Preserve pointer provenance in the Rust backend.

Use the new `Pointer` and `Length` types in the Rust backend to emit
code that uses `*mut c_void` and `usize` instead of `i32` when working
with pointers and array lengths.

To represent `PointerOrI64`, use a `MaybeUninit<u64>`, since that type
can hold any `u64` and is documented to also preserve provenance.

This change happens to get the generated Rust code close to supporting
memory64, however it isn't complete; the abi code still emits hard-coded
`+ 4` offsets for loading the length of a pointer+length pair in memory.

Author: sunfishcode

Cargo.lock

@@ -29,11 +29,11 @@ clap = { version = "4.3.19", features = ["derive"] }
 env_logger = "0.10.0"
 indexmap = "2.0.0"
 
-wasmparser = "0.200.0"
-wasm-encoder = "0.200.0"
-wasm-metadata = "0.200.0"
-wit-parser = "0.200.0"
-wit-component = "0.200.0"
+wasmparser = "0.201.0"
+wasm-encoder = "0.201.0"
+wasm-metadata = "0.201.0"
+wit-parser = "0.201.0"
+wit-component = "0.201.0"
 
 wit-bindgen-core = { path = 'crates/core', version = '0.19.2' }
 wit-bindgen-c = { path = 'crates/c', version = '0.19.2' }

Cargo.toml

@@ -2205,12 +2205,15 @@ impl Bindgen for FunctionBindgen<'_, '_> {
                                 op
                             ));
                         }
-                        Bitcast::I32ToI64 => {
+                        Bitcast::I32ToI64 | Bitcast::PToP64 => {
                             results.push(format!("(int64_t) {}", op));
                         }
-                        Bitcast::I64ToI32 => {
+                        Bitcast::I64ToI32 | Bitcast::P64ToP => {
                             results.push(format!("(int32_t) {}", op));
                         }
+                        Bitcast::I64ToP64 | Bitcast::P64ToI64 => {
+                            results.push(format!("{}", op));
+                        }
                         Bitcast::None => results.push(op.to_string()),
                     }
                 }
@@ -2869,11 +2872,17 @@ impl Bindgen for FunctionBindgen<'_, '_> {
                 }
             }
 
-            Instruction::I32Load { offset } => self.load("int32_t", *offset, operands, results),
+            Instruction::I32Load { offset }
+            | Instruction::PointerLoad { offset }
+            | Instruction::LengthLoad { offset } => {
+                self.load("int32_t", *offset, operands, results)
+            }
             Instruction::I64Load { offset } => self.load("int64_t", *offset, operands, results),
             Instruction::F32Load { offset } => self.load("float", *offset, operands, results),
             Instruction::F64Load { offset } => self.load("double", *offset, operands, results),
-            Instruction::I32Store { offset } => self.store("int32_t", *offset, operands),
+            Instruction::I32Store { offset }
+            | Instruction::PointerStore { offset }
+            | Instruction::LengthStore { offset } => self.store("int32_t", *offset, operands),
             Instruction::I64Store { offset } => self.store("int64_t", *offset, operands),
             Instruction::F32Store { offset } => self.store("float", *offset, operands),
             Instruction::F64Store { offset } => self.store("double", *offset, operands),
@@ -3015,6 +3024,9 @@ fn wasm_type(ty: WasmType) -> &'static str {
         WasmType::I64 => "int64_t",
         WasmType::F32 => "float",
         WasmType::F64 => "double",
+        WasmType::Pointer => "uintptr_t",
+        WasmType::PointerOrI64 => "int64_t",
+        WasmType::Length => "size_t",
     }
 }
 

crates/c/src/lib.rs

@@ -113,6 +113,11 @@ def_instruction! {
         /// it, using the specified constant offset.
         F64Load { offset: i32 } : [1] => [1],
 
+        /// Like `I32Load` or `I64Load`, but for loading pointer values.
+        PointerLoad { offset: i32 } : [1] => [1],
+        /// Like `I32Load` or `I64Load`, but for loading array length values.
+        LengthLoad { offset: i32 } : [1] => [1],
+
         /// Pops an `i32` address from the stack and then an `i32` value.
         /// Stores the value in little-endian at the pointer specified plus the
         /// constant `offset`.
@@ -138,6 +143,11 @@ def_instruction! {
         /// constant `offset`.
         F64Store { offset: i32 } : [2] => [0],
 
+        /// Like `I32Store` or `I64Store`, but for storing pointer values.
+        PointerStore { offset: i32 } : [2] => [0],
+        /// Like `I32Store` or `I64Store`, but for storing array length values.
+        LengthStore { offset: i32 } : [2] => [0],
+
         // Scalar lifting/lowering
 
         /// Converts an interface type `char` value to a 32-bit integer
@@ -526,6 +536,12 @@ pub enum Bitcast {
     I64ToI32,
     I64ToF32,
 
+    // Pointers
+    P64ToI64,
+    I64ToP64,
+    P64ToP,
+    PToP64,
+
     None,
 }
 
@@ -1517,9 +1533,9 @@ impl<'a, B: Bindgen> Generator<'a, B> {
         // and the length into the high address.
         self.lower(ty);
         self.stack.push(addr.clone());
-        self.emit(&Instruction::I32Store { offset: offset + 4 });
+        self.emit(&Instruction::LengthStore { offset: offset + 4 });
         self.stack.push(addr);
-        self.emit(&Instruction::I32Store { offset });
+        self.emit(&Instruction::PointerStore { offset });
     }
 
     fn write_fields_to_memory<'b>(
@@ -1689,9 +1705,9 @@ impl<'a, B: Bindgen> Generator<'a, B> {
         // Read the pointer/len and then perform the standard lifting
         // proceses.
         self.stack.push(addr.clone());
-        self.emit(&Instruction::I32Load { offset });
+        self.emit(&Instruction::PointerLoad { offset });
         self.stack.push(addr);
-        self.emit(&Instruction::I32Load { offset: offset + 4 });
+        self.emit(&Instruction::LengthLoad { offset: offset + 4 });
         self.lift(ty);
     }
 
@@ -1742,9 +1758,9 @@ impl<'a, B: Bindgen> Generator<'a, B> {
         match *ty {
             Type::String => {
                 self.stack.push(addr.clone());
-                self.emit(&Instruction::I32Load { offset });
+                self.emit(&Instruction::PointerLoad { offset });
                 self.stack.push(addr);
-                self.emit(&Instruction::I32Load { offset: offset + 4 });
+                self.emit(&Instruction::LengthLoad { offset: offset + 4 });
                 self.emit(&Instruction::GuestDeallocateString);
             }
 
@@ -1772,9 +1788,9 @@ impl<'a, B: Bindgen> Generator<'a, B> {
                     self.finish_block(0);
 
                     self.stack.push(addr.clone());
-                    self.emit(&Instruction::I32Load { offset });
+                    self.emit(&Instruction::PointerLoad { offset });
                     self.stack.push(addr);
-                    self.emit(&Instruction::I32Load { offset: offset + 4 });
+                    self.emit(&Instruction::LengthLoad { offset: offset + 4 });
                     self.emit(&Instruction::GuestDeallocateList { element });
                 }
 
@@ -1862,7 +1878,12 @@ fn cast(from: WasmType, to: WasmType) -> Bitcast {
     use WasmType::*;
 
     match (from, to) {
-        (I32, I32) | (I64, I64) | (F32, F32) | (F64, F64) => Bitcast::None,
+        (I32, I32)
+        | (I64, I64)
+        | (F32, F32)
+        | (F64, F64)
+        | (Pointer, Pointer)
+        | (Length, Length) => Bitcast::None,
 
         (I32, I64) => Bitcast::I32ToI64,
         (F32, I32) => Bitcast::F32ToI32,
@@ -1875,7 +1896,19 @@ fn cast(from: WasmType, to: WasmType) -> Bitcast {
         (F32, I64) => Bitcast::F32ToI64,
         (I64, F32) => Bitcast::I64ToF32,
 
-        (F32, F64) | (F64, F32) | (F64, I32) | (I32, F64) => unreachable!(),
+        (I64, PointerOrI64) => Bitcast::I64ToP64,
+        (PointerOrI64, I64) => Bitcast::P64ToI64,
+        (Pointer, PointerOrI64) => Bitcast::PToP64,
+        (PointerOrI64, Pointer) => Bitcast::P64ToP,
+
+        (Pointer | PointerOrI64 | Length, _)
+        | (_, Pointer | PointerOrI64 | Length)
+        | (F32, F64)
+        | (F64, F32)
+        | (F64, I32)
+        | (I32, F64) => {
+            unreachable!()
+        }
     }
 }