| 1 | use crate::encoding::types::{FunctionKey, ValtypeEncoder}; |
| 2 | use anyhow::Result; |
| 3 | use indexmap::IndexSet; |
| 4 | use std::collections::HashMap; |
| 5 | use std::mem; |
| 6 | use wasm_encoder::*; |
| 7 | use wit_parser::*; |
| 8 | |
| 9 | /// Encodes the given `package` within `resolve` to a binary WebAssembly |
| 10 | /// representation. |
| 11 | /// |
| 12 | /// This function is the root of the implementation of serializing a WIT package |
| 13 | /// into a WebAssembly representation. The wasm representation serves two |
| 14 | /// purposes: |
| 15 | /// |
| 16 | /// * One is to be a binary encoding of a WIT document which is ideally more |
| 17 | /// stable than the WIT textual format itself. |
| 18 | /// * Another is to provide a clear mapping of all WIT features into the |
| 19 | /// component model through use of its binary representation. |
| 20 | /// |
| 21 | /// The `resolve` provided is a set of packages and types and such and the |
| 22 | /// `package` argument is an ID within the world provided. The documents within |
| 23 | /// `package` will all be encoded into the binary returned. |
| 24 | /// |
| 25 | /// The binary returned can be [`decode`d](crate::decode) to recover the WIT |
| 26 | /// package provided. |
| 27 | pub fn encode(resolve: &Resolve, package: PackageId) -> Result<Vec<u8>> { |
| 28 | let mut component: ComponentBuilder = encode_component(resolve, package)?; |
| 29 | component.raw_custom_section(&crate::base_producers().raw_custom_section()); |
| 30 | Ok(component.finish()) |
| 31 | } |
| 32 | |
| 33 | /// Encodes the given `package` within `resolve` to a binary WebAssembly |
| 34 | /// representation. |
| 35 | /// |
| 36 | /// This function is the root of the implementation of serializing a WIT package |
| 37 | /// into a WebAssembly representation. The wasm representation serves two |
| 38 | /// purposes: |
| 39 | /// |
| 40 | /// * One is to be a binary encoding of a WIT document which is ideally more |
| 41 | /// stable than the WIT textual format itself. |
| 42 | /// * Another is to provide a clear mapping of all WIT features into the |
| 43 | /// component model through use of its binary representation. |
| 44 | /// |
| 45 | /// The `resolve` provided is a set of packages and types and such and the |
| 46 | /// `package` argument is an ID within the world provided. The documents within |
| 47 | /// `package` will all be encoded into the binary returned. |
| 48 | /// |
| 49 | /// The binary returned can be [`decode`d](crate::decode) to recover the WIT |
| 50 | /// package provided. |
| 51 | pub fn encode_component(resolve: &Resolve, package: PackageId) -> Result<ComponentBuilder> { |
| 52 | let mut encoder: Encoder<'_> = Encoder { |
| 53 | component: ComponentBuilder::default(), |
| 54 | resolve, |
| 55 | package, |
| 56 | }; |
| 57 | encoder.run()?; |
| 58 | |
| 59 | let package_metadata: PackageMetadata = PackageMetadata::extract(resolve, package); |
| 60 | encoder.component.custom_section(&CustomSection { |
| 61 | name: PackageMetadata::SECTION_NAME.into(), |
| 62 | data: package_metadata.encode()?.into(), |
| 63 | }); |
| 64 | |
| 65 | Ok(encoder.component) |
| 66 | } |
| 67 | |
| 68 | /// Encodes a `world` as a component type. |
| 69 | pub fn encode_world(resolve: &Resolve, world_id: WorldId) -> Result<ComponentType> { |
| 70 | let mut component = InterfaceEncoder::new(resolve); |
| 71 | let world = &resolve.worlds[world_id]; |
| 72 | log::trace!("encoding world {}" , world.name); |
| 73 | |
| 74 | // This sort is similar in purpose to the sort below in |
| 75 | // `encode_instance`, but different in its sort. The purpose here is |
| 76 | // to ensure that when a document is either printed as WIT or |
| 77 | // encoded as wasm that decoding from those artifacts produces the |
| 78 | // same WIT package. Namely both encoding processes should encode |
| 79 | // things in the same order. |
| 80 | // |
| 81 | // When printing worlds in WIT freestanding function imports are |
| 82 | // printed first, then types. Resource functions are attached to |
| 83 | // types which means that they all come last. Sort all |
| 84 | // resource-related functions here to the back of the `imports` list |
| 85 | // while keeping everything else in front, using a stable sort to |
| 86 | // preserve preexisting ordering. |
| 87 | let mut imports = world.imports.iter().collect::<Vec<_>>(); |
| 88 | imports.sort_by_key(|(_name, import)| match import { |
| 89 | WorldItem::Function(f) => match f.kind { |
| 90 | FunctionKind::Freestanding => 0, |
| 91 | _ => 1, |
| 92 | }, |
| 93 | _ => 0, |
| 94 | }); |
| 95 | |
| 96 | // Encode the imports |
| 97 | for (name, import) in imports { |
| 98 | let name = resolve.name_world_key(name); |
| 99 | log::trace!("encoding import {name}" ); |
| 100 | let ty = match import { |
| 101 | WorldItem::Interface { id, .. } => { |
| 102 | component.interface = Some(*id); |
| 103 | let idx = component.encode_instance(*id)?; |
| 104 | ComponentTypeRef::Instance(idx) |
| 105 | } |
| 106 | WorldItem::Function(f) => { |
| 107 | component.interface = None; |
| 108 | let idx = component.encode_func_type(resolve, f)?; |
| 109 | ComponentTypeRef::Func(idx) |
| 110 | } |
| 111 | WorldItem::Type(t) => { |
| 112 | component.interface = None; |
| 113 | component.import_types = true; |
| 114 | component.encode_valtype(resolve, &Type::Id(*t))?; |
| 115 | component.import_types = false; |
| 116 | continue; |
| 117 | } |
| 118 | }; |
| 119 | component.outer.import(&name, ty); |
| 120 | } |
| 121 | // Encode the exports |
| 122 | for (name, export) in world.exports.iter() { |
| 123 | let name = resolve.name_world_key(name); |
| 124 | log::trace!("encoding export {name}" ); |
| 125 | let ty = match export { |
| 126 | WorldItem::Interface { id, .. } => { |
| 127 | component.interface = Some(*id); |
| 128 | let idx = component.encode_instance(*id)?; |
| 129 | ComponentTypeRef::Instance(idx) |
| 130 | } |
| 131 | WorldItem::Function(f) => { |
| 132 | component.interface = None; |
| 133 | let idx = component.encode_func_type(resolve, f)?; |
| 134 | ComponentTypeRef::Func(idx) |
| 135 | } |
| 136 | WorldItem::Type(_) => unreachable!(), |
| 137 | }; |
| 138 | component.outer.export(&name, ty); |
| 139 | } |
| 140 | |
| 141 | Ok(component.outer) |
| 142 | } |
| 143 | |
| 144 | struct Encoder<'a> { |
| 145 | component: ComponentBuilder, |
| 146 | resolve: &'a Resolve, |
| 147 | package: PackageId, |
| 148 | } |
| 149 | |
| 150 | impl Encoder<'_> { |
| 151 | fn run(&mut self) -> Result<()> { |
| 152 | // Encode all interfaces as component types and then export them. |
| 153 | for (name, &id) in self.resolve.packages[self.package].interfaces.iter() { |
| 154 | let component_ty = self.encode_interface(id)?; |
| 155 | let ty = self.component.type_component(&component_ty); |
| 156 | self.component |
| 157 | .export(name.as_ref(), ComponentExportKind::Type, ty, None); |
| 158 | } |
| 159 | |
| 160 | // For each `world` encode it directly as a component and then create a |
| 161 | // wrapper component that exports that component. |
| 162 | for (name, &world) in self.resolve.packages[self.package].worlds.iter() { |
| 163 | let component_ty = encode_world(self.resolve, world)?; |
| 164 | |
| 165 | let world = &self.resolve.worlds[world]; |
| 166 | let mut wrapper = ComponentType::new(); |
| 167 | wrapper.ty().component(&component_ty); |
| 168 | let pkg = &self.resolve.packages[world.package.unwrap()]; |
| 169 | wrapper.export(&pkg.name.interface_id(name), ComponentTypeRef::Component(0)); |
| 170 | |
| 171 | let ty = self.component.type_component(&wrapper); |
| 172 | self.component |
| 173 | .export(name.as_ref(), ComponentExportKind::Type, ty, None); |
| 174 | } |
| 175 | |
| 176 | Ok(()) |
| 177 | } |
| 178 | |
| 179 | fn encode_interface(&mut self, id: InterfaceId) -> Result<ComponentType> { |
| 180 | // Build a set of interfaces reachable from this document, including the |
| 181 | // interfaces in the document itself. This is used to import instances |
| 182 | // into the component type we're encoding. Note that entire interfaces |
| 183 | // are imported with all their types as opposed to just the needed types |
| 184 | // in an interface for this document. That's done to assist with the |
| 185 | // decoding process where everyone's view of a foreign document agrees |
| 186 | // notably on the order that types are defined in to assist with |
| 187 | // roundtripping. |
| 188 | let mut interfaces = IndexSet::new(); |
| 189 | self.add_live_interfaces(&mut interfaces, id); |
| 190 | |
| 191 | // Seed the set of used names with all exported interfaces to ensure |
| 192 | // that imported interfaces choose different names as the import names |
| 193 | // aren't used during decoding. |
| 194 | let mut used_names = IndexSet::new(); |
| 195 | for id in interfaces.iter() { |
| 196 | let iface = &self.resolve.interfaces[*id]; |
| 197 | if iface.package == Some(self.package) { |
| 198 | let first = used_names.insert(iface.name.as_ref().unwrap().clone()); |
| 199 | assert!(first); |
| 200 | } |
| 201 | } |
| 202 | |
| 203 | let mut encoder = InterfaceEncoder::new(self.resolve); |
| 204 | for interface in interfaces { |
| 205 | encoder.interface = Some(interface); |
| 206 | let iface = &self.resolve.interfaces[interface]; |
| 207 | let name = self.resolve.id_of(interface).unwrap(); |
| 208 | if interface == id { |
| 209 | let idx = encoder.encode_instance(interface)?; |
| 210 | log::trace!("exporting self as {idx}" ); |
| 211 | encoder.outer.export(&name, ComponentTypeRef::Instance(idx)); |
| 212 | } else { |
| 213 | encoder.push_instance(); |
| 214 | for (_, id) in iface.types.iter() { |
| 215 | encoder.encode_valtype(self.resolve, &Type::Id(*id))?; |
| 216 | } |
| 217 | let instance = encoder.pop_instance(); |
| 218 | let idx = encoder.outer.type_count(); |
| 219 | encoder.outer.ty().instance(&instance); |
| 220 | encoder.import_map.insert(interface, encoder.instances); |
| 221 | encoder.instances += 1; |
| 222 | encoder.outer.import(&name, ComponentTypeRef::Instance(idx)); |
| 223 | } |
| 224 | } |
| 225 | |
| 226 | encoder.interface = None; |
| 227 | |
| 228 | Ok(encoder.outer) |
| 229 | } |
| 230 | |
| 231 | /// Recursively add all live interfaces reachable from `id` into the |
| 232 | /// `interfaces` set, and then add `id` to the set. |
| 233 | fn add_live_interfaces(&self, interfaces: &mut IndexSet<InterfaceId>, id: InterfaceId) { |
| 234 | if interfaces.contains(&id) { |
| 235 | return; |
| 236 | } |
| 237 | for id in self.resolve.interface_direct_deps(id) { |
| 238 | self.add_live_interfaces(interfaces, id); |
| 239 | } |
| 240 | assert!(interfaces.insert(id)); |
| 241 | } |
| 242 | } |
| 243 | |
| 244 | struct InterfaceEncoder<'a> { |
| 245 | resolve: &'a Resolve, |
| 246 | outer: ComponentType, |
| 247 | ty: Option<InstanceType>, |
| 248 | func_type_map: HashMap<FunctionKey<'a>, u32>, |
| 249 | type_map: HashMap<TypeId, u32>, |
| 250 | saved_types: Option<(HashMap<TypeId, u32>, HashMap<FunctionKey<'a>, u32>)>, |
| 251 | import_map: HashMap<InterfaceId, u32>, |
| 252 | outer_type_map: HashMap<TypeId, u32>, |
| 253 | instances: u32, |
| 254 | import_types: bool, |
| 255 | interface: Option<InterfaceId>, |
| 256 | } |
| 257 | |
| 258 | impl InterfaceEncoder<'_> { |
| 259 | fn new(resolve: &Resolve) -> InterfaceEncoder<'_> { |
| 260 | InterfaceEncoder { |
| 261 | resolve, |
| 262 | outer: ComponentType::new(), |
| 263 | ty: None, |
| 264 | type_map: Default::default(), |
| 265 | func_type_map: Default::default(), |
| 266 | import_map: Default::default(), |
| 267 | outer_type_map: Default::default(), |
| 268 | instances: 0, |
| 269 | saved_types: None, |
| 270 | import_types: false, |
| 271 | interface: None, |
| 272 | } |
| 273 | } |
| 274 | |
| 275 | fn encode_instance(&mut self, interface: InterfaceId) -> Result<u32> { |
| 276 | self.push_instance(); |
| 277 | let iface = &self.resolve.interfaces[interface]; |
| 278 | let mut type_order = IndexSet::new(); |
| 279 | for (_, id) in iface.types.iter() { |
| 280 | self.encode_valtype(self.resolve, &Type::Id(*id))?; |
| 281 | type_order.insert(*id); |
| 282 | } |
| 283 | |
| 284 | // Sort functions based on whether or not they're associated with |
| 285 | // resources. |
| 286 | // |
| 287 | // This is done here to ensure that when a WIT package is printed as WIT |
| 288 | // then decoded, or if it's printed as Wasm then decoded, the final |
| 289 | // result is the same. When printing via WIT resource methods are |
| 290 | // attached to the resource types themselves meaning that they'll appear |
| 291 | // intermingled with the rest of the types, namely first before all |
| 292 | // other functions. The purpose of this sort is to perform a stable sort |
| 293 | // over all functions by shuffling the resource-related functions first, |
| 294 | // in order of when their associated resource was encoded, and putting |
| 295 | // freestanding functions last. |
| 296 | // |
| 297 | // Note that this is not actually required for correctness, it's |
| 298 | // basically here to make fuzzing happy. |
| 299 | let mut funcs = iface.functions.iter().collect::<Vec<_>>(); |
| 300 | funcs.sort_by_key(|(_name, func)| match func.kind { |
| 301 | FunctionKind::Freestanding => type_order.len(), |
| 302 | FunctionKind::Method(id) | FunctionKind::Constructor(id) | FunctionKind::Static(id) => { |
| 303 | type_order.get_index_of(&id).unwrap() |
| 304 | } |
| 305 | }); |
| 306 | |
| 307 | for (name, func) in funcs { |
| 308 | let ty = self.encode_func_type(self.resolve, func)?; |
| 309 | self.ty |
| 310 | .as_mut() |
| 311 | .unwrap() |
| 312 | .export(name, ComponentTypeRef::Func(ty)); |
| 313 | } |
| 314 | let instance = self.pop_instance(); |
| 315 | let idx = self.outer.type_count(); |
| 316 | self.outer.ty().instance(&instance); |
| 317 | self.import_map.insert(interface, self.instances); |
| 318 | self.instances += 1; |
| 319 | Ok(idx) |
| 320 | } |
| 321 | |
| 322 | fn push_instance(&mut self) { |
| 323 | assert!(self.ty.is_none()); |
| 324 | assert!(self.saved_types.is_none()); |
| 325 | self.saved_types = Some(( |
| 326 | mem::take(&mut self.type_map), |
| 327 | mem::take(&mut self.func_type_map), |
| 328 | )); |
| 329 | self.ty = Some(InstanceType::default()); |
| 330 | } |
| 331 | |
| 332 | fn pop_instance(&mut self) -> InstanceType { |
| 333 | let (types, funcs) = self.saved_types.take().unwrap(); |
| 334 | self.type_map = types; |
| 335 | self.func_type_map = funcs; |
| 336 | mem::take(&mut self.ty).unwrap() |
| 337 | } |
| 338 | } |
| 339 | |
| 340 | impl<'a> ValtypeEncoder<'a> for InterfaceEncoder<'a> { |
| 341 | fn defined_type(&mut self) -> (u32, ComponentDefinedTypeEncoder<'_>) { |
| 342 | match &mut self.ty { |
| 343 | Some(ty) => (ty.type_count(), ty.ty().defined_type()), |
| 344 | None => (self.outer.type_count(), self.outer.ty().defined_type()), |
| 345 | } |
| 346 | } |
| 347 | fn define_function_type(&mut self) -> (u32, ComponentFuncTypeEncoder<'_>) { |
| 348 | match &mut self.ty { |
| 349 | Some(ty) => (ty.type_count(), ty.ty().function()), |
| 350 | None => (self.outer.type_count(), self.outer.ty().function()), |
| 351 | } |
| 352 | } |
| 353 | fn export_type(&mut self, index: u32, name: &'a str) -> Option<u32> { |
| 354 | match &mut self.ty { |
| 355 | Some(ty) => { |
| 356 | assert!(!self.import_types); |
| 357 | let ret = ty.type_count(); |
| 358 | ty.export(name, ComponentTypeRef::Type(TypeBounds::Eq(index))); |
| 359 | Some(ret) |
| 360 | } |
| 361 | None => { |
| 362 | let ret = self.outer.type_count(); |
| 363 | if self.import_types { |
| 364 | self.outer |
| 365 | .import(name, ComponentTypeRef::Type(TypeBounds::Eq(index))); |
| 366 | } else { |
| 367 | self.outer |
| 368 | .export(name, ComponentTypeRef::Type(TypeBounds::Eq(index))); |
| 369 | } |
| 370 | Some(ret) |
| 371 | } |
| 372 | } |
| 373 | } |
| 374 | fn export_resource(&mut self, name: &'a str) -> u32 { |
| 375 | let type_ref = ComponentTypeRef::Type(TypeBounds::SubResource); |
| 376 | match &mut self.ty { |
| 377 | Some(ty) => { |
| 378 | assert!(!self.import_types); |
| 379 | ty.export(name, type_ref); |
| 380 | ty.type_count() - 1 |
| 381 | } |
| 382 | None => { |
| 383 | if self.import_types { |
| 384 | self.outer.import(name, type_ref); |
| 385 | } else { |
| 386 | self.outer.export(name, type_ref); |
| 387 | } |
| 388 | self.outer.type_count() - 1 |
| 389 | } |
| 390 | } |
| 391 | } |
| 392 | fn type_map(&mut self) -> &mut HashMap<TypeId, u32> { |
| 393 | &mut self.type_map |
| 394 | } |
| 395 | fn interface(&self) -> Option<InterfaceId> { |
| 396 | self.interface |
| 397 | } |
| 398 | fn import_type(&mut self, owner: InterfaceId, id: TypeId) -> u32 { |
| 399 | let ty = &self.resolve.types[id]; |
| 400 | let instance = self.import_map[&owner]; |
| 401 | let outer_idx = *self.outer_type_map.entry(id).or_insert_with(|| { |
| 402 | let ret = self.outer.type_count(); |
| 403 | self.outer.alias(Alias::InstanceExport { |
| 404 | instance, |
| 405 | name: ty.name.as_ref().unwrap(), |
| 406 | kind: ComponentExportKind::Type, |
| 407 | }); |
| 408 | ret |
| 409 | }); |
| 410 | match &mut self.ty { |
| 411 | Some(ty) => { |
| 412 | let ret = ty.type_count(); |
| 413 | ty.alias(Alias::Outer { |
| 414 | count: 1, |
| 415 | index: outer_idx, |
| 416 | kind: ComponentOuterAliasKind::Type, |
| 417 | }); |
| 418 | ret |
| 419 | } |
| 420 | None => outer_idx, |
| 421 | } |
| 422 | } |
| 423 | fn func_type_map(&mut self) -> &mut HashMap<FunctionKey<'a>, u32> { |
| 424 | &mut self.func_type_map |
| 425 | } |
| 426 | } |
| 427 | |