diff options
| author | Laurenz <laurmaedje@gmail.com> | 2025-01-30 14:36:15 +0100 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-01-30 13:36:15 +0000 |
| commit | be1fa91a00a9bff6c5eb9744266f252b8cc23fe4 (patch) | |
| tree | 398667e7da4e44d910dbf69f4d6b6d18f9141e0c /crates/typst-library | |
| parent | 7a0d7092bc00ee4f5c0d4887ea3ccf3fbceb2426 (diff) | |
Modular, multi-threaded, transitioning plugins (#5779)
Diffstat (limited to 'crates/typst-library')
| -rw-r--r-- | crates/typst-library/src/foundations/func.rs | 59 | ||||
| -rw-r--r-- | crates/typst-library/src/foundations/mod.rs | 7 | ||||
| -rw-r--r-- | crates/typst-library/src/foundations/module.rs | 20 | ||||
| -rw-r--r-- | crates/typst-library/src/foundations/ops.rs | 1 | ||||
| -rw-r--r-- | crates/typst-library/src/foundations/plugin.rs | 546 | ||||
| -rw-r--r-- | crates/typst-library/src/foundations/scope.rs | 8 | ||||
| -rw-r--r-- | crates/typst-library/src/foundations/value.rs | 11 |
7 files changed, 468 insertions, 184 deletions
diff --git a/crates/typst-library/src/foundations/func.rs b/crates/typst-library/src/foundations/func.rs index cb3eba16..a05deb1f 100644 --- a/crates/typst-library/src/foundations/func.rs +++ b/crates/typst-library/src/foundations/func.rs @@ -9,11 +9,11 @@ use ecow::{eco_format, EcoString}; use typst_syntax::{ast, Span, SyntaxNode}; use typst_utils::{singleton, LazyHash, Static}; -use crate::diag::{bail, SourceResult, StrResult}; +use crate::diag::{bail, At, SourceResult, StrResult}; use crate::engine::Engine; use crate::foundations::{ - cast, repr, scope, ty, Args, CastInfo, Content, Context, Element, IntoArgs, Scope, - Selector, Type, Value, + cast, repr, scope, ty, Args, Bytes, CastInfo, Content, Context, Element, IntoArgs, + PluginFunc, Scope, Selector, Type, Value, }; /// A mapping from argument values to a return value. @@ -151,6 +151,8 @@ enum Repr { Element(Element), /// A user-defined closure. Closure(Arc<LazyHash<Closure>>), + /// A plugin WebAssembly function. + Plugin(Arc<PluginFunc>), /// A nested function with pre-applied arguments. With(Arc<(Func, Args)>), } @@ -164,6 +166,7 @@ impl Func { Repr::Native(native) => Some(native.name), Repr::Element(elem) => Some(elem.name()), Repr::Closure(closure) => closure.name(), + Repr::Plugin(func) => Some(func.name()), Repr::With(with) => with.0.name(), } } @@ -176,6 +179,7 @@ impl Func { Repr::Native(native) => Some(native.title), Repr::Element(elem) => Some(elem.title()), Repr::Closure(_) => None, + Repr::Plugin(_) => None, Repr::With(with) => with.0.title(), } } @@ -186,6 +190,7 @@ impl Func { Repr::Native(native) => Some(native.docs), Repr::Element(elem) => Some(elem.docs()), Repr::Closure(_) => None, + Repr::Plugin(_) => None, Repr::With(with) => with.0.docs(), } } @@ -204,6 +209,7 @@ impl Func { Repr::Native(native) => Some(&native.0.params), Repr::Element(elem) => Some(elem.params()), Repr::Closure(_) => None, + Repr::Plugin(_) => None, Repr::With(with) => with.0.params(), } } @@ -221,6 +227,7 @@ impl Func { Some(singleton!(CastInfo, CastInfo::Type(Type::of::<Content>()))) } Repr::Closure(_) => None, + Repr::Plugin(_) => None, Repr::With(with) => with.0.returns(), } } @@ -231,6 +238,7 @@ impl Func { Repr::Native(native) => native.keywords, Repr::Element(elem) => elem.keywords(), Repr::Closure(_) => &[], + Repr::Plugin(_) => &[], Repr::With(with) => with.0.keywords(), } } @@ -241,6 +249,7 @@ impl Func { Repr::Native(native) => Some(&native.0.scope), Repr::Element(elem) => Some(elem.scope()), Repr::Closure(_) => None, + Repr::Plugin(_) => None, Repr::With(with) => with.0.scope(), } } @@ -266,6 +275,14 @@ impl Func { } } + /// Extract the plugin function, if it is one. + pub fn to_plugin(&self) -> Option<&PluginFunc> { + match &self.repr { + Repr::Plugin(func) => Some(func), + _ => None, + } + } + /// Call the function with the given context and arguments. pub fn call<A: IntoArgs>( &self, @@ -307,6 +324,12 @@ impl Func { context, args, ), + Repr::Plugin(func) => { + let inputs = args.all::<Bytes>()?; + let output = func.call(inputs).at(args.span)?; + args.finish()?; + Ok(Value::Bytes(output)) + } Repr::With(with) => { args.items = with.1.items.iter().cloned().chain(args.items).collect(); with.0.call(engine, context, args) @@ -425,12 +448,30 @@ impl From<Repr> for Func { } } +impl From<&'static NativeFuncData> for Func { + fn from(data: &'static NativeFuncData) -> Self { + Repr::Native(Static(data)).into() + } +} + impl From<Element> for Func { fn from(func: Element) -> Self { Repr::Element(func).into() } } +impl From<Closure> for Func { + fn from(closure: Closure) -> Self { + Repr::Closure(Arc::new(LazyHash::new(closure))).into() + } +} + +impl From<PluginFunc> for Func { + fn from(func: PluginFunc) -> Self { + Repr::Plugin(Arc::new(func)).into() + } +} + /// A Typst function that is defined by a native Rust type that shadows a /// native Rust function. pub trait NativeFunc { @@ -466,12 +507,6 @@ pub struct NativeFuncData { pub returns: LazyLock<CastInfo>, } -impl From<&'static NativeFuncData> for Func { - fn from(data: &'static NativeFuncData) -> Self { - Repr::Native(Static(data)).into() - } -} - cast! { &'static NativeFuncData, self => Func::from(self).into_value(), @@ -525,12 +560,6 @@ impl Closure { } } -impl From<Closure> for Func { - fn from(closure: Closure) -> Self { - Repr::Closure(Arc::new(LazyHash::new(closure))).into() - } -} - cast! { Closure, self => Value::Func(self.into()), diff --git a/crates/typst-library/src/foundations/mod.rs b/crates/typst-library/src/foundations/mod.rs index 2921481b..a790da4f 100644 --- a/crates/typst-library/src/foundations/mod.rs +++ b/crates/typst-library/src/foundations/mod.rs @@ -25,7 +25,8 @@ mod int; mod label; mod module; mod none; -mod plugin; +#[path = "plugin.rs"] +mod plugin_; mod scope; mod selector; mod str; @@ -56,7 +57,7 @@ pub use self::int::*; pub use self::label::*; pub use self::module::*; pub use self::none::*; -pub use self::plugin::*; +pub use self::plugin_::*; pub use self::repr::Repr; pub use self::scope::*; pub use self::selector::*; @@ -114,11 +115,11 @@ pub(super) fn define(global: &mut Scope, inputs: Dict, features: &Features) { global.define_type::<Symbol>(); global.define_type::<Duration>(); global.define_type::<Version>(); - global.define_type::<Plugin>(); global.define_func::<repr::repr>(); global.define_func::<panic>(); global.define_func::<assert>(); global.define_func::<eval>(); + global.define_func::<plugin>(); if features.is_enabled(Feature::Html) { global.define_func::<target>(); } diff --git a/crates/typst-library/src/foundations/module.rs b/crates/typst-library/src/foundations/module.rs index 2001aca1..3ee59c10 100644 --- a/crates/typst-library/src/foundations/module.rs +++ b/crates/typst-library/src/foundations/module.rs @@ -7,14 +7,20 @@ use typst_syntax::FileId; use crate::diag::StrResult; use crate::foundations::{repr, ty, Content, Scope, Value}; -/// An evaluated module, either built-in or resulting from a file. +/// An module of definitions. /// -/// You can access definitions from the module using -/// [field access notation]($scripting/#fields) and interact with it using the -/// [import and include syntaxes]($scripting/#modules). Alternatively, it is -/// possible to convert a module to a dictionary, and therefore access its -/// contents dynamically, using the -/// [dictionary constructor]($dictionary/#constructor). +/// A module +/// - be built-in +/// - stem from a [file import]($scripting/#modules) +/// - stem from a [package import]($scripting/#packages) (and thus indirectly +/// its entrypoint file) +/// - result from a call to the [plugin]($plugin) function +/// +/// You can access definitions from the module using [field access +/// notation]($scripting/#fields) and interact with it using the [import and +/// include syntaxes]($scripting/#modules). Alternatively, it is possible to +/// convert a module to a dictionary, and therefore access its contents +/// dynamically, using the [dictionary constructor]($dictionary/#constructor). /// /// # Example /// ```example diff --git a/crates/typst-library/src/foundations/ops.rs b/crates/typst-library/src/foundations/ops.rs index 7dbdde8f..6c240844 100644 --- a/crates/typst-library/src/foundations/ops.rs +++ b/crates/typst-library/src/foundations/ops.rs @@ -447,7 +447,6 @@ pub fn equal(lhs: &Value, rhs: &Value) -> bool { (Args(a), Args(b)) => a == b, (Type(a), Type(b)) => a == b, (Module(a), Module(b)) => a == b, - (Plugin(a), Plugin(b)) => a == b, (Datetime(a), Datetime(b)) => a == b, (Duration(a), Duration(b)) => a == b, (Dyn(a), Dyn(b)) => a == b, diff --git a/crates/typst-library/src/foundations/plugin.rs b/crates/typst-library/src/foundations/plugin.rs index d41261ed..cbc0f52d 100644 --- a/crates/typst-library/src/foundations/plugin.rs +++ b/crates/typst-library/src/foundations/plugin.rs @@ -4,43 +4,27 @@ use std::sync::{Arc, Mutex}; use ecow::{eco_format, EcoString}; use typst_syntax::Spanned; -use wasmi::{AsContext, AsContextMut}; +use wasmi::Memory; use crate::diag::{bail, At, SourceResult, StrResult}; use crate::engine::Engine; -use crate::foundations::{func, repr, scope, ty, Bytes}; +use crate::foundations::{cast, func, scope, Bytes, Func, Module, Scope, Value}; use crate::loading::{DataSource, Load}; -/// A WebAssembly plugin. +/// Loads a WebAssembly module. /// -/// Typst is capable of interfacing with plugins compiled to WebAssembly. Plugin -/// functions may accept multiple [byte buffers]($bytes) as arguments and return -/// a single byte buffer. They should typically be wrapped in idiomatic Typst -/// functions that perform the necessary conversions between native Typst types -/// and bytes. +/// The resulting [module] will contain one Typst [function] for each function +/// export of the loaded WebAssembly module. /// -/// Plugins run in isolation from your system, which means that printing, -/// reading files, or anything like that will not be supported for security -/// reasons. To run as a plugin, a program needs to be compiled to a 32-bit -/// shared WebAssembly library. Many compilers will use the -/// [WASI ABI](https://wasi.dev/) by default or as their only option (e.g. -/// emscripten), which allows printing, reading files, etc. This ABI will not -/// directly work with Typst. You will either need to compile to a different -/// target or [stub all functions](https://github.com/astrale-sharp/wasm-minimal-protocol/tree/master/crates/wasi-stub). +/// Typst WebAssembly plugins need to follow a specific +/// [protocol]($plugin/#protocol). To run as a plugin, a program needs to be +/// compiled to a 32-bit shared WebAssembly library. Plugin functions may accept +/// multiple [byte buffers]($bytes) as arguments and return a single byte +/// buffer. They should typically be wrapped in idiomatic Typst functions that +/// perform the necessary conversions between native Typst types and bytes. /// -/// # Plugins and Packages -/// Plugins are distributed as packages. A package can make use of a plugin -/// simply by including a WebAssembly file and loading it. Because the -/// byte-based plugin interface is quite low-level, plugins are typically -/// exposed through wrapper functions, that also live in the same package. -/// -/// # Purity -/// Plugin functions must be pure: Given the same arguments, they must always -/// return the same value. The reason for this is that Typst functions must be -/// pure (which is quite fundamental to the language design) and, since Typst -/// function can call plugin functions, this requirement is inherited. In -/// particular, if a plugin function is called twice with the same arguments, -/// Typst might cache the results and call your function only once. +/// For security reasons, plugins run in isolation from your system. This means +/// that printing, reading files, or similar things are not supported. /// /// # Example /// ```example @@ -55,6 +39,50 @@ use crate::loading::{DataSource, Load}; /// #concat("hello", "world") /// ``` /// +/// Since the plugin function returns a module, it can be used with import +/// syntax: +/// ```typ +/// #import plugin("hello.wasm"): concatenate +/// ``` +/// +/// # Purity +/// Plugin functions **must be pure:** A plugin function call most not have any +/// observable side effects on future plugin calls and given the same arguments, +/// it must always return the same value. +/// +/// The reason for this is that Typst functions must be pure (which is quite +/// fundamental to the language design) and, since Typst function can call +/// plugin functions, this requirement is inherited. In particular, if a plugin +/// function is called twice with the same arguments, Typst might cache the +/// results and call your function only once. Moreover, Typst may run multiple +/// instances of your plugin in multiple threads, with no state shared between +/// them. +/// +/// Typst does not enforce plugin function purity (for efficiency reasons), but +/// calling an impure function will lead to unpredictable and irreproducible +/// results and must be avoided. +/// +/// That said, mutable operations _can be_ useful for plugins that require +/// costly runtime initialization. Due to the purity requirement, such +/// initialization cannot be performed through a normal function call. Instead, +/// Typst exposes a [plugin transition API]($plugin.transition), which executes +/// a function call and then creates a derived module with new functions which +/// will observe the side effects produced by the transition call. The original +/// plugin remains unaffected. +/// +/// # Plugins and Packages +/// Any Typst code can make use of a plugin simply by including a WebAssembly +/// file and loading it. However, because the byte-based plugin interface is +/// quite low-level, plugins are typically exposed through a package containing +/// the plugin and idiomatic wrapper functions. +/// +/// # WASI +/// Many compilers will use the [WASI ABI](https://wasi.dev/) by default or as +/// their only option (e.g. emscripten), which allows printing, reading files, +/// etc. This ABI will not directly work with Typst. You will either need to +/// compile to a different target or [stub all +/// functions](https://github.com/astrale-sharp/wasm-minimal-protocol/tree/master/crates/wasi-stub). +/// /// # Protocol /// To be used as a plugin, a WebAssembly module must conform to the following /// protocol: @@ -67,8 +95,8 @@ use crate::loading::{DataSource, Load}; /// lengths, so `usize/size_t` may be preferable), and return one 32-bit /// integer. /// -/// - The function should first allocate a buffer `buf` of length -/// `a_1 + a_2 + ... + a_n`, and then call +/// - The function should first allocate a buffer `buf` of length `a_1 + a_2 + +/// ... + a_n`, and then call /// `wasm_minimal_protocol_write_args_to_buffer(buf.ptr)`. /// /// - The `a_1` first bytes of the buffer now constitute the first argument, the @@ -85,19 +113,21 @@ use crate::loading::{DataSource, Load}; /// then interpreted as an UTF-8 encoded error message. /// /// ## Imports -/// Plugin modules need to import two functions that are provided by the runtime. -/// (Types and functions are described using WAT syntax.) +/// Plugin modules need to import two functions that are provided by the +/// runtime. (Types and functions are described using WAT syntax.) /// -/// - `(import "typst_env" "wasm_minimal_protocol_write_args_to_buffer" (func (param i32)))` +/// - `(import "typst_env" "wasm_minimal_protocol_write_args_to_buffer" (func +/// (param i32)))` /// /// Writes the arguments for the current function into a plugin-allocated -/// buffer. When a plugin function is called, it -/// [receives the lengths](#exports) of its input buffers as arguments. It -/// should then allocate a buffer whose capacity is at least the sum of these -/// lengths. It should then call this function with a `ptr` to the buffer to -/// fill it with the arguments, one after another. +/// buffer. When a plugin function is called, it [receives the +/// lengths](#exports) of its input buffers as arguments. It should then +/// allocate a buffer whose capacity is at least the sum of these lengths. It +/// should then call this function with a `ptr` to the buffer to fill it with +/// the arguments, one after another. /// -/// - `(import "typst_env" "wasm_minimal_protocol_send_result_to_host" (func (param i32 i32)))` +/// - `(import "typst_env" "wasm_minimal_protocol_send_result_to_host" (func +/// (param i32 i32)))` /// /// Sends the output of the current function to the host (Typst). The first /// parameter shall be a pointer to a buffer (`ptr`), while the second is the @@ -106,72 +136,147 @@ use crate::loading::{DataSource, Load}; /// interpreted as an error message, it should be encoded as UTF-8. /// /// # Resources -/// For more resources, check out the -/// [wasm-minimal-protocol repository](https://github.com/astrale-sharp/wasm-minimal-protocol). -/// It contains: +/// For more resources, check out the [wasm-minimal-protocol +/// repository](https://github.com/astrale-sharp/wasm-minimal-protocol). It +/// contains: /// /// - A list of example plugin implementations and a test runner for these /// examples /// - Wrappers to help you write your plugin in Rust (Zig wrapper in /// development) /// - A stubber for WASI -#[ty(scope, cast)] -#[derive(Clone)] -pub struct Plugin(Arc<Repr>); +#[func(scope)] +pub fn plugin( + engine: &mut Engine, + /// A path to a WebAssembly file or raw WebAssembly bytes. + /// + /// For more details about paths, see the [Paths section]($syntax/#paths). + source: Spanned<DataSource>, +) -> SourceResult<Module> { + let data = source.load(engine.world)?; + Plugin::module(data).at(source.span) +} -/// The internal representation of a plugin. -struct Repr { - /// The raw WebAssembly bytes. - bytes: Bytes, - /// The function defined by the WebAssembly module. - functions: Vec<(EcoString, wasmi::Func)>, - /// Owns all data associated with the WebAssembly module. - store: Mutex<Store>, +#[scope] +impl plugin { + /// Calls a plugin function that has side effects and returns a new module + /// with plugin functions that are guaranteed to have observed the results + /// of the mutable call. + /// + /// Note that calling an impure function through a normal function call + /// (without use of the transition API) is forbidden and leads to + /// unpredictable behaviour. Read the [section on purity]($plugin/#purity) + /// for more details. + /// + /// In the example below, we load the plugin `hello-mut.wasm` which exports + /// two functions: The `get()` function retrieves a global array as a + /// string. The `add(value)` function adds a value to the global array. + /// + /// We call `add` via the transition API. The call `mutated.get()` on the + /// derived module will observe the addition. Meanwhile the original module + /// remains untouched as demonstrated by the `base.get()` call. + /// + /// _Note:_ Due to limitations in the internal WebAssembly implementation, + /// the transition API can only guarantee to reflect changes in the plugin's + /// memory, not in WebAssembly globals. If your plugin relies on changes to + /// globals being visible after transition, you might want to avoid use of + /// the transition API for now. We hope to lift this limitation in the + /// future. + /// + /// ```typ + /// #let base = plugin("hello-mut.wasm") + /// #assert.eq(base.get(), "[]") + /// + /// #let mutated = plugin.transition(base.add, "hello") + /// #assert.eq(base.get(), "[]") + /// #assert.eq(mutated.get(), "[hello]") + /// ``` + #[func] + pub fn transition( + /// The plugin function to call. + func: PluginFunc, + /// The byte buffers to call the function with. + #[variadic] + arguments: Vec<Bytes>, + ) -> StrResult<Module> { + func.transition(arguments) + } } -/// Owns all data associated with the WebAssembly module. -type Store = wasmi::Store<StoreData>; +/// A function loaded from a WebAssembly plugin. +#[derive(Debug, Clone, PartialEq, Hash)] +pub struct PluginFunc { + /// The underlying plugin, shared by this and the other functions. + plugin: Arc<Plugin>, + /// The name of the plugin function. + name: EcoString, +} -/// If there was an error reading/writing memory, keep the offset + length to -/// display an error message. -struct MemoryError { - offset: u32, - length: u32, - write: bool, +impl PluginFunc { + /// The name of the plugin function. + pub fn name(&self) -> &str { + &self.name + } + + /// Call the WebAssembly function with the given arguments. + #[comemo::memoize] + #[typst_macros::time(name = "call plugin")] + pub fn call(&self, args: Vec<Bytes>) -> StrResult<Bytes> { + self.plugin.call(&self.name, args) + } + + /// Transition a plugin and turn the result into a module. + #[comemo::memoize] + #[typst_macros::time(name = "transition plugin")] + pub fn transition(&self, args: Vec<Bytes>) -> StrResult<Module> { + self.plugin.transition(&self.name, args).map(Plugin::into_module) + } } -/// The persistent store data used for communication between store and host. -#[derive(Default)] -struct StoreData { - args: Vec<Bytes>, - output: Vec<u8>, - memory_error: Option<MemoryError>, + +cast! { + PluginFunc, + self => Value::Func(self.into()), + v: Func => v.to_plugin().ok_or("expected plugin function")?.clone(), } -#[scope] -impl Plugin { - /// Creates a new plugin from a WebAssembly file. - #[func(constructor)] - pub fn construct( - engine: &mut Engine, - /// A path to a WebAssembly file or raw WebAssembly bytes. - /// - /// For more details about paths, see the [Paths section]($syntax/#paths). - source: Spanned<DataSource>, - ) -> SourceResult<Plugin> { - let data = source.load(engine.world)?; - Plugin::new(data).at(source.span) - } +/// A plugin with potentially multiple instances for multi-threaded +/// execution. +struct Plugin { + /// Shared by all variants of the plugin. + base: Arc<PluginBase>, + /// A pool of plugin instances. + /// + /// When multiple plugin calls run concurrently due to multi-threading, we + /// create new instances whenever we run out of ones. + pool: Mutex<Vec<PluginInstance>>, + /// A snapshot that new instances should be restored to. + snapshot: Option<Snapshot>, + /// A combined hash that incorporates all function names and arguments used + /// in transitions of this plugin, such that this plugin has a deterministic + /// hash and equality check that can differentiate it from "siblings" (same + /// base, different transitions). + fingerprint: u128, } impl Plugin { - /// Create a new plugin from raw WebAssembly bytes. + /// Create a plugin and turn it into a module. #[comemo::memoize] #[typst_macros::time(name = "load plugin")] - pub fn new(bytes: Bytes) -> StrResult<Plugin> { + fn module(bytes: Bytes) -> StrResult<Module> { + Self::new(bytes).map(Self::into_module) + } + + /// Create a new plugin from raw WebAssembly bytes. + fn new(bytes: Bytes) -> StrResult<Self> { let engine = wasmi::Engine::default(); let module = wasmi::Module::new(&engine, bytes.as_slice()) .map_err(|err| format!("failed to load WebAssembly module ({err})"))?; + // Ensure that the plugin exports its memory. + if !matches!(module.get_export("memory"), Some(wasmi::ExternType::Memory(_))) { + bail!("plugin does not export its memory"); + } + let mut linker = wasmi::Linker::new(&engine); linker .func_wrap( @@ -188,58 +293,174 @@ impl Plugin { ) .unwrap(); - let mut store = Store::new(&engine, StoreData::default()); - let instance = linker - .instantiate(&mut store, &module) - .and_then(|pre_instance| pre_instance.start(&mut store)) - .map_err(|e| eco_format!("{e}"))?; + let base = Arc::new(PluginBase { bytes, linker, module }); + let instance = PluginInstance::new(&base, None)?; - // Ensure that the plugin exports its memory. - if !matches!( - instance.get_export(&store, "memory"), - Some(wasmi::Extern::Memory(_)) - ) { - bail!("plugin does not export its memory"); + Ok(Self { + base, + snapshot: None, + fingerprint: 0, + pool: Mutex::new(vec![instance]), + }) + } + + /// Execute a function with access to an instsance. + fn call(&self, func: &str, args: Vec<Bytes>) -> StrResult<Bytes> { + // Acquire an instance from the pool (potentially creating a new one). + let mut instance = self.acquire()?; + + // Execute the call on an instance from the pool. If the call fails, we + // return early and _don't_ return the instance to the pool as it might + // be irrecoverably damaged. + let output = instance.call(func, args)?; + + // Return the instance to the pool. + self.pool.lock().unwrap().push(instance); + + Ok(output) + } + + /// Call a mutable plugin function, producing a new mutable whose functions + /// are guaranteed to be able to observe the mutation. + fn transition(&self, func: &str, args: Vec<Bytes>) -> StrResult<Plugin> { + // Derive a new transition hash from the old one and the function and arguments. + let fingerprint = typst_utils::hash128(&(self.fingerprint, func, &args)); + + // Execute the mutable call on an instance. + let mut instance = self.acquire()?; + + // Call the function. If the call fails, we return early and _don't_ + // return the instance to the pool as it might be irrecoverably damaged. + instance.call(func, args)?; + + // Snapshot the instance after the mutable call. + let snapshot = instance.snapshot(); + + // Create a new plugin and move (this is important!) the used instance + // into it, so that the old plugin won't observe the mutation. Also + // save the snapshot so that instances that are initialized for the + // transitioned plugin's pool observe the mutation. + Ok(Self { + base: self.base.clone(), + snapshot: Some(snapshot), + fingerprint, + pool: Mutex::new(vec![instance]), + }) + } + + /// Acquire an instance from the pool (or create a new one). + fn acquire(&self) -> StrResult<PluginInstance> { + // Don't use match to ensure that the lock is released before we create + // a new instance. + if let Some(instance) = self.pool.lock().unwrap().pop() { + return Ok(instance); } - // Collect exported functions. - let functions = instance - .exports(&store) - .filter_map(|export| { - let name = export.name().into(); - export.into_func().map(|func| (name, func)) - }) - .collect(); + PluginInstance::new(&self.base, self.snapshot.as_ref()) + } - Ok(Plugin(Arc::new(Repr { bytes, functions, store: Mutex::new(store) }))) + /// Turn a plugin into a Typst module containing plugin functions. + fn into_module(self) -> Module { + let shared = Arc::new(self); + + // Build a scope from the collected functions. + let mut scope = Scope::new(); + for export in shared.base.module.exports() { + if matches!(export.ty(), wasmi::ExternType::Func(_)) { + let name = EcoString::from(export.name()); + let func = PluginFunc { plugin: shared.clone(), name: name.clone() }; + scope.define(name, Func::from(func)); + } + } + + Module::anonymous(scope) } +} - /// Call the plugin function with the given `name`. - #[comemo::memoize] - #[typst_macros::time(name = "call plugin")] - pub fn call(&self, name: &str, args: Vec<Bytes>) -> StrResult<Bytes> { - // Find the function with the given name. - let func = self - .0 - .functions - .iter() - .find(|(v, _)| v == name) - .map(|&(_, func)| func) - .ok_or_else(|| { - eco_format!("plugin does not contain a function called {name}") - })?; +impl Debug for Plugin { + fn fmt(&self, f: &mut Formatter) -> fmt::Result { + f.pad("Plugin(..)") + } +} - let mut store = self.0.store.lock().unwrap(); - let ty = func.ty(store.as_context()); +impl PartialEq for Plugin { + fn eq(&self, other: &Self) -> bool { + self.base.bytes == other.base.bytes && self.fingerprint == other.fingerprint + } +} + +impl Hash for Plugin { + fn hash<H: Hasher>(&self, state: &mut H) { + self.base.bytes.hash(state); + self.fingerprint.hash(state); + } +} + +/// Shared by all pooled & transitioned variants of the plugin. +struct PluginBase { + /// The raw WebAssembly bytes. + bytes: Bytes, + /// The compiled WebAssembly module. + module: wasmi::Module, + /// A linker used to create a `Store` for execution. + linker: wasmi::Linker<CallData>, +} + +/// An single plugin instance for single-threaded execution. +struct PluginInstance { + /// The underlying wasmi instance. + instance: wasmi::Instance, + /// The execution store of this concrete plugin instance. + store: wasmi::Store<CallData>, +} + +/// A snapshot of a plugin instance. +struct Snapshot { + /// The number of pages in the main memory. + mem_pages: u32, + /// The data in the main memory. + mem_data: Vec<u8>, +} - // Check function signature. +impl PluginInstance { + /// Create a new execution instance of a plugin, potentially restoring + /// a snapshot. + #[typst_macros::time(name = "create plugin instance")] + fn new(base: &PluginBase, snapshot: Option<&Snapshot>) -> StrResult<PluginInstance> { + let mut store = wasmi::Store::new(base.linker.engine(), CallData::default()); + let instance = base + .linker + .instantiate(&mut store, &base.module) + .and_then(|pre_instance| pre_instance.start(&mut store)) + .map_err(|e| eco_format!("{e}"))?; + + let mut instance = PluginInstance { instance, store }; + if let Some(snapshot) = snapshot { + instance.restore(snapshot); + } + Ok(instance) + } + + /// Call a plugin function with byte arguments. + fn call(&mut self, func: &str, args: Vec<Bytes>) -> StrResult<Bytes> { + let handle = self + .instance + .get_export(&self.store, func) + .unwrap() + .into_func() + .unwrap(); + let ty = handle.ty(&self.store); + + // Check function signature. Do this lazily only when a function is called + // because there might be exported functions like `_initialize` that don't + // match the schema. if ty.params().iter().any(|&v| v != wasmi::core::ValType::I32) { bail!( - "plugin function `{name}` has a parameter that is not a 32-bit integer" + "plugin function `{func}` has a parameter that is not a 32-bit integer" ); } if ty.results() != [wasmi::core::ValType::I32] { - bail!("plugin function `{name}` does not return exactly one 32-bit integer"); + bail!("plugin function `{func}` does not return exactly one 32-bit integer"); } // Check inputs. @@ -260,23 +481,26 @@ impl Plugin { .collect::<Vec<_>>(); // Store the input data. - store.data_mut().args = args; + self.store.data_mut().args = args; // Call the function. let mut code = wasmi::Val::I32(-1); - func.call(store.as_context_mut(), &lengths, std::slice::from_mut(&mut code)) + handle + .call(&mut self.store, &lengths, std::slice::from_mut(&mut code)) .map_err(|err| eco_format!("plugin panicked: {err}"))?; + if let Some(MemoryError { offset, length, write }) = - store.data_mut().memory_error.take() + self.store.data_mut().memory_error.take() { return Err(eco_format!( - "plugin tried to {kind} out of bounds: pointer {offset:#x} is out of bounds for {kind} of length {length}", + "plugin tried to {kind} out of bounds: \ + pointer {offset:#x} is out of bounds for {kind} of length {length}", kind = if write { "write" } else { "read" } )); } // Extract the returned data. - let output = std::mem::take(&mut store.data_mut().output); + let output = std::mem::take(&mut self.store.data_mut().output); // Parse the functions return value. match code { @@ -293,39 +517,63 @@ impl Plugin { Ok(Bytes::new(output)) } - /// An iterator over all the function names defined by the plugin. - pub fn iter(&self) -> impl Iterator<Item = &EcoString> { - self.0.functions.as_slice().iter().map(|(func_name, _)| func_name) + /// Creates a snapshot of this instance from which another one can be + /// initialized. + #[typst_macros::time(name = "save snapshot")] + fn snapshot(&self) -> Snapshot { + let memory = self.memory(); + let mem_pages = memory.size(&self.store); + let mem_data = memory.data(&self.store).to_vec(); + Snapshot { mem_pages, mem_data } } -} -impl Debug for Plugin { - fn fmt(&self, f: &mut Formatter) -> fmt::Result { - f.pad("Plugin(..)") + /// Restores the instance to a snapshot. + #[typst_macros::time(name = "restore snapshot")] + fn restore(&mut self, snapshot: &Snapshot) { + let memory = self.memory(); + let current_size = memory.size(&self.store); + if current_size < snapshot.mem_pages { + memory + .grow(&mut self.store, snapshot.mem_pages - current_size) + .unwrap(); + } + + memory.data_mut(&mut self.store)[..snapshot.mem_data.len()] + .copy_from_slice(&snapshot.mem_data); } -} -impl repr::Repr for Plugin { - fn repr(&self) -> EcoString { - "plugin(..)".into() + /// Retrieves a handle to the plugin's main memory. + fn memory(&self) -> Memory { + self.instance + .get_export(&self.store, "memory") + .unwrap() + .into_memory() + .unwrap() } } -impl PartialEq for Plugin { - fn eq(&self, other: &Self) -> bool { - self.0.bytes == other.0.bytes - } +/// The persistent store data used for communication between store and host. +#[derive(Default)] +struct CallData { + /// Arguments for a current call. + args: Vec<Bytes>, + /// The results of the current call. + output: Vec<u8>, + /// A memory error that occured during execution of the current call. + memory_error: Option<MemoryError>, } -impl Hash for Plugin { - fn hash<H: Hasher>(&self, state: &mut H) { - self.0.bytes.hash(state); - } +/// If there was an error reading/writing memory, keep the offset + length to +/// display an error message. +struct MemoryError { + offset: u32, + length: u32, + write: bool, } /// Write the arguments to the plugin function into the plugin's memory. fn wasm_minimal_protocol_write_args_to_buffer( - mut caller: wasmi::Caller<StoreData>, + mut caller: wasmi::Caller<CallData>, ptr: u32, ) { let memory = caller.get_export("memory").unwrap().into_memory().unwrap(); @@ -346,7 +594,7 @@ fn wasm_minimal_protocol_write_args_to_buffer( /// Extracts the output of the plugin function from the plugin's memory. fn wasm_minimal_protocol_send_result_to_host( - mut caller: wasmi::Caller<StoreData>, + mut caller: wasmi::Caller<CallData>, ptr: u32, len: u32, ) { diff --git a/crates/typst-library/src/foundations/scope.rs b/crates/typst-library/src/foundations/scope.rs index 99c9a37e..b7b4a6d9 100644 --- a/crates/typst-library/src/foundations/scope.rs +++ b/crates/typst-library/src/foundations/scope.rs @@ -167,6 +167,14 @@ impl Scope { Default::default() } + /// Create a new scope with the given capacity. + pub fn with_capacity(capacity: usize) -> Self { + Self { + map: IndexMap::with_capacity(capacity), + ..Default::default() + } + } + /// Create a new scope with duplication prevention. pub fn deduplicating() -> Self { Self { deduplicate: true, ..Default::default() } diff --git a/crates/typst-library/src/foundations/value.rs b/crates/typst-library/src/foundations/value.rs index d9902772..4fa380b4 100644 --- a/crates/typst-library/src/foundations/value.rs +++ b/crates/typst-library/src/foundations/value.rs @@ -15,8 +15,8 @@ use crate::diag::{HintedStrResult, HintedString, StrResult}; use crate::foundations::{ fields, ops, repr, Args, Array, AutoValue, Bytes, CastInfo, Content, Datetime, Decimal, Dict, Duration, Fold, FromValue, Func, IntoValue, Label, Module, - NativeElement, NativeType, NoneValue, Plugin, Reflect, Repr, Resolve, Scope, Str, - Styles, Symbol, SymbolElem, Type, Version, + NativeElement, NativeType, NoneValue, Reflect, Repr, Resolve, Scope, Str, Styles, + Symbol, SymbolElem, Type, Version, }; use crate::layout::{Abs, Angle, Em, Fr, Length, Ratio, Rel}; use crate::text::{RawContent, RawElem, TextElem}; @@ -84,8 +84,6 @@ pub enum Value { Type(Type), /// A module. Module(Module), - /// A WebAssembly plugin. - Plugin(Plugin), /// A dynamic value. Dyn(Dynamic), } @@ -147,7 +145,6 @@ impl Value { Self::Args(_) => Type::of::<Args>(), Self::Type(_) => Type::of::<Type>(), Self::Module(_) => Type::of::<Module>(), - Self::Plugin(_) => Type::of::<Plugin>(), Self::Dyn(v) => v.ty(), } } @@ -251,7 +248,6 @@ impl Debug for Value { Self::Args(v) => Debug::fmt(v, f), Self::Type(v) => Debug::fmt(v, f), Self::Module(v) => Debug::fmt(v, f), - Self::Plugin(v) => Debug::fmt(v, f), Self::Dyn(v) => Debug::fmt(v, f), } } @@ -289,7 +285,6 @@ impl Repr for Value { Self::Args(v) => v.repr(), Self::Type(v) => v.repr(), Self::Module(v) => v.repr(), - Self::Plugin(v) => v.repr(), Self::Dyn(v) => v.repr(), } } @@ -340,7 +335,6 @@ impl Hash for Value { Self::Args(v) => v.hash(state), Self::Type(v) => v.hash(state), Self::Module(v) => v.hash(state), - Self::Plugin(v) => v.hash(state), Self::Dyn(v) => v.hash(state), } } @@ -661,7 +655,6 @@ primitive! { primitive! { Args: "arguments", Args } primitive! { Type: "type", Type } primitive! { Module: "module", Module } -primitive! { Plugin: "plugin", Plugin } impl<T: Reflect> Reflect for Arc<T> { fn input() -> CastInfo { |
