Represent arguments without references directly as a databuf

design/TmpWireFormat.md

@@ -13,8 +13,9 @@ It also does not support all features that we want to support eventually. In
 particular, it does not support subtyping.
 
 Some types have a *specialized argument format* when used directly as a
-function arguments, rather than nested inside a data structure. All others use
-the *general argument format*.
+function arguments, rather than nested inside a data structure. Other types use
+the _general argument format (without references)_ or the _general argument
+format (with references)_.
 
 Each argument of a function is serialized separately. If the function is
 defined with a list of arguments, these all become arguments of the WebAssembly
@@ -36,16 +37,13 @@ Note that there is no terminating `\0`, and the length is implicit as the
 length of the `databuf`.
 
 
-General argument format
------------------------
+General argument format (without references)
+--------------------------------------------
 
-All other arguments are represented as a non-empty `elembuf` where
-
- * the first entry is a `databuf` containing the actual data (see below)
- * all further entries are the references contained in the data.
-
-The `databuf` is generated by an in-order traversal of the data type.
-All numbers are fixed-width and in little endian format.
+Arguments with a type that does not mention any reference types (no actors, no
+shared functions), are represented as a `databuf`. This `databuf` is generated
+by an in-order traversal of the data type.  All numbers are fixed-width and in
+little endian format.
 
  * A `Nat`, `Int` or `Word64` is represented by 8 bytes.
  * A `Word32` is represented by 4 bytes.
@@ -63,9 +61,42 @@ All numbers are fixed-width and in little endian format.
    are statically known.)
  * An `Option` is represented by a single byte `0` if it is `null`, or
    otherwise by a single byte `1` followed by the representation of the value
- * A reference (`actor`, `shared func`) is represented as a 32 bit number (4
-   bytes) that is an index into the surrounding `elembuf`. This is never `0`, as
-   the first entry in the `elembuf` is the `databuf` with the actual data.
  * An empty tuple, the type `Null` and the type `Shared` are represented by
    zero bytes.
 
+
+*Example:* The ActorScript value
+```
+(null, ?4, "!") : (?Text, ?Int, Text)
+```
+is represented as
+```
+00 01 04 00 00 00 00 00 00 00 01 21
+```
+
+General argument format (with references)
+-----------------------------------------
+
+Argument with a type that mentions reference types (actors or shared functions)
+are represented as an `elembuf`:
+
+ * the first entry is a `databuf` contains the data according to the format
+   above.
+ * all further entries are the references contained in the data.
+
+The above format is thus extended with the following case:
+
+ * A reference (`actor`, `shared func`) is represented as a 32 bit number (4
+   bytes). Thus number is an index into the surrounding `elembuf`.
+
+   NB: The index is never never `0`, as the first entry in the `elembuf` is the
+   `databuf` with the actual data.
+
+*Example:* The ActorScript value
+```
+(null, ?console) : (?actor {}, ?actor {log : Text -> () })
+```
+is represented as
+```
+elembuf [databuf [00 01 01 00 00 00], console]
+```

src/compile.ml

@@ -2458,6 +2458,40 @@ module Serialization = struct
   let typ_id : Type.typ -> string = Type.string_of_typ
 
 
+
+  (* Checks whether the serialization of a given type could contain references *)
+  module TS = Set.Make (struct type t = Type.typ let compare = compare end)
+  let has_no_references : Type.typ -> bool = fun t ->
+    let open Type in
+    let seen = ref TS.empty in (* break the cycles *)
+    let rec go t =
+      TS.mem t !seen ||
+      begin
+        seen := TS.add t !seen;
+        match t with
+        | Var _ -> assert false
+        | (Prim _ | Any | Non | Shared | Pre) -> true
+        | Con (c, ts) ->
+          begin match Con.kind c with
+          | Abs _ -> assert false
+          | Def (tbs,t) -> go (open_ ts t) (* TBR this may fail to terminate *)
+          end
+        | Array t -> go t
+        | Tup ts -> List.for_all go ts
+        | Func (Sharable, c, tbs, ts1, ts2) -> false
+        | Func (s, c, tbs, ts1, ts2) ->
+          let ts = open_binds tbs in
+          List.for_all go (List.map (open_ ts) ts1) &&
+          List.for_all go (List.map (open_ ts) ts2)
+        | Opt t -> go t
+        | Async t -> go t
+        | Obj (Actor, fs) -> false
+        | Obj (s, fs) -> List.for_all (fun f -> go f.typ) fs
+        | Mut t -> go t
+        | Serialized t -> go t
+      end
+    in go t
+
   (* Returns data (in bytes) and reference buffer size (in entries) needed *)
   let rec buffer_size env t =
     let open Type in
@@ -2830,10 +2864,20 @@ module Serialization = struct
         G.i (Call (nr (Dfinity.data_externalize_i env))) ^^
         store_unskewed_ptr  ^^
 
-        (* Finally, create elembuf *)
-        get_refs_start ^^
-        get_refs_size ^^ compile_add_const 1l ^^
-        G.i (Call (nr (Dfinity.elem_externalize_i env)))
+        if has_no_references t
+        then
+          (* Sanity check: Really no references *)
+          get_refs_size ^^
+          G.i (Test (Wasm.Values.I32 I32Op.Eqz)) ^^
+          G.if_ (ValBlockType None) G.nop (G.i Unreachable) ^^
+          (* If there are no references, just return the databuf *)
+          get_refs_start ^^
+          load_unskewed_ptr
+        else
+          (* Finally, create elembuf *)
+          get_refs_start ^^
+          get_refs_size ^^ compile_add_const 1l ^^
+          G.i (Call (nr (Dfinity.elem_externalize_i env)))
     )
 
   let deserialize_text env get_databuf =
@@ -2871,28 +2915,36 @@ module Serialization = struct
         let (set_refs_start, get_refs_start) = new_local env "refs_start" in
         let (set_databuf, get_databuf) = new_local env "databuf" in
 
-        (* Allocate space for the elem buffer *)
-        get_elembuf ^^
-        G.i (Call (nr (Dfinity.elem_length_i env))) ^^
-        set_refs_size ^^
-
-        get_refs_size ^^
-        Array.alloc env ^^
-        compile_add_const Array.header_size ^^
-        compile_add_const ptr_unskew ^^
-        set_refs_start ^^
-
-        (* Copy elembuf *)
-        get_refs_start ^^
-        get_refs_size ^^
-        get_elembuf ^^
-        compile_unboxed_const 0l ^^
-        G.i (Call (nr (Dfinity.elem_internalize_i env))) ^^
-
-        (* Get databuf *)
-        get_refs_start ^^
-        load_unskewed_ptr ^^
-        set_databuf ^^
+        begin
+        if has_no_references t
+        then
+          (* We have no elembuf wrapper, so the argument is the databuf *)
+          compile_unboxed_const 0l ^^ set_refs_start ^^
+          get_elembuf ^^ set_databuf
+        else
+          (* Allocate space for the elem buffer *)
+          get_elembuf ^^
+          G.i (Call (nr (Dfinity.elem_length_i env))) ^^
+          set_refs_size ^^
+
+          get_refs_size ^^
+          Array.alloc env ^^
+          compile_add_const Array.header_size ^^
+          compile_add_const ptr_unskew ^^
+          set_refs_start ^^
+
+          (* Copy elembuf *)
+          get_refs_start ^^
+          get_refs_size ^^
+          get_elembuf ^^
+          compile_unboxed_const 0l ^^
+          G.i (Call (nr (Dfinity.elem_internalize_i env))) ^^
+
+          (* Get databuf *)
+          get_refs_start ^^
+          load_unskewed_ptr ^^
+          set_databuf
+        end ^^
 
         (* Allocate space for the data buffer *)
         get_databuf ^^
@@ -2925,6 +2977,7 @@ module Serialization = struct
       | Type.Prim Type.Text -> CustomSections.DataBuf
       | Type.Prim Type.Word32 -> CustomSections.I32
       | Type.Obj (Type.Actor, _) -> CustomSections.ActorRef
+      | t' when has_no_references t' -> CustomSections.DataBuf
       | _ -> CustomSections.ElemBuf
 
 end (* Serialization *)

test/run-dfinity/ok/nary-async.wasm.stderr.ok

@@ -1,3 +1,53 @@
 deserialize: T/77
-serialize: T/77
-buffer_size: T/77
+prelude:103.1-128.2: internal error, File "compile.ml", line 2476, characters 21-27: Assertion failed
+
+Last environment:
+@new_async = func
+Array_init = func
+Array_tabulate = func
+abs = func
+btstWord16 = func
+btstWord32 = func
+btstWord64 = func
+btstWord8 = func
+charToWord32 = func
+clzWord16 = func
+clzWord32 = func
+clzWord64 = func
+clzWord8 = func
+ctzWord16 = func
+ctzWord32 = func
+ctzWord64 = func
+ctzWord8 = func
+hashInt = func
+ignore = func
+intToWord16 = func
+intToWord32 = func
+intToWord64 = func
+intToWord8 = func
+natToWord16 = func
+natToWord32 = func
+natToWord64 = func
+natToWord8 = func
+popcntWord16 = func
+popcntWord32 = func
+popcntWord64 = func
+popcntWord8 = func
+print = func
+printInt = func
+range = func
+revrange = func
+shrsWord16 = func
+shrsWord32 = func
+shrsWord64 = func
+shrsWord8 = func
+word16ToInt = func
+word16ToNat = func
+word32ToChar = func
+word32ToInt = func
+word32ToNat = func
+word64ToInt = func
+word64ToNat = func
+word8ToInt = func
+word8ToNat = func
+