don't try to clean up the COP pointer during global destruction

At this point there should be no further debugging code trying to
use the COP pointer.

On threaded builds the call to CopFILEAV() could cause a segmentation
fault, since during late destruction gv_fetchfile() could return
NULL, which the threaded version of CopFILEAV() would blindly call
GvAV() on.

1 files changed, 5 insertions, 1 deletions

diff --git a/op.c b/op.c
index 8e50ee0df0..e1584cfde1 100644
--- a/op.c
+++ b/op.c
@@ -1316,7 +1316,11 @@ S_cop_free(pTHX_ COP* cop)
 {
     PERL_ARGS_ASSERT_COP_FREE;
 
-    if (cop->op_type == OP_DBSTATE) {
+    /* If called during global destruction PL_defstash might be NULL and there
+       shouldn't be any code running that will trip over the bad cop address.
+       This also avoids uselessly creating the AV after it's been destroyed.
+    */
+    if (cop->op_type == OP_DBSTATE && PL_phase != PERL_PHASE_DESTRUCT) {
         /* Remove the now invalid op from the line number information.
            This could cause a freed memory overwrite if the debugger tried to
            set a breakpoint on this line.