Implement the GC in Rust

This implements the current garbage collector in Rust. No changes were
made to the GC design -- it's just ports the one implemented in code
generator to Rust.

The goals are:

- Evaluate Rust for Motoko's RTS implementation
- Make the collector easier to read, understand, modify, and extend.

Current status: the code is complete in the sense that there aren't any
missing features/passes etc., but it's has bugs. I'm not sure how to
debug Wasm yet. There are also lots of TODOs in the code, mostly for
documentation.

Submitting a PR to get early feedback.

.gitignore

@@ -3,6 +3,7 @@
 _out
 _output
 _build
+target
 
 **/*~
 /result*

default.nix

@@ -22,16 +22,18 @@ let haskellPackages = nixpkgs.haskellPackages.override {
       overrides = import nix/haskell-packages.nix nixpkgs subpath;
     }; in
 let
-  llvmBuildInputs = [
+  rtsBuildInputs = [
     nixpkgs.clang_10 # for native/wasm building
     nixpkgs.lld_10 # for wasm building
+    nixpkgs.rustc
+    nixpkgs.cargo
   ];
 
-  # When compiling natively, we want to use `clang` (which is a nixpkgs
-  # provided wrapper that sets various include paths etc).
-  # But for some reason it does not handle building for Wasm well, so
-  # there we use plain clang-10. There is no stdlib there anyways.
   llvmEnv = ''
+    # When compiling natively, we want to use `clang` (which is a nixpkgs
+    # provided wrapper that sets various include paths etc).
+    # But for some reason it does not handle building for Wasm well, so
+    # there we use plain clang-10. There is no stdlib there anyways.
     export CLANG="${nixpkgs.clang_10}/bin/clang"
     export WASM_CLANG="clang-10"
     export WASM_LD=wasm-ld
@@ -129,7 +131,7 @@ rec {
     src = subpath ./rts;
     nativeBuildInputs = [ nixpkgs.makeWrapper ];
 
-    buildInputs = llvmBuildInputs;
+    buildInputs = rtsBuildInputs;
 
     preBuild = ''
       ${llvmEnv}
@@ -203,7 +205,7 @@ rec {
             wasmtime
             nixpkgs.sources.esm
           ] ++
-          llvmBuildInputs;
+          rtsBuildInputs;
 
         checkPhase = ''
             patchShebangs .

nix/default.nix

@@ -34,6 +34,17 @@ let
             wasmtime = self.callPackage ./wasmtime.nix {};
           }
         )
+        # nixpkgs's rustc does not include the wasm32-unknown-unknown target, so
+        # lets add it here.
+        (self: super: {
+          rustc = super.rustc.overrideAttrs (old: {
+	    configureFlags = self.lib.lists.forEach old.configureFlags (flag:
+              if self.lib.strings.hasPrefix "--target=" flag
+              then flag + ",wasm32-unknown-unknown,wasm32-unknown-emscripten"
+              else flag
+            );
+          });
+        })
       ];
     };
 in

rts/Makefile

@@ -1,3 +1,5 @@
+SHELL:=bash -O globstar
+
 CLANG ?= clang-10
 WASM_CLANG ?= clang-10
 WASM_LD ?= wasm-ld-10
@@ -154,14 +156,28 @@ _build/native/tommath_%.o: %.c rts.h buf.h | _build/native
 _build/wasm/musl_%.o: %.c | _build/wasm
 	$(WASM_CLANG) $(CLANG_FLAGS) $(WASM_FLAGS) $(MUSL_FLAGS) $< --output $@
 
+.PHONY: _build/wasm/motoko_rts.o
+_build/wasm/motoko_rts.o: | _build/wasm
+	cd motoko-rts && \
+	  cargo rustc --target=wasm32-unknown-emscripten --release -v -- -Crelocation-model=pic --emit=obj
+	cp motoko-rts/target/wasm32-unknown-emscripten/release/deps/motoko_rts*.o $@
+
+.PHONY: _build/native/motoko_rts.o
+_build/native/motoko_rts.o: | _build/native
+	cd motoko-rts && \
+	  cargo rustc --release -v -- -Crelocation-model=pic --emit=obj
+	cp motoko-rts/target/release/deps/motoko_rts*.o $@
+
 RTS_WASM_O=$(RTSFILES:%=_build/wasm/%.o)
 RTS_NATIVE_O=$(RTSFILES:%=_build/native/%.o)
+RTS_RUST_WASM_O=_build/wasm/motoko_rts.o
+RTS_RUST_NATIVE_O=_build/native/motoko_rts.o
 
 #
 # The actual RTS, as we ship it with the compiler
 #
 
-mo-rts.wasm: $(RTS_WASM_O) $(TOMMATH_WASM_O) $(MUSL_WASM_O)
+mo-rts.wasm: $(RTS_WASM_O) $(RTS_RUST_WASM_O) $(TOMMATH_WASM_O) $(MUSL_WASM_O)
 	$(WASM_LD) -o $@ \
 		--import-memory --shared --no-entry --gc-sections \
 		--export=__wasm_call_ctors \
@@ -171,12 +187,11 @@ mo-rts.wasm: $(RTS_WASM_O) $(TOMMATH_WASM_O) $(MUSL_WASM_O)
 # A simple program to do simple tests of rts.c, using native compilation
 #
 
-test_rts: test_rts.c $(RTS_NATIVE_O) $(TOMMATH_NATIVE_O)
+test_rts: test_rts.c $(RTS_NATIVE_O) $(RTS_RUST_NATIVE_O) $(TOMMATH_NATIVE_O)
 	$(CLANG) -o $@ $+
 
-test_leb128: test_leb128.c $(RTS_NATIVE_O) $(TOMMATH_NATIVE_O)
+test_leb128: test_leb128.c $(RTS_NATIVE_O) $(RTS_RUST_NATIVE_O) $(TOMMATH_NATIVE_O)
 	$(CLANG) -o $@ $+
 
 clean:
-	rm -rf _build mo-rts.wasm test_rts test_leb128
-
+	rm -rf _build mo-rts.wasm test_rts test_leb128 motoko-rts/target

rts/motoko-rts/.vim/coc-settings.json

@@ -0,0 +1,3 @@
+{
+    "rust-analyzer.cargo.target": "wasm32-unknown-emscripten"
+}

rts/motoko-rts/Cargo.toml

@@ -0,0 +1,7 @@
+[package]
+name = "motoko-rts"
+version = "0.1.0"
+authors = ["Ömer Sinan Ağacan <omeragacan@gmail.com>"]
+edition = "2018"
+
+[dependencies]

rts/motoko-rts/src/array.rs

@@ -0,0 +1,21 @@
+use crate::common::rts_trap_with;
+use crate::types::{skew, Array, SkewedPtr};
+
+/// Returns address of Nth payload field of an array
+#[no_mangle]
+pub unsafe extern "C" fn array_field_addr(array: SkewedPtr, idx: u32) -> SkewedPtr {
+    let array_ptr = array.unskew() as *const Array;
+
+    if idx >= (*array_ptr).len {
+        rts_trap_with("Array index out of bounds\0".as_ptr());
+    }
+
+    let payload_begin = array_ptr.offset(1) as *const u32;
+    skew(payload_begin.offset(idx as isize) as usize)
+}
+
+/// Index an array. Does not check bounds.
+pub unsafe fn array_idx_unchecked(array_ptr: *const Array, idx: u32) -> SkewedPtr {
+    let payload_begin = array_ptr.offset(1) as *const u32;
+    SkewedPtr(*payload_begin.offset(idx as isize) as usize)
+}

rts/motoko-rts/src/common.rs

@@ -0,0 +1,3 @@
+extern "C" {
+    pub fn rts_trap_with(msg: *const u8) -> !;
+}