Binary serialization for reals

Added new "encode_real" methods for handling real_t, and used them for vector types. Types are encoded based on compilation setting.

But for decoding, always check how it was encoded. This way, serialized data is cross-compatible with Godot compiled with singles and Godot compiled with doubles. At least, in theory.

2 files changed, 368 insertions, 169 deletions

diff --git a/core/io/marshalls.cpp b/core/io/marshalls.cpp
index 4c58c84c14..c447e11ee7 100644
--- a/core/io/marshalls.cpp
+++ b/core/io/marshalls.cpp
@@ -111,6 +111,9 @@ Error decode_variant(Variant &r_variant, const uint8_t *p_buffer, int p_len, int
 		*r_len = 4;
 	}
 
+	// Note: We cannot use sizeof(real_t) for decoding, in case a different size is encoded.
+	// Decoding math types always checks for the encoded size, while encoding always uses compilation setting.
+	// This does lead to some code duplication for decoding, but compatibility is the priority.
 	switch (type & ENCODE_MASK) {
 		case Variant::NIL: {
 			r_variant = Variant();
@@ -144,18 +147,18 @@ Error decode_variant(Variant &r_variant, const uint8_t *p_buffer, int p_len, int
 		} break;
 		case Variant::FLOAT: {
 			if (type & ENCODE_FLAG_64) {
-				ERR_FAIL_COND_V(len < 8, ERR_INVALID_DATA);
+				ERR_FAIL_COND_V((size_t)len < sizeof(double), ERR_INVALID_DATA);
 				double val = decode_double(buf);
 				r_variant = val;
 				if (r_len) {
-					(*r_len) += 8;
+					(*r_len) += sizeof(double);
 				}
 			} else {
-				ERR_FAIL_COND_V(len < 4, ERR_INVALID_DATA);
+				ERR_FAIL_COND_V((size_t)len < sizeof(float), ERR_INVALID_DATA);
 				float val = decode_float(buf);
 				r_variant = val;
 				if (r_len) {
-					(*r_len) += 4;
+					(*r_len) += sizeof(float);
 				}
 			}
 
@@ -172,15 +175,25 @@ Error decode_variant(Variant &r_variant, const uint8_t *p_buffer, int p_len, int
 
 		// math types
 		case Variant::VECTOR2: {
-			ERR_FAIL_COND_V(len < 4 * 2, ERR_INVALID_DATA);
 			Vector2 val;
-			val.x = decode_float(&buf[0]);
-			val.y = decode_float(&buf[4]);
-			r_variant = val;
+			if (type & ENCODE_FLAG_64) {
+				ERR_FAIL_COND_V((size_t)len < sizeof(double) * 2, ERR_INVALID_DATA);
+				val.x = decode_double(&buf[0]);
+				val.y = decode_double(&buf[sizeof(double)]);
 
-			if (r_len) {
-				(*r_len) += 4 * 2;
+				if (r_len) {
+					(*r_len) += sizeof(double) * 2;
+				}
+			} else {
+				ERR_FAIL_COND_V((size_t)len < sizeof(float) * 2, ERR_INVALID_DATA);
+				val.x = decode_float(&buf[0]);
+				val.y = decode_float(&buf[sizeof(float)]);
+
+				if (r_len) {
+					(*r_len) += sizeof(float) * 2;
+				}
 			}
+			r_variant = val;
 
 		} break;
 		case Variant::VECTOR2I: {
@@ -196,17 +209,29 @@ Error decode_variant(Variant &r_variant, const uint8_t *p_buffer, int p_len, int
 
 		} break;
 		case Variant::RECT2: {
-			ERR_FAIL_COND_V(len < 4 * 4, ERR_INVALID_DATA);
 			Rect2 val;
-			val.position.x = decode_float(&buf[0]);
-			val.position.y = decode_float(&buf[4]);
-			val.size.x = decode_float(&buf[8]);
-			val.size.y = decode_float(&buf[12]);
-			r_variant = val;
+			if (type & ENCODE_FLAG_64) {
+				ERR_FAIL_COND_V((size_t)len < sizeof(double) * 4, ERR_INVALID_DATA);
+				val.position.x = decode_double(&buf[0]);
+				val.position.y = decode_double(&buf[sizeof(double)]);
+				val.size.x = decode_double(&buf[sizeof(double) * 2]);
+				val.size.y = decode_double(&buf[sizeof(double) * 3]);
 
-			if (r_len) {
-				(*r_len) += 4 * 4;
+				if (r_len) {
+					(*r_len) += sizeof(double) * 4;
+				}
+			} else {
+				ERR_FAIL_COND_V((size_t)len < sizeof(float) * 4, ERR_INVALID_DATA);
+				val.position.x = decode_float(&buf[0]);
+				val.position.y = decode_float(&buf[sizeof(float)]);
+				val.size.x = decode_float(&buf[sizeof(float) * 2]);
+				val.size.y = decode_float(&buf[sizeof(float) * 3]);
+
+				if (r_len) {
+					(*r_len) += sizeof(float) * 4;
+				}
 			}
+			r_variant = val;
 
 		} break;
 		case Variant::RECT2I: {
@@ -224,16 +249,27 @@ Error decode_variant(Variant &r_variant, const uint8_t *p_buffer, int p_len, int
 
 		} break;
 		case Variant::VECTOR3: {
-			ERR_FAIL_COND_V(len < 4 * 3, ERR_INVALID_DATA);
 			Vector3 val;
-			val.x = decode_float(&buf[0]);
-			val.y = decode_float(&buf[4]);
-			val.z = decode_float(&buf[8]);
-			r_variant = val;
+			if (type & ENCODE_FLAG_64) {
+				ERR_FAIL_COND_V((size_t)len < sizeof(double) * 3, ERR_INVALID_DATA);
+				val.x = decode_double(&buf[0]);
+				val.y = decode_double(&buf[sizeof(double)]);
+				val.z = decode_double(&buf[sizeof(double) * 2]);
 
-			if (r_len) {
-				(*r_len) += 4 * 3;
+				if (r_len) {
+					(*r_len) += sizeof(double) * 3;
+				}
+			} else {
+				ERR_FAIL_COND_V((size_t)len < sizeof(float) * 3, ERR_INVALID_DATA);
+				val.x = decode_float(&buf[0]);
+				val.y = decode_float(&buf[sizeof(float)]);
+				val.z = decode_float(&buf[sizeof(float) * 2]);
+
+				if (r_len) {
+					(*r_len) += sizeof(float) * 3;
+				}
 			}
+			r_variant = val;
 
 		} break;
 		case Variant::VECTOR3I: {
@@ -250,101 +286,177 @@ Error decode_variant(Variant &r_variant, const uint8_t *p_buffer, int p_len, int
 
 		} break;
 		case Variant::TRANSFORM2D: {
-			ERR_FAIL_COND_V(len < 4 * 6, ERR_INVALID_DATA);
 			Transform2D val;
-			for (int i = 0; i < 3; i++) {
-				for (int j = 0; j < 2; j++) {
-					val.elements[i][j] = decode_float(&buf[(i * 2 + j) * 4]);
+			if (type & ENCODE_FLAG_64) {
+				ERR_FAIL_COND_V((size_t)len < sizeof(double) * 6, ERR_INVALID_DATA);
+				for (int i = 0; i < 3; i++) {
+					for (int j = 0; j < 2; j++) {
+						val.elements[i][j] = decode_double(&buf[(i * 2 + j) * sizeof(double)]);
+					}
 				}
-			}
 
-			r_variant = val;
+				if (r_len) {
+					(*r_len) += sizeof(double) * 6;
+				}
+			} else {
+				ERR_FAIL_COND_V((size_t)len < sizeof(float) * 6, ERR_INVALID_DATA);
+				for (int i = 0; i < 3; i++) {
+					for (int j = 0; j < 2; j++) {
+						val.elements[i][j] = decode_float(&buf[(i * 2 + j) * sizeof(float)]);
+					}
+				}
 
-			if (r_len) {
-				(*r_len) += 4 * 6;
+				if (r_len) {
+					(*r_len) += sizeof(float) * 6;
+				}
 			}
+			r_variant = val;
 
 		} break;
 		case Variant::PLANE: {
-			ERR_FAIL_COND_V(len < 4 * 4, ERR_INVALID_DATA);
 			Plane val;
-			val.normal.x = decode_float(&buf[0]);
-			val.normal.y = decode_float(&buf[4]);
-			val.normal.z = decode_float(&buf[8]);
-			val.d = decode_float(&buf[12]);
-			r_variant = val;
+			if (type & ENCODE_FLAG_64) {
+				ERR_FAIL_COND_V((size_t)len < sizeof(double) * 4, ERR_INVALID_DATA);
+				val.normal.x = decode_double(&buf[0]);
+				val.normal.y = decode_double(&buf[sizeof(double)]);
+				val.normal.z = decode_double(&buf[sizeof(double) * 2]);
+				val.d = decode_double(&buf[sizeof(double) * 3]);
 
-			if (r_len) {
-				(*r_len) += 4 * 4;
+				if (r_len) {
+					(*r_len) += sizeof(double) * 4;
+				}
+			} else {
+				ERR_FAIL_COND_V((size_t)len < sizeof(float) * 4, ERR_INVALID_DATA);
+				val.normal.x = decode_float(&buf[0]);
+				val.normal.y = decode_float(&buf[sizeof(float)]);
+				val.normal.z = decode_float(&buf[sizeof(float) * 2]);
+				val.d = decode_float(&buf[sizeof(float) * 3]);
+
+				if (r_len) {
+					(*r_len) += sizeof(float) * 4;
+				}
 			}
+			r_variant = val;
 
 		} break;
 		case Variant::QUATERNION: {
-			ERR_FAIL_COND_V(len < 4 * 4, ERR_INVALID_DATA);
 			Quaternion val;
-			val.x = decode_float(&buf[0]);
-			val.y = decode_float(&buf[4]);
-			val.z = decode_float(&buf[8]);
-			val.w = decode_float(&buf[12]);
-			r_variant = val;
+			if (type & ENCODE_FLAG_64) {
+				ERR_FAIL_COND_V((size_t)len < sizeof(double) * 4, ERR_INVALID_DATA);
+				val.x = decode_double(&buf[0]);
+				val.y = decode_double(&buf[sizeof(double)]);
+				val.z = decode_double(&buf[sizeof(double) * 2]);
+				val.w = decode_double(&buf[sizeof(double) * 3]);
 
-			if (r_len) {
-				(*r_len) += 4 * 4;
+				if (r_len) {
+					(*r_len) += sizeof(double) * 4;
+				}
+			} else {
+				ERR_FAIL_COND_V((size_t)len < sizeof(float) * 4, ERR_INVALID_DATA);
+				val.x = decode_float(&buf[0]);
+				val.y = decode_float(&buf[sizeof(float)]);
+				val.z = decode_float(&buf[sizeof(float) * 2]);
+				val.w = decode_float(&buf[sizeof(float) * 3]);
+
+				if (r_len) {
+					(*r_len) += sizeof(float) * 4;
+				}
 			}
+			r_variant = val;
 
 		} break;
 		case Variant::AABB: {
-			ERR_FAIL_COND_V(len < 4 * 6, ERR_INVALID_DATA);
 			AABB val;
-			val.position.x = decode_float(&buf[0]);
-			val.position.y = decode_float(&buf[4]);
-			val.position.z = decode_float(&buf[8]);
-			val.size.x = decode_float(&buf[12]);
-			val.size.y = decode_float(&buf[16]);
-			val.size.z = decode_float(&buf[20]);
-			r_variant = val;
+			if (type & ENCODE_FLAG_64) {
+				ERR_FAIL_COND_V((size_t)len < sizeof(double) * 6, ERR_INVALID_DATA);
+				val.position.x = decode_double(&buf[0]);
+				val.position.y = decode_double(&buf[sizeof(double)]);
+				val.position.z = decode_double(&buf[sizeof(double) * 2]);
+				val.size.x = decode_double(&buf[sizeof(double) * 3]);
+				val.size.y = decode_double(&buf[sizeof(double) * 4]);
+				val.size.z = decode_double(&buf[sizeof(double) * 5]);
 
-			if (r_len) {
-				(*r_len) += 4 * 6;
+				if (r_len) {
+					(*r_len) += sizeof(double) * 6;
+				}
+			} else {
+				ERR_FAIL_COND_V((size_t)len < sizeof(float) * 6, ERR_INVALID_DATA);
+				val.position.x = decode_float(&buf[0]);
+				val.position.y = decode_float(&buf[sizeof(float)]);
+				val.position.z = decode_float(&buf[sizeof(float) * 2]);
+				val.size.x = decode_float(&buf[sizeof(float) * 3]);
+				val.size.y = decode_float(&buf[sizeof(float) * 4]);
+				val.size.z = decode_float(&buf[sizeof(float) * 5]);
+
+				if (r_len) {
+					(*r_len) += sizeof(float) * 6;
+				}
 			}
+			r_variant = val;
 
 		} break;
 		case Variant::BASIS: {
-			ERR_FAIL_COND_V(len < 4 * 9, ERR_INVALID_DATA);
 			Basis val;
-			for (int i = 0; i < 3; i++) {
-				for (int j = 0; j < 3; j++) {
-					val.elements[i][j] = decode_float(&buf[(i * 3 + j) * 4]);
+			if (type & ENCODE_FLAG_64) {
+				ERR_FAIL_COND_V((size_t)len < sizeof(double) * 9, ERR_INVALID_DATA);
+				for (int i = 0; i < 3; i++) {
+					for (int j = 0; j < 3; j++) {
+						val.elements[i][j] = decode_double(&buf[(i * 3 + j) * sizeof(double)]);
+					}
 				}
-			}
 
-			r_variant = val;
+				if (r_len) {
+					(*r_len) += sizeof(double) * 9;
+				}
+			} else {
+				ERR_FAIL_COND_V((size_t)len < sizeof(float) * 9, ERR_INVALID_DATA);
+				for (int i = 0; i < 3; i++) {
+					for (int j = 0; j < 3; j++) {
+						val.elements[i][j] = decode_float(&buf[(i * 3 + j) * sizeof(float)]);
+					}
+				}
 
-			if (r_len) {
-				(*r_len) += 4 * 9;
+				if (r_len) {
+					(*r_len) += sizeof(float) * 9;
+				}
 			}
+			r_variant = val;
 
 		} break;
 		case Variant::TRANSFORM3D: {
-			ERR_FAIL_COND_V(len < 4 * 12, ERR_INVALID_DATA);
 			Transform3D val;
-			for (int i = 0; i < 3; i++) {
-				for (int j = 0; j < 3; j++) {
-					val.basis.elements[i][j] = decode_float(&buf[(i * 3 + j) * 4]);
+			if (type & ENCODE_FLAG_64) {
+				ERR_FAIL_COND_V((size_t)len < sizeof(double) * 12, ERR_INVALID_DATA);
+				for (int i = 0; i < 3; i++) {
+					for (int j = 0; j < 3; j++) {
+						val.basis.elements[i][j] = decode_double(&buf[(i * 3 + j) * sizeof(double)]);
+					}
 				}
-			}
-			val.origin[0] = decode_float(&buf[36]);
-			val.origin[1] = decode_float(&buf[40]);
-			val.origin[2] = decode_float(&buf[44]);
+				val.origin[0] = decode_double(&buf[sizeof(double) * 9]);
+				val.origin[1] = decode_double(&buf[sizeof(double) * 10]);
+				val.origin[2] = decode_double(&buf[sizeof(double) * 11]);
 
-			r_variant = val;
+				if (r_len) {
+					(*r_len) += sizeof(double) * 12;
+				}
+			} else {
+				ERR_FAIL_COND_V((size_t)len < sizeof(float) * 12, ERR_INVALID_DATA);
+				for (int i = 0; i < 3; i++) {
+					for (int j = 0; j < 3; j++) {
+						val.basis.elements[i][j] = decode_float(&buf[(i * 3 + j) * sizeof(float)]);
+					}
+				}
+				val.origin[0] = decode_float(&buf[sizeof(float) * 9]);
+				val.origin[1] = decode_float(&buf[sizeof(float) * 10]);
+				val.origin[2] = decode_float(&buf[sizeof(float) * 11]);
 
-			if (r_len) {
-				(*r_len) += 4 * 12;
+				if (r_len) {
+					(*r_len) += sizeof(float) * 12;
+				}
 			}
+			r_variant = val;
 
 		} break;
-
 		// misc types
 		case Variant::COLOR: {
 			ERR_FAIL_COND_V(len < 4 * 4, ERR_INVALID_DATA);
@@ -356,9 +468,8 @@ Error decode_variant(Variant &r_variant, const uint8_t *p_buffer, int p_len, int
 			r_variant = val;
 
 			if (r_len) {
-				(*r_len) += 4 * 4;
+				(*r_len) += 4 * 4; // Colors should always be in single-precision.
 			}
-
 		} break;
 		case Variant::STRING_NAME: {
 			String str;
@@ -463,7 +574,7 @@ Error decode_variant(Variant &r_variant, const uint8_t *p_buffer, int p_len, int
 					buf += 4;
 					len -= 4;
 					if (r_len) {
-						(*r_len) += 4;
+						(*r_len) += 4; // Size of count number.
 					}
 
 					for (int i = 0; i < count; i++) {
@@ -516,7 +627,7 @@ Error decode_variant(Variant &r_variant, const uint8_t *p_buffer, int p_len, int
 			len -= 4;
 
 			if (r_len) {
-				(*r_len) += 4;
+				(*r_len) += 4; // Size of count number.
 			}
 
 			Dictionary d;
@@ -559,7 +670,7 @@ Error decode_variant(Variant &r_variant, const uint8_t *p_buffer, int p_len, int
 			len -= 4;
 
 			if (r_len) {
-				(*r_len) += 4;
+				(*r_len) += 4; // Size of count number.
 			}
 
 			Array varr;
@@ -716,9 +827,8 @@ Error decode_variant(Variant &r_variant, const uint8_t *p_buffer, int p_len, int
 			len -= 4;
 
 			if (r_len) {
-				(*r_len) += 4;
+				(*r_len) += 4; // Size of count number.
 			}
-			//printf("string count: %i\n",count);
 
 			for (int32_t i = 0; i < count; i++) {
 				String str;
@@ -739,30 +849,57 @@ Error decode_variant(Variant &r_variant, const uint8_t *p_buffer, int p_len, int
 			buf += 4;
 			len -= 4;
 
-			ERR_FAIL_MUL_OF(count, 4 * 2, ERR_INVALID_DATA);
-			ERR_FAIL_COND_V(count < 0 || count * 4 * 2 > len, ERR_INVALID_DATA);
 			Vector<Vector2> varray;
 
-			if (r_len) {
-				(*r_len) += 4;
-			}
-
-			if (count) {
-				varray.resize(count);
-				Vector2 *w = varray.ptrw();
+			if (type & ENCODE_FLAG_64) {
+				ERR_FAIL_MUL_OF(count, sizeof(double) * 2, ERR_INVALID_DATA);
+				ERR_FAIL_COND_V(count < 0 || count * sizeof(double) * 2 > (size_t)len, ERR_INVALID_DATA);
 
-				for (int32_t i = 0; i < count; i++) {
-					w[i].x = decode_float(buf + i * 4 * 2 + 4 * 0);
-					w[i].y = decode_float(buf + i * 4 * 2 + 4 * 1);
+				if (r_len) {
+					(*r_len) += 4; // Size of count number.
 				}
 
-				int adv = 4 * 2 * count;
+				if (count) {
+					varray.resize(count);
+					Vector2 *w = varray.ptrw();
+
+					for (int32_t i = 0; i < count; i++) {
+						w[i].x = decode_double(buf + i * sizeof(double) * 2 + sizeof(double) * 0);
+						w[i].y = decode_double(buf + i * sizeof(double) * 2 + sizeof(double) * 1);
+					}
+
+					int adv = sizeof(double) * 2 * count;
+
+					if (r_len) {
+						(*r_len) += adv;
+					}
+					len -= adv;
+					buf += adv;
+				}
+			} else {
+				ERR_FAIL_MUL_OF(count, sizeof(float) * 2, ERR_INVALID_DATA);
+				ERR_FAIL_COND_V(count < 0 || count * sizeof(float) * 2 > (size_t)len, ERR_INVALID_DATA);
 
 				if (r_len) {
-					(*r_len) += adv;
+					(*r_len) += 4; // Size of count number.
 				}
-			}
 
+				if (count) {
+					varray.resize(count);
+					Vector2 *w = varray.ptrw();
+
+					for (int32_t i = 0; i < count; i++) {
+						w[i].x = decode_float(buf + i * sizeof(float) * 2 + sizeof(float) * 0);
+						w[i].y = decode_float(buf + i * sizeof(float) * 2 + sizeof(float) * 1);
+					}
+
+					int adv = sizeof(float) * 2 * count;
+
+					if (r_len) {
+						(*r_len) += adv;
+					}
+				}
+			}
 			r_variant = varray;
 
 		} break;
@@ -772,32 +909,61 @@ Error decode_variant(Variant &r_variant, const uint8_t *p_buffer, int p_len, int
 			buf += 4;
 			len -= 4;
 
-			ERR_FAIL_MUL_OF(count, 4 * 3, ERR_INVALID_DATA);
-			ERR_FAIL_COND_V(count < 0 || count * 4 * 3 > len, ERR_INVALID_DATA);
-
 			Vector<Vector3> varray;
 
-			if (r_len) {
-				(*r_len) += 4;
-			}
-
-			if (count) {
-				varray.resize(count);
-				Vector3 *w = varray.ptrw();
+			if (type & ENCODE_FLAG_64) {
+				ERR_FAIL_MUL_OF(count, sizeof(double) * 3, ERR_INVALID_DATA);
+				ERR_FAIL_COND_V(count < 0 || count * sizeof(double) * 3 > (size_t)len, ERR_INVALID_DATA);
 
-				for (int32_t i = 0; i < count; i++) {
-					w[i].x = decode_float(buf + i * 4 * 3 + 4 * 0);
-					w[i].y = decode_float(buf + i * 4 * 3 + 4 * 1);
-					w[i].z = decode_float(buf + i * 4 * 3 + 4 * 2);
+				if (r_len) {
+					(*r_len) += 4; // Size of count number.
 				}
 
-				int adv = 4 * 3 * count;
+				if (count) {
+					varray.resize(count);
+					Vector3 *w = varray.ptrw();
+
+					for (int32_t i = 0; i < count; i++) {
+						w[i].x = decode_double(buf + i * sizeof(double) * 3 + sizeof(double) * 0);
+						w[i].y = decode_double(buf + i * sizeof(double) * 3 + sizeof(double) * 1);
+						w[i].z = decode_double(buf + i * sizeof(double) * 3 + sizeof(double) * 2);
+					}
+
+					int adv = sizeof(double) * 3 * count;
+
+					if (r_len) {
+						(*r_len) += adv;
+					}
+					len -= adv;
+					buf += adv;
+				}
+			} else {
+				ERR_FAIL_MUL_OF(count, sizeof(float) * 3, ERR_INVALID_DATA);
+				ERR_FAIL_COND_V(count < 0 || count * sizeof(float) * 3 > (size_t)len, ERR_INVALID_DATA);
 
 				if (r_len) {
-					(*r_len) += adv;
+					(*r_len) += 4; // Size of count number.
 				}
-			}
 
+				if (count) {
+					varray.resize(count);
+					Vector3 *w = varray.ptrw();
+
+					for (int32_t i = 0; i < count; i++) {
+						w[i].x = decode_float(buf + i * sizeof(float) * 3 + sizeof(float) * 0);
+						w[i].y = decode_float(buf + i * sizeof(float) * 3 + sizeof(float) * 1);
+						w[i].z = decode_float(buf + i * sizeof(float) * 3 + sizeof(float) * 2);
+					}
+
+					int adv = sizeof(float) * 3 * count;
+
+					if (r_len) {
+						(*r_len) += adv;
+					}
+					len -= adv;
+					buf += adv;
+				}
+			}
 			r_variant = varray;
 
 		} break;
@@ -813,7 +979,7 @@ Error decode_variant(Variant &r_variant, const uint8_t *p_buffer, int p_len, int
 			Vector<Color> carray;
 
 			if (r_len) {
-				(*r_len) += 4;
+				(*r_len) += 4; // Size of count number.
 			}
 
 			if (count) {
@@ -821,6 +987,7 @@ Error decode_variant(Variant &r_variant, const uint8_t *p_buffer, int p_len, int
 				Color *w = carray.ptrw();
 
 				for (int32_t i = 0; i < count; i++) {
+					// Colors should always be in single-precision.
 					w[i].r = decode_float(buf + i * 4 * 4 + 4 * 0);
 					w[i].g = decode_float(buf + i * 4 * 4 + 4 * 1);
 					w[i].b = decode_float(buf + i * 4 * 4 + 4 * 2);
@@ -882,7 +1049,7 @@ Error encode_variant(const Variant &p_variant, uint8_t *r_buffer, int &r_len, bo
 			double d = p_variant;
 			float f = d;
 			if (double(f) != d) {
-				flags |= ENCODE_FLAG_64; //always encode real as double
+				flags |= ENCODE_FLAG_64;
 			}
 		} break;
 		case Variant::OBJECT: {
@@ -1013,11 +1180,11 @@ Error encode_variant(const Variant &p_variant, uint8_t *r_buffer, int &r_len, bo
 		case Variant::VECTOR2: {
 			if (buf) {
 				Vector2 v2 = p_variant;
-				encode_float(v2.x, &buf[0]);
-				encode_float(v2.y, &buf[4]);
+				encode_real(v2.x, &buf[0]);
+				encode_real(v2.y, &buf[sizeof(real_t)]);
 			}
 
-			r_len += 2 * 4;
+			r_len += 2 * sizeof(real_t);
 
 		} break;
 		case Variant::VECTOR2I: {
@@ -1033,12 +1200,12 @@ Error encode_variant(const Variant &p_variant, uint8_t *r_buffer, int &r_len, bo
 		case Variant::RECT2: {
 			if (buf) {
 				Rect2 r2 = p_variant;
-				encode_float(r2.position.x, &buf[0]);
-				encode_float(r2.position.y, &buf[4]);
-				encode_float(r2.size.x, &buf[8]);
-				encode_float(r2.size.y, &buf[12]);
+				encode_real(r2.position.x, &buf[0]);
+				encode_real(r2.position.y, &buf[sizeof(real_t)]);
+				encode_real(r2.size.x, &buf[sizeof(real_t) * 2]);
+				encode_real(r2.size.y, &buf[sizeof(real_t) * 3]);
 			}
-			r_len += 4 * 4;
+			r_len += 4 * sizeof(real_t);
 
 		} break;
 		case Variant::RECT2I: {
@@ -1055,12 +1222,12 @@ Error encode_variant(const Variant &p_variant, uint8_t *r_buffer, int &r_len, bo
 		case Variant::VECTOR3: {
 			if (buf) {
 				Vector3 v3 = p_variant;
-				encode_float(v3.x, &buf[0]);
-				encode_float(v3.y, &buf[4]);
-				encode_float(v3.z, &buf[8]);
+				encode_real(v3.x, &buf[0]);
+				encode_real(v3.y, &buf[sizeof(real_t)]);
+				encode_real(v3.z, &buf[sizeof(real_t) * 2]);
 			}
 
-			r_len += 3 * 4;
+			r_len += 3 * sizeof(real_t);
 
 		} break;
 		case Variant::VECTOR3I: {
@@ -1079,50 +1246,50 @@ Error encode_variant(const Variant &p_variant, uint8_t *r_buffer, int &r_len, bo
 				Transform2D val = p_variant;
 				for (int i = 0; i < 3; i++) {
 					for (int j = 0; j < 2; j++) {
-						memcpy(&buf[(i * 2 + j) * 4], &val.elements[i][j], sizeof(float));
+						memcpy(&buf[(i * 2 + j) * sizeof(real_t)], &val.elements[i][j], sizeof(real_t));
 					}
 				}
 			}
 
-			r_len += 6 * 4;
+			r_len += 6 * sizeof(real_t);
 
 		} break;
 		case Variant::PLANE: {
 			if (buf) {
 				Plane p = p_variant;
-				encode_float(p.normal.x, &buf[0]);
-				encode_float(p.normal.y, &buf[4]);
-				encode_float(p.normal.z, &buf[8]);
-				encode_float(p.d, &buf[12]);
+				encode_real(p.normal.x, &buf[0]);
+				encode_real(p.normal.y, &buf[sizeof(real_t)]);
+				encode_real(p.normal.z, &buf[sizeof(real_t) * 2]);
+				encode_real(p.d, &buf[sizeof(real_t) * 3]);
 			}
 
-			r_len += 4 * 4;
+			r_len += 4 * sizeof(real_t);
 
 		} break;
 		case Variant::QUATERNION: {
 			if (buf) {
 				Quaternion q = p_variant;
-				encode_float(q.x, &buf[0]);
-				encode_float(q.y, &buf[4]);
-				encode_float(q.z, &buf[8]);
-				encode_float(q.w, &buf[12]);
+				encode_real(q.x, &buf[0]);
+				encode_real(q.y, &buf[sizeof(real_t)]);
+				encode_real(q.z, &buf[sizeof(real_t) * 2]);
+				encode_real(q.w, &buf[sizeof(real_t) * 3]);
 			}
 
-			r_len += 4 * 4;
+			r_len += 4 * sizeof(real_t);
 
 		} break;
 		case Variant::AABB: {
 			if (buf) {
 				AABB aabb = p_variant;
-				encode_float(aabb.position.x, &buf[0]);
-				encode_float(aabb.position.y, &buf[4]);
-				encode_float(aabb.position.z, &buf[8]);
-				encode_float(aabb.size.x, &buf[12]);
-				encode_float(aabb.size.y, &buf[16]);
-				encode_float(aabb.size.z, &buf[20]);
+				encode_real(aabb.position.x, &buf[0]);
+				encode_real(aabb.position.y, &buf[sizeof(real_t)]);
+				encode_real(aabb.position.z, &buf[sizeof(real_t) * 2]);
+				encode_real(aabb.size.x, &buf[sizeof(real_t) * 3]);
+				encode_real(aabb.size.y, &buf[sizeof(real_t) * 4]);
+				encode_real(aabb.size.z, &buf[sizeof(real_t) * 5]);
 			}
 
-			r_len += 6 * 4;
+			r_len += 6 * sizeof(real_t);
 
 		} break;
 		case Variant::BASIS: {
@@ -1130,12 +1297,12 @@ Error encode_variant(const Variant &p_variant, uint8_t *r_buffer, int &r_len, bo
 				Basis val = p_variant;
 				for (int i = 0; i < 3; i++) {
 					for (int j = 0; j < 3; j++) {
-						memcpy(&buf[(i * 3 + j) * 4], &val.elements[i][j], sizeof(float));
+						memcpy(&buf[(i * 3 + j) * sizeof(real_t)], &val.elements[i][j], sizeof(real_t));
 					}
 				}
 			}
 
-			r_len += 9 * 4;
+			r_len += 9 * sizeof(real_t);
 
 		} break;
 		case Variant::TRANSFORM3D: {
@@ -1143,16 +1310,16 @@ Error encode_variant(const Variant &p_variant, uint8_t *r_buffer, int &r_len, bo
 				Transform3D val = p_variant;
 				for (int i = 0; i < 3; i++) {
 					for (int j = 0; j < 3; j++) {
-						memcpy(&buf[(i * 3 + j) * 4], &val.basis.elements[i][j], sizeof(float));
+						memcpy(&buf[(i * 3 + j) * sizeof(real_t)], &val.basis.elements[i][j], sizeof(real_t));
 					}
 				}
 
-				encode_float(val.origin.x, &buf[36]);
-				encode_float(val.origin.y, &buf[40]);
-				encode_float(val.origin.z, &buf[44]);
+				encode_real(val.origin.x, &buf[sizeof(real_t) * 9]);
+				encode_real(val.origin.y, &buf[sizeof(real_t) * 10]);
+				encode_real(val.origin.z, &buf[sizeof(real_t) * 11]);
 			}
 
-			r_len += 12 * 4;
+			r_len += 12 * sizeof(real_t);
 
 		} break;
 
@@ -1166,7 +1333,7 @@ Error encode_variant(const Variant &p_variant, uint8_t *r_buffer, int &r_len, bo
 				encode_float(c.a, &buf[12]);
 			}
 
-			r_len += 4 * 4;
+			r_len += 4 * 4; // Colors should always be in single-precision.
 
 		} break;
 		case Variant::RID: {
@@ -1441,13 +1608,13 @@ Error encode_variant(const Variant &p_variant, uint8_t *r_buffer, int &r_len, bo
 				for (int i = 0; i < len; i++) {
 					Vector2 v = data.get(i);
 
-					encode_float(v.x, &buf[0]);
-					encode_float(v.y, &buf[4]);
-					buf += 4 * 2;
+					encode_real(v.x, &buf[0]);
+					encode_real(v.y, &buf[sizeof(real_t)]);
+					buf += sizeof(real_t) * 2;
 				}
 			}
 
-			r_len += 4 * 2 * len;
+			r_len += sizeof(real_t) * 2 * len;
 
 		} break;
 		case Variant::PACKED_VECTOR3_ARRAY: {
@@ -1465,14 +1632,14 @@ Error encode_variant(const Variant &p_variant, uint8_t *r_buffer, int &r_len, bo
 				for (int i = 0; i < len; i++) {
 					Vector3 v = data.get(i);
 
-					encode_float(v.x, &buf[0]);
-					encode_float(v.y, &buf[4]);
-					encode_float(v.z, &buf[8]);
-					buf += 4 * 3;
+					encode_real(v.x, &buf[0]);
+					encode_real(v.y, &buf[sizeof(real_t)]);
+					encode_real(v.z, &buf[sizeof(real_t) * 2]);
+					buf += sizeof(real_t) * 3;
 				}
 			}
 
-			r_len += 4 * 3 * len;
+			r_len += sizeof(real_t) * 3 * len;
 
 		} break;
 		case Variant::PACKED_COLOR_ARRAY: {
@@ -1494,7 +1661,7 @@ Error encode_variant(const Variant &p_variant, uint8_t *r_buffer, int &r_len, bo
 					encode_float(c.g, &buf[4]);
 					encode_float(c.b, &buf[8]);
 					encode_float(c.a, &buf[12]);
-					buf += 4 * 4;
+					buf += 4 * 4; // Colors should always be in single-precision.
 				}
 			}
 
diff --git a/core/io/marshalls.h b/core/io/marshalls.h
index 7fac708f97..3ebed914a3 100644
--- a/core/io/marshalls.h
+++ b/core/io/marshalls.h
@@ -31,10 +31,18 @@
 #ifndef MARSHALLS_H
 #define MARSHALLS_H
 
+#include "core/math/math_defs.h"
 #include "core/object/ref_counted.h"
 #include "core/typedefs.h"
 #include "core/variant/variant.h"
 
+// uintr_t is only for pairing with real_t, and we only need it in here.
+#ifdef REAL_T_IS_DOUBLE
+typedef uint64_t uintr_t;
+#else
+typedef uint32_t uintr_t;
+#endif
+
 /**
   * Miscellaneous helpers for marshalling data types, and encoding
   * in an endian independent way
@@ -50,6 +58,12 @@ union MarshallDouble {
 	double d; ///< double
 };
 
+// Behaves like one of the above, depending on compilation setting.
+union MarshallReal {
+	uintr_t i;
+	real_t r;
+};
+
 static inline unsigned int encode_uint16(uint16_t p_uint, uint8_t *p_arr) {
 	for (int i = 0; i < 2; i++) {
 		*p_arr = p_uint & 0xFF;
@@ -96,6 +110,24 @@ static inline unsigned int encode_double(double p_double, uint8_t *p_arr) {
 	return sizeof(uint64_t);
 }
 
+static inline unsigned int encode_uintr(uintr_t p_uint, uint8_t *p_arr) {
+	for (size_t i = 0; i < sizeof(uintr_t); i++) {
+		*p_arr = p_uint & 0xFF;
+		p_arr++;
+		p_uint >>= 8;
+	}
+
+	return sizeof(uintr_t);
+}
+
+static inline unsigned int encode_real(real_t p_real, uint8_t *p_arr) {
+	MarshallReal mr;
+	mr.r = p_real;
+	encode_uintr(mr.i, p_arr);
+
+	return sizeof(uintr_t);
+}
+
 static inline int encode_cstring(const char *p_string, uint8_t *p_data) {
 	int len = 0;