New lightweight Carp has a require. If Carp is used in a __DIE__ handler

this causes a POPSTACK panic. The problem seems to be that although
die_where() has unwound the tail of perl_vdie() top_env setjmp has been
set to resume execution there. Avoiding setting CATCH_SET(TRUE) in 
call_sv() avoids this. So invent a new G_NOCATCH flag to disable
messing with CATCH_SET() in call_sv, use it in perl_vdie().
Add test to op/eval.t which will fail (panic) if bug comes back.
>>> I AM NOT CONVINCED THIS IS CORRECT LONG TERM FIX <<<

p4raw-id: //depot/perl@3988

1 files changed, 8 insertions, 2 deletions

diff --git a/perl.c b/perl.c
index d81187989f..8c8b84c387 100644
--- a/perl.c
+++ b/perl.c
@@ -1225,10 +1225,16 @@ Perl_call_sv(pTHX_ SV *sv, I32 flags)
 	PL_op->op_private |= OPpENTERSUB_DB;
 
     if (!(flags & G_EVAL)) {
-	CATCH_SET(TRUE);
+        /* G_NOCATCH is a hack for perl_vdie using this path to call
+	   a __DIE__ handler */
+        if (!(flags & G_NOCATCH)) {
+	    CATCH_SET(TRUE);
+	}
 	call_xbody((OP*)&myop, FALSE);
 	retval = PL_stack_sp - (PL_stack_base + oldmark);
-	CATCH_SET(FALSE);
+        if (!(flags & G_NOCATCH)) {
+	    CATCH_SET(FALSE);
+	}
     }
     else {
 	cLOGOP->op_other = PL_op;