compact IEEE 754 double encoding in external binary format for ei

Implement the compact IEEE 754 double encoding in external binary
format for ei. Encoding for ei now always produces the NEW_FLOAT_EXT
format. Decoding and term printing handle both the old ERL_FLOAT_EXT
encoding and the new NEW_FLOAT_EXT encoding. Legacy erl_interface code
also handles the new encoding, but still produces the ERL_FLOAT_EXT
encoding by default.

Also enable the DFLAG_NEW_FLOATS distribution flag.

Reduce the number of copies of the code for encoding and decoding
doubles throughout ei and erl_interface by instead calling the ei
encoding and decoding functions wherever possible.

Restore commented-out float tests in ei_decode_SUITE and
ei_encode_SUITE in lib/erl_interface/test. Modify them to make them
match the style of other tests in the same suites.

These changes are based on an ei float patch from Serge Aleynikov
originally submitted against R12B-2 in July 2008.

2 files changed, 22 insertions, 8 deletions

diff --git a/lib/erl_interface/src/legacy/decode_term.c b/lib/erl_interface/src/legacy/decode_term.c
index ef29d6f57d..ddd1682f95 100644
--- a/lib/erl_interface/src/legacy/decode_term.c
+++ b/lib/erl_interface/src/legacy/decode_term.c
@@ -59,6 +59,7 @@ int ei_decode_term(const char *buf, int *index, void *t)
       return ei_decode_long(buf,index,NULL);
 
     case ERL_FLOAT_EXT:  
+    case NEW_FLOAT_EXT:
       return ei_decode_double(buf,index,NULL);
 
     case ERL_ATOM_EXT:
diff --git a/lib/erl_interface/src/legacy/erl_marshal.c b/lib/erl_interface/src/legacy/erl_marshal.c
index 4b5f28178f..ad0cc5672c 100644
--- a/lib/erl_interface/src/legacy/erl_marshal.c
+++ b/lib/erl_interface/src/legacy/erl_marshal.c
@@ -102,6 +102,7 @@ void erl_init_marshal(void)
     cmp_array[ERL_SMALL_INTEGER_EXT] = 1;
     cmp_array[ERL_INTEGER_EXT]       = 1;
     cmp_array[ERL_FLOAT_EXT]         = 1;
+    cmp_array[NEW_FLOAT_EXT]         = 1;
     cmp_array[ERL_SMALL_BIG_EXT]     = 1;
     cmp_array[ERL_LARGE_BIG_EXT]     = 1;
     cmp_array[ERL_ATOM_EXT]          = 2;
@@ -124,6 +125,7 @@ void erl_init_marshal(void)
     cmp_num_class[ERL_SMALL_INTEGER_EXT] = SMALL;
     cmp_num_class[ERL_INTEGER_EXT]       = SMALL;
     cmp_num_class[ERL_FLOAT_EXT]         = FLOAT;
+    cmp_num_class[NEW_FLOAT_EXT]         = FLOAT;
     cmp_num_class[ERL_SMALL_BIG_EXT]     = BIG;
     cmp_num_class[ERL_LARGE_BIG_EXT]     = BIG;
     init_cmp_num_class_p = 0;
@@ -1008,10 +1010,13 @@ static ETERM *erl_decode_it(unsigned char **ext)
 	return ep;
 
     case ERL_FLOAT_EXT:
+    case NEW_FLOAT_EXT:
 	ERL_TYPE(ep) = ERL_FLOAT;
-	if (sscanf((char *) *ext, "%lf", &ff) != 1)
+	cp = (char *) *ext;
+	i = -1;
+	if (ei_decode_double(cp, &i, &ff) == -1)
 	    goto failure;
-	*ext += 31;
+	*ext += i;
 	ep->uval.fval.f = ff;
 	return ep;
 
@@ -1176,6 +1181,7 @@ unsigned char erl_ext_type(unsigned char *ext)
     case ERL_LARGE_TUPLE_EXT:
 	return ERL_TUPLE;
     case ERL_FLOAT_EXT:
+    case NEW_FLOAT_EXT:
 	return ERL_FLOAT;
     case ERL_BINARY_EXT:
 	return ERL_BINARY;
@@ -1218,6 +1224,7 @@ int erl_ext_size(unsigned char *t)
     case ERL_BINARY_EXT:
     case ERL_STRING_EXT:
     case ERL_FLOAT_EXT:
+    case NEW_FLOAT_EXT:
     case ERL_SMALL_BIG_EXT:
     case ERL_LARGE_BIG_EXT:
 	return 0;
@@ -1332,6 +1339,9 @@ static int jump(unsigned char **ext)
     case ERL_FLOAT_EXT:
 	*ext += 31;
 	break;
+    case NEW_FLOAT_EXT:
+	*ext += 8;
+	break;
     case ERL_BINARY_EXT:
 	i = (**ext << 24) | ((*ext)[1] << 16) |((*ext)[2] << 8) | (*ext)[3];
 	*ext += 4+i;
@@ -1696,12 +1706,15 @@ static int cmp_exe2(unsigned char **e1, unsigned char **e2)
       }
       return 0;
     case ERL_FLOAT_EXT:
-      if (sscanf((char *) *e1, "%lf", &ff1) != 1)
-	return -1;
-      *e1 += 31;
-      if (sscanf((char *) *e2, "%lf", &ff2) != 1)
-	return -1;
-      *e2 += 31;
+    case NEW_FLOAT_EXT:
+      i = -1;
+      if (ei_decode_double((char *) *e1, &i, &ff1) != 0)
+        return -1;
+      *e1 += i;
+      j = -1;
+      if (ei_decode_double((char *) *e2, &j, &ff2) != 0)
+        return -1;
+      *e2 += j;
       return cmp_floats(ff1,ff2);
 
     case ERL_BINARY_EXT: