- /* Freeing ops on the stack, and the op_latefree / op_latefreed /
- * op_attached flags:
- *
- * When we pop tokens off the stack during error recovery, or when
- * we pop all the tokens off the stack after a die during a shift or
- * reduce (i.e. Perl_croak somewhere in yylex() or in one of the
- * newFOO() functions), then it's possible that some of these tokens are
- * of type opval, pointing to an OP. All these ops are orphans; each is
- * its own miniature subtree that has not yet been attached to a
- * larger tree. In this case, we should clearly free the op (making
- * sure, for each op we free that we have PL_comppad pointing to the
- * right place for freeing any SVs attached to the op in threaded
- * builds.
- *
- * However, there is a particular problem if we die in newFOO() called
- * by a reducing action; e.g.
- *
- * foo : bar baz boz
- * { $$ = newFOO($1,$2,$3) }
- *
- * where
- * OP *newFOO { ....; if (...) croak; .... }
- *
- * In this case, when we come to clean bar baz and boz off the stack,
- * we don't know whether newFOO() has already:
- * * freed them
- * * left them as is
- * * attached them to part of a larger tree
- * * attached them to PL_compcv
- * * attached them to PL_compcv then freed it (as in BEGIN {die } )
- *
- * To get round this problem, we set the flag op_latefree on every op
- * that gets pushed onto the parser stack. If op_free() sees this
- * flag, it clears the op and frees any children,, but *doesn't* free
- * the op itself; instead it sets the op_latefreed flag. This means
- * that we can safely call op_free() multiple times on each stack op.
- * So, when clearing the stack, we first, for each op that was being
- * reduced, call op_free with op_latefree=1. This ensures that all ops
- * hanging off these op are freed, but the reducing ops themselves are
- * just undefed. Then we set op_latefreed=0 on *all* ops on the stack
- * and free them. A little thought should convince you that this
- * two-part approach to the reducing ops should handle the first three
- * cases above safely.
- *
- * In the case of attaching to PL_compcv (currently just newATTRSUB
- * does this), then we set the op_attached flag on the op that has
- * been so attached, then avoid doing the final op_free during
- * cleanup, on the assumption that it will happen (or has already
- * happened) when PL_compcv is freed.
- *
- * Note this is fairly fragile mechanism. A more robust approach
- * would be to use two of these flag bits as 2-bit reference count
- * field for each op, indicating whether it is pointed to from:
- * * a parent op
- * * the parser stack
- * * a CV
- * but this would involve reworking all code (core and external) that
- * manipulate op trees.
- *
- * XXX DAPM 17/1/07 I've decided its too fragile for now, and so have
- * disabled it */
-
-#define DISABLE_STACK_FREE
-
-
-#ifdef DISABLE_STACK_FREE