make re_intuit_string() return correct string

Fix #123469 - Bug in split function, with utf8 strings

Each regex has two SV pointers, check_substr and check_utf8,
which hold a constant string (if any) corresponding to the longest
constant string in the regexp.

When the regex is first compiled, only one pointer is set, depending on
whether the pattern is utf8 or not; but subsequent usage of the regex can
instantiate the other pointer too. So which of the two strings
re_intuit_string() should return should be based on the UTF8ness of the
pattern, not whether check_substr is set.

1 files changed, 16 insertions, 1 deletions

diff --git a/t/op/split.t b/t/op/split.t
index 9afdd6e411..5d5c19dc39 100644
--- a/t/op/split.t
+++ b/t/op/split.t
@@ -6,7 +6,7 @@ BEGIN {
     set_up_inc('../lib');
 }
 
-plan tests => 125;
+plan tests => 131;
 
 $FS = ':';
 
@@ -374,6 +374,21 @@ is($cnt, scalar(@ary));
 }
 
 {
+    # LATIN SMALL LETTER A WITH DIAERESIS, CYRILLIC SMALL LETTER I
+    for my $pattern ("\x{e4}", "\x{0437}") {
+        utf8::upgrade $pattern;
+        my @res;
+        for my $str ("a${pattern}b", "axb", "a${pattern}b") {
+            @split = split /$pattern/, $str;
+            push @res, scalar(@split);
+        }
+        is($res[0], 2);
+        is($res[1], 1);
+        is($res[2], 2, '#123469 - split with utf8 pattern after handling non-utf8 EXPR');
+    }
+}
+
+{
     is (\@a, \@{"a"}, '@a must be global for following test');
     $p="";
     $n = @a = split /,/,$p;