";
+    is($extend_count, undef, "test that EXTEND has not been called prior to initialization");
+    $t[0]=3;
+    $t[1]=1;
+    $t[2]=2;
+    is($extend_count, undef, "test that EXTEND has not been called during initialization");
+    @t= sort @t;
+    is($extend_count, 3, "test that EXTEND was called with an argument of 3 by pp_sort()");
+    is("@t","1 2 3","test that sorting the tied array worked even though EXTEND is a no-op");
+}
+
 
 # Test optimisations of reversed sorts. As we now guarantee stability by
 # default, # optimisations which do not provide this are bogus.