Add cursor rule documenting zig javascriptcore bindings

2026-02-02 15:08:46 +00:00 · 2025-03-24 03:10:15 -07:00
parent b37054697b
commit 46c43d954c
1 changed files with 488 additions and 0 deletions
--- a/.cursor/rules/zig-javascriptcore-classes.mdc
+++ b/.cursor/rules/zig-javascriptcore-classes.mdc
@@ -0,0 +1,488 @@
+---
+description: How Zig works with JavaScriptCore bindings generator
+globs: 
+alwaysApply: false
+---
+# Bun's JavaScriptCore Class Bindings Generator
+
+This document explains how Bun's class bindings generator works to bridge Zig and JavaScript code through JavaScriptCore (JSC).
+
+## Architecture Overview
+
+Bun's binding system creates a seamless bridge between JavaScript and Zig, allowing Zig implementations to be exposed as JavaScript classes. The system has several key components:
+
+1. **Zig Implementation** (.zig files)
+2. **JavaScript Interface Definition** (.classes.ts files)
+3. **Generated Code** (C++/Zig files that connect everything)
+
+## Class Definition Files
+
+### JavaScript Interface (.classes.ts)
+
+The `.classes.ts` files define the JavaScript API using a declarative approach:
+
+```typescript
+// Example: encoding.classes.ts
+define({
+  name: "TextDecoder", 
+  constructor: true,
+  JSType: "object",
+  finalize: true,
+  proto: {
+    decode: {
+      // Function definition
+      args: 1,
+    },
+    encoding: {
+      // Getter with caching
+      getter: true,
+      cache: true,
+    },
+    fatal: {
+      // Read-only property
+      getter: true, 
+    },
+    ignoreBOM: {
+      // Read-only property
+      getter: true,
+    }
+  }
+});
+```
+
+Each class definition specifies:
+- The class name
+- Whether it has a constructor
+- JavaScript type (object, function, etc.)
+- Properties and methods in the `proto` field
+- Caching strategy for properties
+- Finalization requirements
+
+### Zig Implementation (.zig)
+
+The Zig files implement the native functionality:
+
+```zig
+// Example: TextDecoder.zig
+pub const TextDecoder = struct {
+    // Internal state
+    encoding: []const u8,
+    fatal: bool,
+    ignoreBOM: bool,
+    
+    // Use generated bindings
+    pub usingnamespace JSC.Codegen.JSTextDecoder;
+    pub usingnamespace bun.New(@This());
+    
+    // Constructor implementation - note use of globalObject
+    pub fn constructor(
+        globalObject: *JSGlobalObject,
+        callFrame: *JSC.CallFrame,
+    ) bun.JSError!*TextDecoder {
+        // Implementation
+    }
+    
+    // Prototype methods - note return type includes JSError
+    pub fn decode(
+        this: *TextDecoder,
+        globalObject: *JSGlobalObject,
+        callFrame: *JSC.CallFrame,
+    ) bun.JSError!JSC.JSValue {
+        // Implementation
+    }
+    
+    // Getters
+    pub fn getEncoding(this: *TextDecoder, globalObject: *JSGlobalObject) JSC.JSValue {
+        return JSC.JSValue.createStringFromUTF8(globalObject, this.encoding);
+    }
+    
+    pub fn getFatal(this: *TextDecoder, globalObject: *JSGlobalObject) JSC.JSValue {
+        return JSC.JSValue.jsBoolean(this.fatal);
+    }
+    
+    // Cleanup - note standard pattern of using deinit/deref
+    pub fn deinit(this: *TextDecoder) void {
+        // Release any retained resources
+    }
+    
+    pub fn finalize(this: *TextDecoder) void {
+        this.deinit();
+        // Or sometimes this is used to free memory instead
+        bun.default_allocator.destroy(this);
+    }
+};
+```
+
+Key components in the Zig file:
+- The struct containing native state
+- `usingnamespace JSC.Codegen.JS<ClassName>` to include generated code
+- `usingnamespace bun.New(@This())` for object creation helpers
+- Constructor and methods using `bun.JSError!JSValue` return type for proper error handling
+- Consistent use of `globalObject` parameter name instead of `ctx`
+- Methods matching the JavaScript interface
+- Getters/setters for properties
+- Proper resource cleanup pattern with `deinit()` and `finalize()`
+
+## Code Generation System
+
+The binding generator produces C++ code that connects JavaScript and Zig:
+
+1. **JSC Class Structure**: Creates C++ classes for the JS object, prototype, and constructor
+2. **Memory Management**: Handles GC integration through JSC's WriteBarrier
+3. **Method Binding**: Connects JS function calls to Zig implementations
+4. **Type Conversion**: Converts between JS values and Zig types
+5. **Property Caching**: Implements the caching system for properties
+
+The generated C++ code includes:
+- A JSC wrapper class (`JSTextDecoder`)
+- A prototype class (`JSTextDecoderPrototype`)
+- A constructor function (`JSTextDecoderConstructor`)
+- Function bindings (`TextDecoderPrototype__decodeCallback`)
+- Property getters/setters (`TextDecoderPrototype__encodingGetterWrap`)
+
+## CallFrame Access
+
+The `CallFrame` object provides access to JavaScript execution context:
+
+```zig
+pub fn decode(
+    this: *TextDecoder, 
+    globalObject: *JSGlobalObject,
+    callFrame: *JSC.CallFrame
+) bun.JSError!JSC.JSValue {
+    // Get arguments
+    const input = callFrame.argument(0);
+    const options = callFrame.argument(1);
+    
+    // Get this value
+    const thisValue = callFrame.thisValue();
+    
+    // Implementation with error handling
+    if (input.isUndefinedOrNull()) {
+        return globalObject.throw("Input cannot be null or undefined", .{});
+    }
+    
+    // Return value or throw error
+    return JSC.JSValue.jsString(globalObject, "result");
+}
+```
+
+CallFrame methods include:
+- `argument(i)`: Get the i-th argument
+- `argumentCount()`: Get the number of arguments
+- `thisValue()`: Get the `this` value
+- `callee()`: Get the function being called
+
+## Property Caching and GC-Owned Values
+
+The `cache: true` option in property definitions enables JSC's WriteBarrier to efficiently store values:
+
+```typescript
+encoding: {
+  getter: true,
+  cache: true, // Enable caching
+}
+```
+
+### C++ Implementation
+
+In the generated C++ code, caching uses JSC's WriteBarrier:
+
+```cpp
+JSC_DEFINE_CUSTOM_GETTER(TextDecoderPrototype__encodingGetterWrap, (...)) {
+    auto& vm = JSC::getVM(lexicalGlobalObject);
+    Zig::GlobalObject *globalObject = reinterpret_cast<Zig::GlobalObject*>(lexicalGlobalObject);
+    auto throwScope = DECLARE_THROW_SCOPE(vm);
+    JSTextDecoder* thisObject = jsCast<JSTextDecoder*>(JSValue::decode(encodedThisValue));
+    JSC::EnsureStillAliveScope thisArg = JSC::EnsureStillAliveScope(thisObject);
+    
+    // Check for cached value and return if present
+    if (JSValue cachedValue = thisObject->m_encoding.get())
+        return JSValue::encode(cachedValue);
+    
+    // Get value from Zig implementation
+    JSC::JSValue result = JSC::JSValue::decode(
+        TextDecoderPrototype__getEncoding(thisObject->wrapped(), globalObject)
+    );
+    RETURN_IF_EXCEPTION(throwScope, {});
+    
+    // Store in cache for future access
+    thisObject->m_encoding.set(vm, thisObject, result);
+    RELEASE_AND_RETURN(throwScope, JSValue::encode(result));
+}
+```
+
+### Zig Accessor Functions
+
+For each cached property, the generator creates Zig accessor functions that allow Zig code to work with these GC-owned values:
+
+```zig
+// External function declarations
+extern fn TextDecoderPrototype__encodingSetCachedValue(JSC.JSValue, *JSC.JSGlobalObject, JSC.JSValue) callconv(JSC.conv) void;
+extern fn TextDecoderPrototype__encodingGetCachedValue(JSC.JSValue) callconv(JSC.conv) JSC.JSValue;
+
+/// `TextDecoder.encoding` setter
+/// This value will be visited by the garbage collector.
+pub fn encodingSetCached(thisValue: JSC.JSValue, globalObject: *JSC.JSGlobalObject, value: JSC.JSValue) void {
+    JSC.markBinding(@src());
+    TextDecoderPrototype__encodingSetCachedValue(thisValue, globalObject, value);
+}
+
+/// `TextDecoder.encoding` getter
+/// This value will be visited by the garbage collector.
+pub fn encodingGetCached(thisValue: JSC.JSValue) ?JSC.JSValue {
+    JSC.markBinding(@src());
+    const result = TextDecoderPrototype__encodingGetCachedValue(thisValue);
+    if (result == .zero)
+        return null;
+
+    return result;
+}
+```
+
+### Benefits of GC-Owned Values
+
+This system provides several key benefits:
+
+1. **Automatic Memory Management**: The JavaScriptCore GC tracks and manages these values
+2. **Proper Garbage Collection**: The WriteBarrier ensures values are properly visited during GC
+3. **Consistent Access**: Zig code can easily get/set these cached JS values
+4. **Performance**: Cached values avoid repeated computation or serialization  
+
+### Use Cases
+
+GC-owned cached values are particularly useful for:
+
+1. **Computed Properties**: Store expensive computation results
+2. **Lazily Created Objects**: Create objects only when needed, then cache them
+3. **References to Other Objects**: Store references to other JS objects that need GC tracking
+4. **Memoization**: Cache results based on input parameters
+
+The WriteBarrier mechanism ensures that any JS values stored in this way are properly tracked by the garbage collector.
+
+## Memory Management and Finalization
+
+The binding system handles memory management across the JavaScript/Zig boundary:
+
+1. **Object Creation**: JavaScript `new TextDecoder()` creates both a JS wrapper and a Zig struct
+2. **Reference Tracking**: JSC's GC tracks all JS references to the object
+3. **Finalization**: When the JS object is collected, the finalizer releases Zig resources
+
+Bun uses a consistent pattern for resource cleanup:
+
+```zig
+// Resource cleanup method - separate from finalization
+pub fn deinit(this: *TextDecoder) void {
+    // Release resources like strings
+    this._encoding.deref(); // String deref pattern
+    
+    // Free any buffers
+    if (this.buffer) |buffer| {
+        bun.default_allocator.free(buffer);
+    }
+}
+
+// Called by the GC when object is collected
+pub fn finalize(this: *TextDecoder) void {
+    JSC.markBinding(@src()); // For debugging
+    this.deinit(); // Clean up resources
+    bun.default_allocator.destroy(this); // Free the object itself
+}
+```
+
+Some objects that hold references to other JS objects use `.deref()` instead:
+
+```zig
+pub fn finalize(this: *SocketAddress) void {
+    JSC.markBinding(@src());
+    this._presentation.deref(); // Release references
+    this.destroy();
+}
+```
+
+## Error Handling with JSError
+
+Bun uses `bun.JSError!JSValue` return type for proper error handling:
+
+```zig
+pub fn decode(
+    this: *TextDecoder, 
+    globalObject: *JSGlobalObject,
+    callFrame: *JSC.CallFrame
+) bun.JSError!JSC.JSValue {
+    // Throwing an error
+    if (callFrame.argumentCount() < 1) {
+        return globalObject.throw("Missing required argument", .{});
+    }
+    
+    // Or returning a success value
+    return JSC.JSValue.jsString(globalObject, "Success!");
+}
+```
+
+This pattern allows Zig functions to:
+1. Return JavaScript values on success
+2. Throw JavaScript exceptions on error
+3. Propagate errors automatically through the call stack
+
+## Type Safety and Error Handling
+
+The binding system includes robust error handling:
+
+```cpp
+// Example of type checking in generated code
+JSTextDecoder* thisObject = jsDynamicCast<JSTextDecoder*>(callFrame->thisValue());
+if (UNLIKELY(!thisObject)) {
+    scope.throwException(lexicalGlobalObject, 
+        Bun::createInvalidThisError(lexicalGlobalObject, callFrame->thisValue(), "TextDecoder"_s));
+    return {};
+}
+```
+
+## Prototypal Inheritance
+
+The binding system creates proper JavaScript prototype chains:
+
+1. **Constructor**: JSTextDecoderConstructor with standard .prototype property
+2. **Prototype**: JSTextDecoderPrototype with methods and properties
+3. **Instances**: Each JSTextDecoder instance with __proto__ pointing to prototype
+
+This ensures JavaScript inheritance works as expected:
+
+```cpp
+// From generated code
+void JSTextDecoderConstructor::finishCreation(VM& vm, JSC::JSGlobalObject* globalObject, JSTextDecoderPrototype* prototype)
+{
+    Base::finishCreation(vm, 0, "TextDecoder"_s, PropertyAdditionMode::WithoutStructureTransition);
+    
+    // Set up the prototype chain
+    putDirectWithoutTransition(vm, vm.propertyNames->prototype, prototype, PropertyAttribute::DontEnum | PropertyAttribute::DontDelete | PropertyAttribute::ReadOnly);
+    ASSERT(inherits(info()));
+}
+```
+
+## Performance Considerations
+
+The binding system is optimized for performance:
+
+1. **Direct Pointer Access**: JavaScript objects maintain a direct pointer to Zig objects
+2. **Property Caching**: WriteBarrier caching avoids repeated native calls for stable properties 
+3. **Memory Management**: JSC garbage collection integrated with Zig memory management
+4. **Type Conversion**: Fast paths for common JavaScript/Zig type conversions
+
+## Creating a New Class Binding
+
+To create a new class binding in Bun:
+
+1. **Define the class interface** in a `.classes.ts` file:
+   ```typescript
+   define({
+     name: "MyClass",
+     constructor: true,
+     finalize: true,
+     proto: {
+       myMethod: {
+         args: 1,
+       },
+       myProperty: {
+         getter: true,
+         cache: true,
+       }
+     }
+   });
+   ```
+
+2. **Implement the native functionality** in a `.zig` file:
+   ```zig
+   pub const MyClass = struct {
+       // State
+       value: []const u8,
+       
+       // Generated bindings
+       pub usingnamespace JSC.Codegen.JSMyClass;
+       pub usingnamespace bun.New(@This());
+       
+       // Constructor
+       pub fn constructor(
+           globalObject: *JSGlobalObject,
+           callFrame: *JSC.CallFrame,
+       ) bun.JSError!*MyClass {
+           const arg = callFrame.argument(0);
+           // Implementation
+       }
+       
+       // Method
+       pub fn myMethod(
+           this: *MyClass,
+           globalObject: *JSGlobalObject,
+           callFrame: *JSC.CallFrame,
+       ) bun.JSError!JSC.JSValue {
+           // Implementation
+       }
+       
+       // Getter
+       pub fn getMyProperty(this: *MyClass, globalObject: *JSGlobalObject) JSC.JSValue {
+           return JSC.JSValue.jsString(globalObject, this.value);
+       }
+       
+       // Resource cleanup
+       pub fn deinit(this: *MyClass) void {
+           // Clean up resources
+       }
+       
+       pub fn finalize(this: *MyClass) void {
+           this.deinit();
+           bun.default_allocator.destroy(this);
+       }
+   };
+   ```
+
+3. **The binding generator** creates all necessary C++ and Zig glue code to connect JavaScript and Zig, including:
+   - C++ class definitions
+   - Method and property bindings
+   - Memory management utilities
+   - GC integration code
+
+## Generated Code Structure
+
+The binding generator produces several components:
+
+### 1. C++ Classes
+
+For each Zig class, the system generates:
+
+- **JS<Class>**: Main wrapper that holds a pointer to the Zig object (`JSTextDecoder`)
+- **JS<Class>Prototype**: Contains methods and properties (`JSTextDecoderPrototype`)
+- **JS<Class>Constructor**: Implementation of the JavaScript constructor (`JSTextDecoderConstructor`)
+
+### 2. C++ Methods and Properties
+
+- **Method Callbacks**: `TextDecoderPrototype__decodeCallback`
+- **Property Getters/Setters**: `TextDecoderPrototype__encodingGetterWrap`
+- **Initialization Functions**: `finishCreation` methods for setting up the class
+
+### 3. Zig Bindings
+
+- **External Function Declarations**:
+  ```zig
+  extern fn TextDecoderPrototype__decode(*TextDecoder, *JSC.JSGlobalObject, *JSC.CallFrame) callconv(JSC.conv) JSC.EncodedJSValue;
+  ```
+
+- **Cached Value Accessors**:
+  ```zig
+  pub fn encodingGetCached(thisValue: JSC.JSValue) ?JSC.JSValue { ... }
+  pub fn encodingSetCached(thisValue: JSC.JSValue, globalObject: *JSC.JSGlobalObject, value: JSC.JSValue) void { ... }
+  ```
+
+- **Constructor Helpers**:
+  ```zig
+  pub fn create(globalObject: *JSC.JSGlobalObject) bun.JSError!JSC.JSValue { ... }
+  ```
+
+### 4. GC Integration
+
+- **Memory Cost Calculation**: `estimatedSize` method
+- **Child Visitor Methods**: `visitChildrenImpl` and `visitAdditionalChildren`
+- **Heap Analysis**: `analyzeHeap` for debugging memory issues
+
+This architecture makes it possible to implement high-performance native functionality in Zig while exposing a clean, idiomatic JavaScript API to users.