improve ubsan support

Provide `--enable-ubsan` to simplify `-fsanitize=undefined` builds,
where alignment and floating-point divide-by-zero need to be
disabled to get useful results.

Also, repair undefined behavior exposed by the test suite. Most of the
repairs are avoiding `memset(..., NULL, 0)` --- which is an unhelpful
requirement, IMO. The other two repairs are worthwhile but unlikely to
have caused trouble.

racket/src/configure

@@ -827,6 +827,7 @@ enable_ffipoll
 enable_gprof
 enable_gcov
 enable_noopt
+enable_ubsan
 '
       ac_precious_vars='build_alias
 host_alias
@@ -1481,6 +1482,7 @@ Optional Features:
   --enable-gcov           compile to gather gcov statistics (gcc3 only)
   --enable-strip          strip debug on install (usually enabled by default)
   --enable-noopt          drop -O C flags (useful for low-level debugging)
+  --enable-ubsan          compile with -fsanitize=undefined)
 
 Some influential environment variables:
   CC          C compiler command
@@ -2824,6 +2826,11 @@ if test "${enable_noopt+set}" = set; then :
   enableval=$enable_noopt;
 fi
 
+# Check whether --enable-ubsan was given.
+if test "${enable_ubsan+set}" = set; then :
+  enableval=$enable_ubsan;
+fi
+
 
 ###### Some flags imply other flags #######
 
@@ -6791,6 +6798,16 @@ if test "${enable_noopt}" = "yes" ; then
   PREFLAGS=`echo "$PREFLAGS" | awk "$AWKPRG"`
 fi
 
+############## ubsan ################
+
+if test "${enable_ubsan}" = "yes" ; then
+   UBSAN="-fsanitize=undefined -fno-sanitize=alignment -fno-sanitize=float-divide-by-zero"
+   CFLAGS="$CFLAGS $UBSAN"
+   CPPFLAGS="$CPPFLAGS $UBSAN"
+   PREFLAGS="$PREFLAGS $UBSAN"
+   LDFLAGS="$LDFLAGS $UBSAN"
+fi
+
 ############## usersetup ################
 
 if test "${enable_usersetup}" != "yes" ; then

racket/src/racket/configure.ac

@@ -86,6 +86,7 @@ AC_ARG_ENABLE(gcov,    [  --enable-gcov           compile to gather gcov statist
 
 AC_ARG_ENABLE(noopt,   [  --enable-strip          strip debug on install (usually enabled by default)])
 AC_ARG_ENABLE(noopt,   [  --enable-noopt          drop -O C flags (useful for low-level debugging)])
+AC_ARG_ENABLE(ubsan,   [  --enable-ubsan          compile with -fsanitize=undefined)])
 
 ###### Some flags imply other flags #######
 
@@ -1629,6 +1630,16 @@ if test "${enable_noopt}" = "yes" ; then
   PREFLAGS=`echo "$PREFLAGS" | awk "$AWKPRG"`
 fi
 
+############## ubsan ################
+
+if test "${enable_ubsan}" = "yes" ; then
+   UBSAN="-fsanitize=undefined -fno-sanitize=alignment -fno-sanitize=float-divide-by-zero"
+   CFLAGS="$CFLAGS $UBSAN"
+   CPPFLAGS="$CPPFLAGS $UBSAN"
+   PREFLAGS="$PREFLAGS $UBSAN"
+   LDFLAGS="$LDFLAGS $UBSAN"
+fi
+
 ############## usersetup ################
 
 if test "${enable_usersetup}" != "yes" ; then  

racket/src/racket/gc2/mem_account.c

@@ -136,7 +136,8 @@ inline static int create_blank_owner_set(NewGC *gc)
   gc->owner_table_size = curr_size;
 
   naya = (OTEntry **)ofm_malloc(curr_size * sizeof(OTEntry*));
-  memcpy(naya, owner_table, old_size*sizeof(OTEntry*));
+  if (old_size)
+    memcpy(naya, owner_table, old_size*sizeof(OTEntry*));
   gc->owner_table = owner_table = naya;
   bzero(((char*)owner_table) + (sizeof(OTEntry*) * old_size),
         (curr_size - old_size) * sizeof(OTEntry*));

racket/src/racket/gc2/roots.c

@@ -17,7 +17,8 @@ static void grow_roots(Roots *roots) {
   roots->size = roots->size ? ( 2 * roots->size ) : 500;
   new_roots   = (uintptr_t *)ofm_malloc(sizeof(uintptr_t) * (roots->size + 1));
 
-  memcpy((void *)new_roots, (void *)roots->roots, sizeof(uintptr_t) * roots->count);
+  if (roots->count)
+    memcpy((void *)new_roots, (void *)roots->roots, sizeof(uintptr_t) * roots->count);
 
   if (roots->roots)
     free(roots->roots);

racket/src/racket/gc2/vm_osx.c

@@ -362,6 +362,10 @@ void exception_thread(void *shared_thread_state)
     /* block until we get an exception message */
     retval = mach_msg(message, MACH_RCV_MSG, 0, sizeof(mach_exc_msg_t), 
 		      exc_port, MACH_MSG_TIMEOUT_NONE, MACH_PORT_NULL);
+    if(retval != KERN_SUCCESS) {
+      GCPRINT(GCOUTF, "Message receive failed: %s\n", mach_error_string(retval));
+      abort();
+    }
     /* forward off the handling of this message */
     if(!exc_server(message, reply)) {
       GCPRINT(GCOUTF, "INTERNAL ERROR: exc_server() didn't like something\n");
@@ -370,6 +374,10 @@ void exception_thread(void *shared_thread_state)
     /* send the message back out to the thread */
     retval = mach_msg(reply, MACH_SEND_MSG, sizeof(mach_reply_msg_t), 0, 
 		      MACH_PORT_NULL, MACH_MSG_TIMEOUT_NONE, MACH_PORT_NULL);
+    if(retval != KERN_SUCCESS) {
+      GCPRINT(GCOUTF, "Message send failed: %s\n", mach_error_string(retval));
+      abort();
+    }
   }
 }
 

racket/src/racket/sgc/sgc.c

@@ -1372,7 +1372,8 @@ static void *realloc_collect_temp(void *v, intptr_t oldsize, intptr_t newsize)
   void *naya;
   
   naya = platform_plain_sector((newsize + SECTOR_SEGMENT_SIZE - 1) >> LOG_SECTOR_SEGMENT_SIZE, 0);
-  memcpy(naya, v, oldsize);
+  if (oldsize)
+    memcpy(naya, v, oldsize);
   if (v)
     munmap(v, (oldsize + SECTOR_SEGMENT_SIZE - 1) >> LOG_SECTOR_SEGMENT_SIZE);
 

racket/src/racket/src/eval.c

@@ -1125,7 +1125,8 @@ void scheme_new_mark_segment(Scheme_Thread *p)
   seg = scheme_malloc_allow_interior(sizeof(Scheme_Cont_Mark) * SCHEME_MARK_SEGMENT_SIZE);
   segs[c] = seg;
 
-  memcpy(segs, p->cont_mark_stack_segments, c * sizeof(Scheme_Cont_Mark *));
+  if (c)
+    memcpy(segs, p->cont_mark_stack_segments, c * sizeof(Scheme_Cont_Mark *));
   
   p->cont_mark_seg_count++;
   p->cont_mark_stack_segments = segs;

racket/src/racket/src/fun.c

@@ -5308,7 +5308,8 @@ static Scheme_Meta_Continuation *clone_meta_cont(Scheme_Meta_Continuation *mc,
       } else {
         Scheme_Cont_Mark *cp;
         cp = MALLOC_N(Scheme_Cont_Mark, naya->cont_mark_total);
-        memcpy(cp, mc->cont_mark_stack_copied, naya->cont_mark_total * sizeof(Scheme_Cont_Mark));
+        if (naya->cont_mark_total)
+          memcpy(cp, mc->cont_mark_stack_copied, naya->cont_mark_total * sizeof(Scheme_Cont_Mark));
         clear_cm_copy_caches(cp, naya->cont_mark_total);
         naya->cont_mark_stack_copied = cp;
         naya->cm_caches = 0;

racket/src/racket/src/jit.c

@@ -1248,7 +1248,7 @@ int scheme_generate_flonum_local_unboxing(mz_jit_state *jitter, int push, int no
 /*========================================================================*/
 
 #ifdef JIT_PRECISE_GC
-static Scheme_Object example_so = { scheme_native_closure_type, 0 };
+static struct { Scheme_Object so; int filler; } example_so = { { scheme_native_closure_type, 0}, 0 };
 #endif
 
 static Scheme_Native_Lambda *create_native_lambda(Scheme_Lambda *lam, int clear_code_after_jit,

racket/src/racket/src/marshal.c

@@ -852,7 +852,8 @@ static Scheme_Object *write_lambda(Scheme_Object *obj)
           /* Need to grow the array */
           Scheme_Object **a;
           a = MALLOC_N(Scheme_Object *, (pos ? 2 * pos : 32));
-          memcpy(a, mt->cdata_map, pos * sizeof(Scheme_Object *));
+          if (pos)
+            memcpy(a, mt->cdata_map, pos * sizeof(Scheme_Object *));
           mt->cdata_map = a;
         }
         mt->cdata_counter++;

racket/src/racket/src/optimize.c

@@ -2677,7 +2677,8 @@ static void register_local_argument_types(Scheme_App_Rec *app, Scheme_App2_Rec *
               } else if (predicate_implies(lam->ir_info->arg_types[i], pred)) {
                 /* widen */
                 lam->ir_info->arg_types[i] = pred;
-                lam->ir_info->arg_type_contributors[i] |= (1 << (nth_app-1));
+                if (nth_app)
+                  lam->ir_info->arg_type_contributors[i] |= (1 << (nth_app-1));
               } else
                 widen_to_top = 1;
             } else

racket/src/racket/src/portfun.c

@@ -665,7 +665,8 @@ string_write_bytes(Scheme_Output_Port *port,
     memcpy(is->string, old, is->index);
   }
 
-  memcpy(is->string + is->index, str + d, len);
+  if (len)
+    memcpy(is->string + is->index, str + d, len);
   is->index += len;
 
   return len;

racket/src/racket/src/print.c

@@ -1235,7 +1235,8 @@ static void print_this_string(PrintParams *pp, const char *str, int offset, int
     memcpy(pp->print_buffer, oldstr, pp->print_position);
   }
 
-  memcpy(pp->print_buffer + pp->print_position, str + offset, len);
+  if (len)
+    memcpy(pp->print_buffer + pp->print_position, str + offset, len);
   pp->print_position += len;
   pp->print_offset += len;
 
@@ -4144,7 +4145,8 @@ void scheme_set_type_printer(Scheme_Type stype, Scheme_Type_Printer printer)
     Scheme_Type_Printer *naya;
     naya = MALLOC_N(Scheme_Type_Printer, stype + 10);
     memset(naya, 0, sizeof(Scheme_Type_Printer) * (stype + 10));
-    memcpy(naya, printers, sizeof(Scheme_Type_Printer) * printers_count);
+    if (printers_count)
+      memcpy(naya, printers, sizeof(Scheme_Type_Printer) * printers_count);
     printers_count = stype + 10;
     printers = naya;
   }

racket/src/racket/src/regexp.c

@@ -2448,7 +2448,8 @@ static void stack_room(Regwork *rw, int amt)
     sz = rw->rewind_stack_size * 2;
     if (!sz) sz = MATCH_STACK_SIZE;
     p = (rxpos *)scheme_malloc_atomic(sizeof(rxpos)*sz);
-    memcpy(p, rw->rewind_stack, rw->rewind_stack_size * sizeof(rxpos));
+    if (rw->rewind_stack_size)
+      memcpy(p, rw->rewind_stack, rw->rewind_stack_size * sizeof(rxpos));
     rw->rewind_stack = p;
     rw->rewind_stack_size = sz;
   }
@@ -2964,7 +2965,8 @@ static void read_more_from_lazy_string(Regwork *rw, rxpos need_total)
                               0 /* not UTF-16 */);
     tlen = blen + ls->blen;
     s = (char *)scheme_malloc_atomic(tlen);
-    memcpy(s, ls->s, ls->blen);
+    if (ls->blen)
+      memcpy(s, ls->s, ls->blen);
     scheme_utf8_encode(ls->chars, ls->start + ls->done, ls->start + ls->done + amt,
                        (unsigned char *)s, ls->blen,
                        0 /* not UTF-16 */);
@@ -3012,7 +3014,8 @@ static void read_more_from_regport(Regwork *rw, rxpos need_total)
       size = 16;
 
     naya = (char *)scheme_malloc_atomic(size);
-    memcpy(naya, rw->instr, rw->input_end);
+    if (rw->input_end)
+      memcpy(naya, rw->instr, rw->input_end);
 
     rw->instr = naya;
     rw->instr_size = size;
@@ -5916,7 +5919,8 @@ static Scheme_Object *gen_replace(const char *name, int argc, Scheme_Object *arg
 	total = len + prefix_len + (startpd - srcoffset);
 	
 	naya = (char *)scheme_malloc_atomic(total + more + 1);
-	memcpy(naya, prefix, prefix_len);
+        if (prefix_len)
+          memcpy(naya, prefix, prefix_len);
 	memcpy(naya + prefix_len, source + srcoffset, startpd - srcoffset);
 	memcpy(naya + prefix_len + (startpd - srcoffset), insert, len);
         if (more) {

racket/src/racket/src/struct.c

@@ -4697,11 +4697,13 @@ static Scheme_Object *make_name(const char *pre, const char *tn, int ltn,
   
   memcpy(name, pre, lp);
   total = lp;
-  memcpy(name + total, (ltn < 0) ? SCHEME_SYM_VAL((Scheme_Object *)tn) : tn, xltn);
+  if (xltn)
+    memcpy(name + total, (ltn < 0) ? SCHEME_SYM_VAL((Scheme_Object *)tn) : tn, xltn);
   total += xltn;
   memcpy(name + total, post1, lp1);
   total += lp1;
-  memcpy(name + total, (lfn < 0) ? SCHEME_SYM_VAL((Scheme_Object *)fn) : fn, xlfn);
+  if (xlfn)
+    memcpy(name + total, (lfn < 0) ? SCHEME_SYM_VAL((Scheme_Object *)fn) : fn, xlfn);
   total += xlfn;
   memcpy(name + total, post2, lp2);
   total += lp2;
@@ -5092,7 +5094,8 @@ static Scheme_Object *_make_struct_type(Scheme_Object *base,
       }
       
       pa = MALLOC_N(Scheme_Object *, i + num_props);
-      memcpy(pa, struct_type->props, sizeof(Scheme_Object *) * i);
+      if (i)
+        memcpy(pa, struct_type->props, sizeof(Scheme_Object *) * i);
 
       num_props = i;
 

racket/src/racket/src/thread.c

@@ -1720,7 +1720,8 @@ static Scheme_Object *make_custodian_box(int argc, Scheme_Object *argv[])
       cust_box_alloc = 2 * cust_box_alloc;
     }
     cbs = (Scheme_Custodian_Box **)scheme_malloc_atomic(cust_box_alloc * sizeof(Scheme_Custodian_Box *));
-    memcpy(cbs, cust_boxes, cust_box_count * sizeof(Scheme_Custodian_Box *));
+    if (cust_box_count)
+      memcpy(cbs, cust_boxes, cust_box_count * sizeof(Scheme_Custodian_Box *));
     cust_boxes = cbs;
   }
   cust_boxes[cust_box_count++] = cb;
@@ -6284,7 +6285,8 @@ void scheme_add_evt_worker(Evt ***evt_array,
     if (new_size < _scheme_last_type_)
       new_size = _scheme_last_type_;
     nevts = MALLOC_N(Evt*, new_size);
-    memcpy(nevts, (*evt_array), (*evt_size) * sizeof(Evt*));
+    if (*evt_size)
+      memcpy(nevts, (*evt_array), (*evt_size) * sizeof(Evt*));
     (*evt_array) = nevts;
     (*evt_size) = new_size;
   }

racket/src/racket/src/validate.c

@@ -99,7 +99,8 @@ static void clearing_stack_push(struct Validate_Clearing *vc, int pos, int val)
     int *a, sz;
     sz = (vc->stacksize ? 2 * vc->stacksize : 32);
     a = (int *)scheme_malloc_atomic(sizeof(int) * sz);
-    memcpy(a, vc->stack, vc->stacksize * sizeof(int));
+    if (vc->stacksize)
+      memcpy(a, vc->stack, vc->stacksize * sizeof(int));
     vc->stacksize = sz;
     vc->stack = a;
   }
@@ -114,7 +115,8 @@ static void noclear_stack_push(struct Validate_Clearing *vc, int pos)
     int *a, sz;
     sz = (vc->ncstacksize ? 2 * vc->ncstacksize : 32);
     a = (int *)scheme_malloc_atomic(sizeof(int) * sz);
-    memcpy(a, vc->ncstack, vc->ncstacksize * sizeof(int));
+    if (vc->ncstacksize)
+      memcpy(a, vc->ncstack, vc->ncstacksize * sizeof(int));
     vc->ncstacksize = sz;
     vc->ncstack = a;
   }