diff --git a/src/lapi.c b/src/lapi.c
index d87b090..df0ca5b 100644
--- a/src/lapi.c
+++ b/src/lapi.c
@@ -599,13 +599,11 @@ LUA_API const char *lua_pushsubstring (lua_State *L, int idx, size_t start, size
     }
     if (ss != NULL) break;
     if (nextsscluster(cluster) == NULL) {  /* need new cluster? */
-      olddebt = g->GCdebt;
       next = luaM_newvector(L, SUBSTR_CLUSTER_SIZE, TString);
-      g->GCdebt = olddebt;
       memset(next, 0, SUBSTR_CLUSTER_SIZE * sizeof(TString));
       nextsscluster(cluster) = next;  /* chain next cluster in list */
       nextsscluster(next) = NULL;  /* ensure next pointer is NULL */
-      clusterid(next) = clusterid(cluster) + 1; /* set cluster number */
+      rawclusterid(next) = (clusterid(cluster) + 1) | (g->currentwhite & bitmask(WHITE0BIT) ? 0 : ~CLUSTERID_MASK); /* set cluster number */
       ((bitmap_unit*)next)[BITMAP_SKIP] = 0xFFFF;  /* always mark first entry as used by bitmap */
       nextsscluster(cluster) = next;
     }
diff --git a/src/lgc.c b/src/lgc.c
index 1e81493..4b0592d 100644
--- a/src/lgc.c
+++ b/src/lgc.c
@@ -525,26 +525,27 @@ static lu_mem traversestack (global_State *g, lua_State *th) {
 
 
 static int traverserope (global_State *g, TString *r) {
-  int size = sizeof(TString);
-  if (r->tsr.left) {
+  if (r->tsr.left)
     markobject(g, r->tsr.left);
-    //size += sizeof(TString);
-  }
-  if (r->tsr.right) {
+  if (r->tsr.right)
     markobject(g, r->tsr.right);
-    //size += sizeof(TString);
-  }
-  if (r->tsr.res) {
+  if (r->tsr.res)
     markobject(g, r->tsr.res);
-    //size += sizeof(TString);
+  if ((rawclusterid(r->tsr.cluster) & ~CLUSTERID_MASK ? 1 : 0) != (g->currentwhite & bitmask(WHITE0BIT) ? 1 : 0)) {
+    rawclusterid(r->tsr.cluster) = clusterid(r->tsr.cluster) | (g->currentwhite & bitmask(WHITE0BIT) ? ~CLUSTERID_MASK : 0);
+    return sizeof(TString);
   }
-  return size;
+  return 0;
 }
 
 
 static int traversesubstr (global_State *g, TString *ss) {
   markobject(g, ss->tss.str);
-  return sizeof(TString);
+  if ((rawclusterid(ss->tss.cluster) & ~CLUSTERID_MASK ? 1 : 0) != (g->currentwhite & bitmask(WHITE0BIT) ? 1 : 0)) {
+    rawclusterid(ss->tss.cluster) = clusterid(ss->tss.cluster) | (g->currentwhite & bitmask(WHITE0BIT) ? ~CLUSTERID_MASK : 0);
+    return sizeof(TString);
+  }
+  return 0;
 }
 
 
@@ -1089,8 +1090,12 @@ static lu_mem singlestep (lua_State *L) {
   global_State *g = G(L);
   switch (g->gcstate) {
     case GCSpause: {
+      TString* cluster;
       /* start to count memory traversed */
-      g->GCmemtrav = g->strt.size * sizeof(GCObject*);
+      g->GCmemtrav = g->strt.size * sizeof(GCObject*) +
+        g->ropestacksize * sizeof(TString*) +
+        ROPE_CLUSTER_SIZE * sizeof(TString) +
+        SUBSTR_CLUSTER_SIZE * sizeof(TString);  /* "static" blocks */
       lua_assert(!isgenerational(g));
       restartcollection(g);
       g->gcstate = GCSpropagate;
diff --git a/src/lstate.c b/src/lstate.c
index ccf9edb..1258faf 100644
--- a/src/lstate.c
+++ b/src/lstate.c
@@ -200,7 +200,6 @@ static void f_luaopen (lua_State *L, void *ud) {
   g->memerrmsg = luaS_newliteral(L, MEMERRMSG);
   luaS_fix(g->memerrmsg);  /* it should never be collected */
   /* allocate rope and substring clusters */
-  olddebt = g->GCdebt;  /* don't count debt of these, they are never found by GC */
   g->ropestack = luaM_newvector(L, g->ropestacksize, TString *);
   g->ropeclusters = luaM_newvector(L, ROPE_CLUSTER_SIZE, TString);
   memset(g->ropeclusters, 0, ROPE_CLUSTER_SIZE * sizeof(TString));
@@ -213,9 +212,6 @@ static void f_luaopen (lua_State *L, void *ud) {
   ((unsigned long*)g->ssclusters)[BITMAP_SKIP] = 0xFFFF;  /* always mark first entry as used by bitmap */
   g->ssfreecluster = g->ssclusters;
   memset(g->allowedcfuncs, 0, sizeof(g->allowedcfuncs));  /* set all funclists to NULL */
-  /* mark new allocations as compensated & reset debt */
-  //g->totalbytes += g->GCdebt - olddebt;
-  g->GCdebt = olddebt;
   g->gcrunning = 1;  /* allow gc */
   g->version = lua_version(NULL);
   luai_userstateopen(L);
@@ -257,19 +253,17 @@ static void close_state (lua_State *L) {
   luaM_freearray(L, G(L)->strt.hash, G(L)->strt.size);
   luaZ_freebuffer(L, &g->buff);
   freestack(L);
-  olddebt = g->GCdebt;
   luaM_freearray(L, g->ropestack, g->ropestacksize);
   while (cluster != NULL) {
     next = *(TString**)cluster;
-    luaM_free(L, cluster);
+    luaM_freemem(L, cluster, ROPE_CLUSTER_SIZE * sizeof(TString));
     cluster = next;
   }
   while (sscluster != NULL) {
     ssnext = *(TString**)sscluster;
-    luaM_free(L, sscluster);
+    luaM_freemem(L, sscluster, SUBSTR_CLUSTER_SIZE * sizeof(TString));
     sscluster = ssnext;
   }
-  g->GCdebt = olddebt;
   //lua_assert(gettotalbytes(g) == sizeof(LG));
   if (g->lockstate) lua_unlock(L);
   _lua_freelock(g->lock);
diff --git a/src/lstring.c b/src/lstring.c
index cf4df22..d124e79 100644
--- a/src/lstring.c
+++ b/src/lstring.c
@@ -209,13 +209,11 @@ TString *luaS_concat (lua_State *L, TString *l, TString *r) {
     }
     if (rope != NULL) break;
     if (nextropecluster(cluster) == NULL) {  /* need new cluster? */
-      olddebt = g->GCdebt;
       next = luaM_newvector(L, ROPE_CLUSTER_SIZE, TString);
-      g->GCdebt = olddebt;
       memset(next, 0, ROPE_CLUSTER_SIZE * sizeof(TString));
       nextropecluster(cluster) = next;  /* chain next cluster in list */
       nextropecluster(next) = NULL;  /* ensure next pointer is NULL */
-      clusterid(next) = clusterid(cluster) + 1; /* set cluster number */
+      rawclusterid(next) = (clusterid(cluster) + 1) | (g->currentwhite & bitmask(WHITE0BIT) ? 0 : ~CLUSTERID_MASK); /* set cluster number */
       ((bitmap_unit*)next)[BITMAP_SKIP] = 0xFFFF;  /* always mark first entry as used by bitmap */
       nextropecluster(cluster) = next;
     }
@@ -239,20 +237,21 @@ TString *luaS_build (lua_State *L, TString *rope) {
   TString *s;
   TString **stack;
   TString *orig = rope;
-  l_mem olddebt;
+  int freemem = 0;
+  void* alloc_ud;
   if (rope->tsr.tt == LUA_TSHRSTR || rope->tsr.tt == LUA_TLNGSTR || rope->tsr.tt == LUA_TSUBSTR) return cast(TString *, rope);
   if (rope->tsr.res || rope->tsr.left == NULL || rope->tsr.right == NULL) return rope->tsr.res;
-  if (rope->tsr.len >= ROPE_ALLOC_MIN_SIZE) buffer = cur = luaM_newvector(L, rope->tsr.len, char);
-  else buffer = cur = luaZ_openspace(L, &G(L)->buff, rope->tsr.len);
+  if (rope->tsr.len >= ROPE_ALLOC_MIN_SIZE) {
+    buffer = cur = G(L)->frealloc(G(L)->ud, NULL, 0, rope->tsr.len * sizeof(char));
+    freemem = 1;
+  } else buffer = cur = luaZ_openspace(L, &G(L)->buff, rope->tsr.len);
   stack = G(L)->ropestack;
   do {
     int b = 0;
     while (rope->tsr.left->tsr.tt == LUA_TROPSTR && rope->tsr.left->tsr.res == NULL) {
       if (stack - G(L)->ropestack == G(L)->ropestacksize - 1) {
         TString **oldbase = G(L)->ropestack;
-        olddebt = G(L)->GCdebt;
         luaM_reallocvector(L, G(L)->ropestack, G(L)->ropestacksize, G(L)->ropestacksize + G(L)->ropestacksize, TString *);
-        G(L)->GCdebt = olddebt;
         stack = G(L)->ropestack + (stack - oldbase);
         G(L)->ropestacksize += G(L)->ropestacksize;
       }
@@ -287,7 +286,9 @@ TString *luaS_build (lua_State *L, TString *rope) {
     rope = rope->tsr.right;
   } while (stack >= G(L)->ropestack);
   s = luaS_newlstr(L, buffer, cur - buffer);
-  if (orig->tsr.len >= ROPE_ALLOC_MIN_SIZE) luaM_free(L, buffer);
+  if (freemem) {
+    G(L)->frealloc(G(L)->ud, buffer, orig->tsr.len * sizeof(char), 0);
+  }
   orig->tsr.res = s;
   orig->tsr.left = orig->tsr.right = NULL;  /* release left & right nodes (we don't need them anymore) */
   /* mark the string as black so it doesn't accidentally get freed */
@@ -318,7 +319,6 @@ static void freeropeclusters (lua_State *L) {
   TString *cluster, *last = NULL;
   bitmap_unit *bitmap;
   int i, empty, full, kept = 1, setfree = 0;
-  l_mem olddebt;
   for (cluster = G(L)->ropeclusters; cluster != NULL; last = cluster, cluster = nextropecluster(cluster)) {
     bitmap = (bitmap_unit*)cluster + BITMAP_SKIP;
     empty = (bitmap[0] & ~(bitmap_unit)(0xFFFF)) == 0;  /* ignore first entry use bit */
@@ -334,14 +334,14 @@ static void freeropeclusters (lua_State *L) {
     if (empty) {  /* entire cluster unused? */
       if (kept) {
         kept--;  /* leave one empty cluster allocated */
+        setfree = 1;
+        G(L)->ropefreecluster = cluster;
       } else {
         /* unlink and free cluster */
         nextropecluster(last) = nextropecluster(cluster);
         if (G(L)->ropefreecluster == cluster)
           G(L)->ropefreecluster = nextropecluster(cluster);
-        olddebt = G(L)->GCdebt;
-        luaM_free(L, cluster);
-        G(L)->GCdebt = olddebt;
+        luaM_freemem(L, cluster, ROPE_CLUSTER_SIZE * sizeof(TString));
         cluster = last;
       }
     } else if (full && !setfree) {
@@ -355,7 +355,6 @@ static void freessclusters (lua_State *L) {
   TString *cluster, *last = NULL;
   bitmap_unit *bitmap;
   int i, empty, full, kept = 1, setfree = 0;
-  l_mem olddebt;
   for (cluster = G(L)->ssclusters; cluster != NULL; last = cluster, cluster = nextsscluster(cluster)) {
     bitmap = (bitmap_unit*)cluster + BITMAP_SKIP;
     empty = (bitmap[0] & ~(bitmap_unit)(0xFFFF)) == 0;  /* ignore first entry use bit */
@@ -371,14 +370,14 @@ static void freessclusters (lua_State *L) {
     if (empty) {  /* entire cluster unused? */
       if (kept) {
         kept--;  /* leave one empty cluster allocated */
+        setfree = 1;
+        G(L)->ssfreecluster = cluster;
       } else {
         /* unlink and free cluster */
         nextsscluster(last) = nextsscluster(cluster);
         if (G(L)->ssfreecluster == cluster)
           G(L)->ssfreecluster = nextsscluster(cluster);
-        olddebt = G(L)->GCdebt;
-        luaM_free(L, cluster);
-        G(L)->GCdebt = olddebt;
+        luaM_freemem(L, cluster, SUBSTR_CLUSTER_SIZE * sizeof(TString));
         cluster = last;
       }
     } else if (full && !setfree) {
diff --git a/src/lstring.h b/src/lstring.h
index 9c2deea..37fa11c 100644
--- a/src/lstring.h
+++ b/src/lstring.h
@@ -16,12 +16,14 @@
 #define SUBSTR_CLUSTER_SIZE ((sizeof(TString) * 16 - sizeof(void*) * 2) * 8)
 #define BITMAP_UNIT_SIZE (sizeof(bitmap_unit) * 8)
 #define BITMAP_SKIP (sizeof(void*) / sizeof(bitmap_unit) * 2)
+#define CLUSTERID_MASK LONG_MAX
 
 #define sizestring(s)	(sizeof(union TString)+((s)->len+1)*sizeof(char))
 
 #define sizeudata(u)	(sizeof(union Udata)+(u)->len)
 
-#define clusterid(l) (*(bitmap_unit*)((void**)(l) + 1))
+#define clusterid(l) (*(bitmap_unit*)((void**)(l) + 1) & CLUSTERID_MASK)
+#define rawclusterid(l) (*(bitmap_unit*)((void**)(l) + 1))
 #define nextropecluster(l) (*(TString**)(l))
 #define nextsscluster(l) (*(TString**)(l))