Don’t attempt to inline my sub (){$outer_var}

$ perl5.18.2 -Ilib  -Mfeature=lexical_subs -e ' my $x; my sub a(){$x}; print a'
The lexical_subs feature is experimental at -e line 1.
Segmentation fault: 11

Same in blead.

When calls to the sub are compiled (the ‘a’ in ‘print a’) the value
of the lexical variable cannot possibly known, because the sub hasn’t
been cloned yet, and all we have is the closure prototype.

A potentially constant closure prototype is marked CvCONST and
cv_const_sv_or_av (called by the code in toke.c that handles bare-
words) thinks that CvCONST means we have a constant XSUB.  Only lexi-
cal subs allow a closure prototype to reach that function.

We shouldn’t mark the closure prototype as CvCONST to begin with.
Because when we do, the ‘constant’ is retrieved from CvXUBANY, which
is a union shared by CvSTART.  Then toke.c does SvREFCNT_inc on
CvSTART, and screws up the op tree.

1 files changed, 6 insertions, 1 deletions

diff --git a/t/op/lexsub.t b/t/op/lexsub.t
index 385aaeedd1..e170555517 100644
--- a/t/op/lexsub.t
+++ b/t/op/lexsub.t
@@ -7,7 +7,7 @@ BEGIN {
     *bar::is = *is;
     *bar::like = *like;
 }
-plan 143;
+plan 144;
 
 # -------------------- Errors with feature disabled -------------------- #
 
@@ -788,6 +788,11 @@ is runperl(switches => ['-lXMfeature=:all'],
   print handel,"ok ",curr_test()," - no 'No comma allowed' after my sub\n";
   curr_test(curr_test()+1);
 }
+{
+  my $x = 43;
+  my sub y :prototype() {$x};
+  is y, 43, 'my sub that looks like constant closure';
+}
 
 # -------------------- Interactions (and misc tests) -------------------- #