Merge B::SVOP::{gv,sv} into the common B::OP accessor XS code.

On this platform, this saves about 700 bytes.

diff --git a/ext/B/B.xs b/ext/B/B.xs
index bdbaa75..a0259a0 100644 (file)
--- a/ext/B/B.xs
+++ b/ext/B/B.xs
@@ -835,7 +835,8 @@ threadsv_names()
 #define PADOFFSETp     0x10000
 #define U8p            0x20000
 #define U32p           0x30000
-#define IVp            0x40000
+#define SVp            0x40000
+#define IVp            0x50000
 
 #define OP_next_ix             OPp | offsetof(struct op, op_next)
 #define OP_sibling_ix          OPp | offsetof(struct op, op_sibling)
@@ -862,6 +863,10 @@ threadsv_names()
 #define PMOP_pmoffset_ix       IVp | offsetof(struct pmop, op_pmoffset)
 #endif
 
+#  Yes, B::SV::sv and B::SV::gv really do end up generating identical code.
+#define SVOP_sv_ix             SVp | offsetof(struct svop, op_sv)
+#define SVOP_gv_ix             SVp | offsetof(struct svop, op_sv)
+
 MODULE = B     PACKAGE = B::OP         PREFIX = OP_
 
 size_t
@@ -892,6 +897,8 @@ next(o)
        B::LOOP::nextop = LOOP_nextop_ix
        B::LOOP::lastop = LOOP_lastop_ix
        B::PMOP::pmflags = PMOP_pmflags_ix
+       B::SVOP::sv = SVOP_sv_ix
+       B::SVOP::gv = SVOP_gv_ix
     PREINIT:
        char *ptr;
        SV *ret;
@@ -914,6 +921,9 @@ next(o)
        case (U8)(U32p >> 16):
            ret = sv_2mortal(newSVuv(*((U32*)ptr)));
            break;
+       case (U8)(SVp >> 16):
+           ret = make_sv_object(aTHX_ NULL, *((SV **)ptr));
+           break;
 #ifdef USE_ITHREADS
        case (U8)(IVp >> 16):
            ret = sv_2mortal(newSViv(*((IV*)ptr)));
@@ -1125,20 +1135,6 @@ BOOT:
 #endif
 }
 
-
-#define SVOP_sv(o)     cSVOPo->op_sv
-#define SVOP_gv(o)     ((GV*)cSVOPo->op_sv)
-
-MODULE = B     PACKAGE = B::SVOP               PREFIX = SVOP_
-
-B::SV
-SVOP_sv(o)
-       B::SVOP o
-
-B::GV
-SVOP_gv(o)
-       B::SVOP o
-
 #define PADOP_padix(o) o->op_padix
 #define PADOP_sv(o)    (o->op_padix ? PAD_SVl(o->op_padix) : Nullsv)
 #define PADOP_gv(o)    ((o->op_padix \