Update documentation for class file parsing, class loading, frame interpreter, JNI, object management, and FFI. Increment crate version to 0.2.0;

2025-12-26 18:10:52 +10:30 · 2025-12-26 18:10:52 +10:30 · 24939df1b7
commit 24939df1b7
parent 7fcf00b77f
10 changed files with 960 additions and 114 deletions
--- a/README.md
+++ b/README.md
@ -1,146 +1,99 @@
 # RoastVM

-A Java Virtual Machine (JVM) implementation written in Rust, capable of parsing and executing Java class files and bytecode.
+A Java Virtual Machine (JVM) implementation written in Rust.

 ## Overview

-RoastVM is an educational/experimental JVM implementation that demonstrates the core components and execution model of the Java Virtual Machine. The project uses Rust's type safety and modern tooling to build a simplified but functional JVM interpreter.
+RoastVM is an educational/experimental JVM implementation that executes Java bytecode. It supports class file parsing,
+bytecode interpretation, JNI native methods, and includes a boot image system for loading the Java standard library.

 ## Features

-### Currently Implemented
-
- **Class File Parsing**: Full support for reading and deserializing binary Java class files (`.class`) with magic number `0xCAFEBABE`
- **Constant Pool Management**: Handles 20+ constant pool entry types (UTF8, Integer, Float, Long, Double, Class, String, MethodRef, FieldRef, InterfaceMethodRef, NameAndType, MethodHandle, MethodType, InvokeDynamic, etc.)
- **Dynamic Class Loading**: On-demand class loading with superclass and interface resolution, caching via DashMap
- **Class Initialization**: Automatic `<clinit>` method execution following JVM Spec 5.5, with recursive initialization tracking
- **Bytecode Execution**: Interpreter for 50+ JVM bytecode instructions including:
-  - Constants: `aconst_null`, `iconst_*`, `lconst_*`, `fconst_*`, `dconst_*`, `bipush`, `sipush`, `ldc`, `ldc_w`, `ldc2_w`
-  - Load/Store: `iload`, `lload`, `fload`, `dload`, `aload`, `istore`, `lstore`, `fstore`, `dstore`, `astore` (including `_0-3` variants)
-  - Array operations: `iaload`, `laload`, `faload`, `daload`, `aaload`, `baload`, `caload`, `saload`, `iastore`, `lastore`, `fastore`, `dastore`, `aastore`, `bastore`, `castore`, `sastore`, `arraylength`
-  - Stack manipulation: `pop`, `pop2`, `dup`, `dup_x1`, `dup_x2`, `dup2`, `dup2_x1`, `dup2_x2`, `swap`
-  - Arithmetic: `iadd`, `ladd`, `fadd`, `dadd`, `isub`, `lsub`, `fsub`, `dsub`, `imul`, `lmul`, `fmul`, `dmul`, `idiv`, `ldiv`, `fdiv`, `ddiv`, `irem`, `lrem`, `frem`, `drem`, `ineg`, `lneg`, `fneg`, `dneg`
-  - Bitwise: `ishl`, `lshl`, `ishr`, `lshr`, `iushr`, `lushr`, `iand`, `land`, `ior`, `lor`, `ixor`, `lxor`
-  - Type conversions: `i2l`, `i2f`, `i2d`, `l2i`, `l2f`, `l2d`, `f2i`, `f2l`, `f2d`, `d2i`, `d2l`, `d2f`, `i2b`, `i2c`, `i2s`
-  - Comparisons: `lcmp`, `fcmpl`, `fcmpg`, `dcmpl`, `dcmpg`
-  - Control flow: `ifeq`, `ifne`, `iflt`, `ifge`, `ifgt`, `ifle`, `if_icmp*`, `if_acmp*`, `goto`, `ifnull`, `ifnonnull`
-  - Object operations: `new`, `newarray`, `anewarray`, `multianewarray`, `checkcast`, `instanceof`
-  - Field access: `getstatic`, `putstatic`, `getfield`, `putfield`
-  - Method invocation: `invokevirtual`, `invokespecial`, `invokestatic`, `invokeinterface`
-  - Returns: `ireturn`, `lreturn`, `freturn`, `dreturn`, `areturn`, `return`
- **Object Model**: Full object creation, field storage, and array support (primitive and reference arrays)
- **JNI Support**: Implementation of 80+ JNI functions for native method integration
- **Native Library Loading**: Dynamic loading of native libraries (DLLs on Windows)
- **Stack Traces**: Detailed stack trace generation with line number mapping from class file attributes
- **Module System**: Support for loading classes from 7z binary image archives (JDK modules)
- **Frame-based Execution**: Proper execution context with program counter, operand stack, and local variables
-
-### In Development
-
- Additional bytecode instructions (`tableswitch`, `lookupswitch`, `monitorenter`, `monitorexit`, etc.)
- Exception handling (`athrow`, try/catch blocks)
- Garbage collection (basic object manager exists)
- Reflection API
- Multi-threading support
- Method handles and `invokedynamic`
-
-## Architecture
-
-### Core Components
-
- **`Vm`** (`vm.rs`): Main virtual machine controller managing threads, class loader, and native library loading
- **`VmThread`** (`thread.rs`): Thread of execution managing the frame stack and method invocation
- **`Frame`** (`lib.rs`): Execution context for a method with PC, operand stack, and local variables
- **`ClassLoader`** (`class_loader.rs`): Handles dynamic class loading, linking, and initialization
- **`RuntimeClass`** (`class.rs`): Runtime representation of a loaded class with initialization state tracking
- **`ClassFile`** (`class_file/`): Binary parser for Java class files using the `deku` library
- **`ConstantPool`** (`class_file/constant_pool.rs`): Constant pool resolution and management
- **`ObjectManager`** (`objects/object_manager.rs`): Object allocation and garbage collection management
- **`JNI`** (`jni.rs`): Java Native Interface implementation
-
-### Execution Flow
-
-1. **Loading**: `ClassFile::from_bytes()` parses binary class file data
-2. **Resolution**: `ClassLoader` converts `ClassFile` to `RuntimeClass`, resolving dependencies
-3. **Initialization**: Class initializers (`<clinit>`) execute per JVM Spec 5.5
-4. **Execution**: `VmThread` invokes the main method, creating a `Frame`
-5. **Interpretation**: `Frame` iterates through bytecode operations, executing each instruction
-6. **Stack Operations**: Instructions manipulate the operand stack and local variables
+- **Class File Parsing** - Full `.class` file support using deku for binary parsing
+- **Bytecode Interpreter** - 200+ JVM instructions implemented
+- **Object Model** - Objects, arrays, monitors, and string interning
+- **JNI Support** - 250+ JNI functions for native method integration
+- **Boot Image** - Load JDK classes from 7z module archives
+- **Native FFI** - Dynamic library loading via libffi

 ## Project Structure

 ```
 roast-vm/
-├── Cargo.toml                      # Workspace configuration
-└── crates/
-    ├── core/                       # Main JVM implementation (roast-vm-core)
-    │   ├── Cargo.toml
-    │   └── src/
-    │       ├── main.rs             # Entry point (binary: roast)
-    │       ├── lib.rs              # Frame and bytecode execution
-    │       ├── vm.rs               # Virtual Machine controller
-    │       ├── thread.rs           # Thread execution management
-    │       ├── class.rs            # RuntimeClass definition
-    │       ├── class_loader.rs     # ClassLoader implementation
-    │       ├── class_file/         # Binary class file parser
-    │       │   ├── class_file.rs   # ClassFile parser (magic 0xCAFEBABE)
-    │       │   └── constant_pool.rs
-    │       ├── objects/            # Object model
-    │       │   ├── object.rs       # Object representation
-    │       │   ├── array.rs        # Array support
-    │       │   └── object_manager.rs
-    │       ├── jni.rs              # JNI implementation
-    │       ├── instructions.rs     # Bytecode opcode definitions
-    │       ├── attributes.rs       # Class file attributes
-    │       ├── value.rs            # Value and stack types
-    │       ├── error.rs            # Error handling and stack traces
-    │       ├── native_libraries.rs # Native library management
-    │       └── bimage.rs           # Binary image (7z) reader
+├── crates/
+│   ├── core/                    # Main VM implementation
+│   │   └── src/
+│   │       ├── main.rs          # Entry point
+│   │       ├── vm.rs            # VM controller
+│   │       ├── thread.rs        # Thread execution
+│   │       ├── class_loader.rs  # Class loading
+│   │       ├── bimage.rs        # Boot image reader
+│   │       ├── frame/           # Stack frames & interpreter
+│   │       ├── class_file/      # Class file parser
+│   │       ├── objects/         # Object/array model
+│   │       └── native/          # JNI infrastructure
+│   │
+│   └── roast-vm-sys/            # Native methods (cdylib)
+│       └── src/                 # JNI implementations
 │
-    └── roast-vm-sys/               # Native methods bridge (cdylib)
-        ├── Cargo.toml
-        └── src/
-            ├── lib.rs              # Native method implementations
-            ├── system.rs           # System native calls
-            ├── class.rs            # Class native operations
-            └── object.rs           # Object native operations
+├── lib/                         # Boot image location
+├── data/                        # Default classpath
+└── docs/                        # Detailed documentation
 ```

-## Dependencies
+## Documentation

- **`deku`**: Binary parsing and serialization for class files
- **`dashmap`**: Concurrent HashMap for class and object storage
- **`jni`**: Java Native Interface bindings
- **`libloading`**: Dynamic library loading
- **`libffi`**: Foreign function interface for native calls
- **`sevenz-rust2`**: 7z archive reading for module system support
- **`log`** / **`env_logger`**: Logging infrastructure
- **`itertools`**: Iterator utilities
- **`colored`**: Colored console output
+Detailed implementation docs are in the `docs/` folder:
+
+- [Class File Parsing](docs/class-file-parsing.md) - Binary format, constant pool, attributes
+- [Frame & Interpreter](docs/frame-interpreter.md) - Stack frames, opcode dispatch
+- [Object Management](docs/object-management.md) - Objects, arrays, monitors
+- [JNI](docs/jni.md) - JNIEnv structure, native invocation
+- [Native/FFI](docs/native-ffi.md) - Library loading, libffi integration
+- [roast-vm-sys](docs/roast-vm-sys.md) - Native method implementations
+- [Class Loading](docs/class-loading.md) - Boot image, classpath, RuntimeClass

 ## Building

 ```bash
-# Build the project
 cargo build
-
-# Build with optimizations
 cargo build --release
-
-# Run tests
 cargo test
+```

-# Run with logging
+## Running
+
+```bash
+# Run with default classpath (./data)
+cargo run
+
+# Run with custom classpath
+cargo run -- /path/to/classes
+
+# With debug logging
 RUST_LOG=debug cargo run
 ```

-## Current Status
+## Dependencies

-This project is in early development (v0.1.0). The core infrastructure for class loading, bytecode execution, object creation, JNI support, and stack traces is functional. Many JVM features remain in development.
+| Crate                                                 | Purpose                   |
+|-------------------------------------------------------|---------------------------|
+| [deku](https://crates.io/crates/deku)                 | Binary class file parsing |
+| [dashmap](https://crates.io/crates/dashmap)           | Concurrent maps           |
+| [jni](https://crates.io/crates/jni)                   | JNI type definitions      |
+| [libloading](https://crates.io/crates/libloading)     | Dynamic library loading   |
+| [libffi](https://crates.io/crates/libffi)             | Native function calls     |
+| [sevenz-rust2](https://crates.io/crates/sevenz-rust2) | Boot image archives       |
+| [parking_lot](https://crates.io/crates/parking_lot)   | Synchronization           |

+## Status

+Early development (v0.2.0). Core class loading, bytecode execution, and JNI are functional. Exception handling and GC
+are in progress.
+
+**Vendor**: infernap12

 ## References

 - [JVM Specification](https://docs.oracle.com/javase/specs/jvms/se25/html/index.html)
- [Java Class File Format](https://docs.oracle.com/javase/specs/jvms/se25/html/jvms-4.html)
+- [JNI Specification](https://docs.oracle.com/en/java/javase/25/docs/specs/jni/index.html)
--- a/crates/core/Cargo.toml
+++ b/crates/core/Cargo.toml
@ -1,6 +1,6 @@
 [package]
 name = "roast-vm-core"
-version = "0.1.5"
+version = "0.2.0"
 edition = "2024"
 publish = ["nexus"]

--- a/crates/roast-vm-sys/Cargo.toml
+++ b/crates/roast-vm-sys/Cargo.toml
@ -1,6 +1,6 @@
 [package]
 name = "roast-vm-sys"
-version = "0.1.5"
+version = "0.2.0"
 edition = "2024"
 publish = ["nexus"]

--- a/docs/class-file-parsing.md
+++ b/docs/class-file-parsing.md
@ -0,0 +1,156 @@
+# Class File Parsing
+
+**Location**: `crates/core/src/class_file/`
+
+The class file parser uses the **deku** library for declarative binary deserialization with automatic validation.
+
+## Components
+
+| File | Purpose |
+|------|---------|
+| `class_file.rs` | Main ClassFile struct with version, constant pool, fields, methods, attributes |
+| `constant_pool.rs` | ConstantPoolGet/ConstantPoolExt traits for pool access and resolution |
+| `attributes.rs` | Attribute parsing (Code, LineNumberTable, LocalVariableTable, BootstrapMethods) |
+| `mod.rs` | Access flag definitions (ClassFlags, MethodFlags, FieldFlags) |
+
+## Key Types
+
+```rust
+pub struct ClassFile {
+    pub minor_version: u16,
+    pub major_version: u16,
+    pub constant_pool: Arc<ConstantPoolOwned>,
+    pub access_flags: u16,
+    pub this_class: u16,
+    pub super_class: u16,
+    pub interfaces: Vec<u16>,
+    pub fields: Vec<FieldInfo>,
+    pub methods: Vec<MethodInfo>,
+    pub attributes: Vec<AttributeInfo>,
+}
+
+pub struct FieldInfo {
+    pub access_flags: u16,
+    pub name_index: u16,
+    pub descriptor_index: u16,
+    pub attributes: Vec<AttributeInfo>,
+}
+
+pub struct MethodInfo {
+    pub access_flags: u16,
+    pub name_index: u16,
+    pub descriptor_index: u16,
+    pub attributes: Vec<AttributeInfo>,
+}
+```
+
+## Constant Pool
+
+Trait-based architecture with two layers:
+
+### ConstantPoolGet Trait
+Low-level accessors:
+- `get_constant()`: Resolve by index (accounts for 64-bit entries)
+- Type-specific getters: `get_i32()`, `get_utf8_info()`, `get_class_info()`, `get_method_ref()`, etc.
+- Implemented via `pool_get_impl!` macro
+
+### ConstantPoolExt Trait
+High-level operations:
+- `get_string()`: Fetch UTF-8 strings with CESU-8 decoding
+- `resolve_class_name()`: Trace class references through constant pool
+- `resolve_method_ref()` / `resolve_interface_method_ref()`: Resolve method references
+- `resolve_field()`: Resolve field references with type descriptors
+- `parse_attribute()`: Convert raw attribute bytes to typed Attribute enum
+
+### Constant Pool Entry Types (20 types)
+
+```rust
+pub enum ConstantPoolEntry {
+    Utf8(ConstantUtf8Info),
+    Integer(i32), Float(f32), Long(i64), Double(f64),
+    Class(ConstantClassInfo),
+    String(ConstantStringInfo),
+    FieldRef(ConstantFieldrefInfo),
+    MethodRef(ConstantMethodrefInfo),
+    InterfaceMethodRef(ConstantInterfaceMethodrefInfo),
+    NameAndType(ConstantNameAndTypeInfo),
+    MethodHandle(ConstantMethodHandleInfo),
+    MethodType(ConstantMethodTypeInfo),
+    Dynamic(ConstantDynamicInfo),
+    InvokeDynamic(ConstantInvokeDynamicInfo),
+    Module(ConstantModuleInfo),
+    Package(ConstantPackageInfo),
+}
+```
+
+## Attributes
+
+Recursive attribute parsing with support for:
+
+- **Code**: Method bytecode with max_stack, max_locals, exception tables, nested attributes
+- **LineNumberTable**: Maps bytecode offsets to source line numbers
+- **LocalVariableTable**: Local variable debugging info (name, descriptor, PC range)
+- **BootstrapMethods**: Dynamic invocation bootstrap method references
+- **StackMapTable**, **Exceptions**, **InnerClasses**: Parsed as raw byte vectors
+- **SourceFile**, **Signature**: Index-based attribute data
+- **Unknown**: Fallback for unrecognized attributes
+
+### Code Attribute Structure
+
+```rust
+pub struct CodeAttribute {
+    pub max_stack: u16,
+    pub max_locals: u16,
+    pub code_length: u32,
+    pub code: Vec<u8>,
+    pub exception_table: Vec<ExceptionTableEntry>,
+    pub attributes: Vec<AttributeInfo>,  // Recursive
+}
+```
+
+## Access Flags
+
+Bitfield structures for parsing JVM access flags:
+
+- **ClassFlags**: PUBLIC, FINAL, INTERFACE, ABSTRACT, SYNTHETIC, ANNOTATION, ENUM, MODULE
+- **FieldFlags**: PUBLIC, PRIVATE, PROTECTED, STATIC, FINAL, VOLATILE, TRANSIENT, SYNTHETIC, ENUM
+- **MethodFlags**: PUBLIC, PRIVATE, PROTECTED, STATIC, FINAL, SYNCHRONIZED, BRIDGE, VARARGS, NATIVE, ABSTRACT, STRICT, SYNTHETIC
+
+## Validation
+
+Validation occurs at multiple levels:
+
+1. **Binary Format** (Automatic via Deku):
+   - Magic number (0xCAFEBABE)
+   - Big-endian byte order
+   - Type-safe parsing with error propagation
+
+2. **Constant Pool**:
+   - Index bounds checking in `get_constant()`
+   - Type validation: Each accessor checks the entry type matches expected type
+   - CESU-8 decoding errors caught from Java-style UTF-8 strings
+
+3. **Class Structure** (ClassLoader):
+   - Debug assertions for Object class having super_class = 0
+   - Non-Object classes must have valid super class reference
+   - Interfaces must inherit from Object
+
+## Descriptor Parsing
+
+```rust
+// Method descriptor: (II)I  ->  two ints, return int
+MethodDescriptor::parse("(II)I")?
+
+// Field descriptor: Ljava/lang/String;  ->  String class type
+FieldType::parse("Ljava/lang/String;")?
+
+// Array descriptor: [[I  ->  2D int array
+FieldType::parse("[[I")?
+```
+
+## Error Handling
+
+- `DekuError`: Binary parsing failures
+- `ConstantPoolError`: Pool access with Generic, DescriptorParseError, Cesu8DecodingError variants
+- `VmError`: Higher-level VM-specific errors
+- `DescParseError`: Invalid method/field descriptor syntax
--- a/docs/class-loading.md
+++ b/docs/class-loading.md
@ -0,0 +1,134 @@
+# Class Loading and Boot Image
+
+**Location**: `crates/core/src/class_loader.rs`, `crates/core/src/bimage.rs`
+
+## Boot Image (Bimage)
+
+A 7z archive containing precompiled Java standard library classes.
+
+### Structure
+
+```rust
+pub struct Bimage {
+    image: ArchiveReader<File>,       // 7z archive reader
+    modules: Vec<String>,              // Available modules
+    packages: HashMap<String, String>, // Package -> Module mapping
+    pub total_access_time: Duration,   // Performance tracking
+}
+```
+
+- **Default Location**: `./lib/modules`
+- **Format**: 7z compressed archive
+- **Structure**: `<module>/classes/<class>.class`
+- **Default Module**: `java.base` (used when no module is specified)
+
+## Class Loading Flow
+
+The `ClassLoader` manages class resolution with a two-tier fallback mechanism:
+
+### Process
+
+1. **Check Cache**: Look in `DashMap<(String, LoaderId), Arc<RuntimeClass>>` for already-loaded classes
+2. **Try Bimage**: Attempt to load from boot image via `bimage.get_class(module, class_fqn)`
+3. **Fallback to Disk**: If not in bimage, load from filesystem at `{CLASSPATH}/{class_name}.class`
+4. **Parse & Cache**: Parse ClassFile using deku, create RuntimeClass, store in cache
+
+### Key Method
+
+```rust
+pub fn load_class(&mut self, what: &str, loader: LoaderId) -> Result<Arc<RuntimeClass>, VmError> {
+    let bytes = self.bimage
+        .and_then(|b| b.get_class("", what).ok())
+        .or_else(|_| Self::load_class_from_disk(what))
+        .map_err(|_| VmError::LoaderError(...))?;
+
+    let (_, cf) = ClassFile::from_bytes(bytes.as_ref())?;
+    let runtime = self.runtime_class(cf);
+
+    // Store with loader ID for multi-loader support
+    self.classes.insert((class_fqn, loader), Arc::new(runtime));
+}
+```
+
+## Classpath Handling
+
+### Resolution Priority
+
+1. **Bimage (boot image)** - Primary source for standard library
+2. **Command-line argument (arg[1])** - User-provided classpath
+3. **Default `./data` directory** - Fallback location
+
+### Implementation
+
+```rust
+fn load_class_from_disk(what: &str) -> Result<Vec<u8>, String> {
+    let class_path = std::env::args()
+        .nth(1)
+        .unwrap_or("./data".to_string())
+        .replace("\\", "/");
+
+    let path = format!("{class_path}/{what}.class");
+    // Load file from disk
+}
+```
+
+## Bootstrap Process
+
+**Location**: `crates/core/src/vm.rs` - `boot_strap()` method
+
+### Steps
+
+1. **Create VM**: `ClassLoader::with_bimage()` - initializes with boot image
+2. **Load Core Classes**: Preloads essential VM classes
+3. **Create Primitive Classes**: Synthetic class objects for primitive types
+4. **Initialize Classes**: Run static initializers (`<clinit>`)
+
+### Core Classes Loaded
+
+```rust
+let classes = vec![
+    "java/lang/String",
+    "java/lang/System",
+    "java/lang/Class",
+    "java/lang/Object",
+    "java/lang/Thread",
+    "java/lang/ThreadGroup",
+    "java/lang/Module",
+    "java/lang/reflect/Method",
+    // ...
+];
+```
+
+## RuntimeClass
+
+Runtime representation of a loaded class:
+
+### Cached Data
+
+- **Superclass Chain**: `super_classes: Vec<Arc<RuntimeClass>>`
+- **Interface Hierarchy**: `super_interfaces: Vec<Arc<RuntimeClass>>`
+- **Component Type**: For array classes, reference to element type
+- **Initialization State**: Thread-safe `InitState` enum
+
+### Initialization States
+
+```rust
+pub enum InitState {
+    NotInitialized,
+    Initializing(ThreadId),  // Track which thread is initializing
+    Initialized,
+}
+```
+
+### Method/Field Resolution
+
+- `find_method()` - Searches class then walks up superclass chain
+- `find_field()` - Same recursive behavior for fields
+- `is_assignable_into()` - Checks type compatibility with array covariance
+
+## Multi-Loader Support
+
+Classes are keyed by `(class_name, LoaderId)` tuple to support:
+- Different class loaders loading same-named classes
+- Isolation between class loader namespaces
+- Proper class identity checks
--- a/docs/frame-interpreter.md
+++ b/docs/frame-interpreter.md
@ -0,0 +1,141 @@
+# Frame and Bytecode Interpreter
+
+**Location**: `crates/core/src/frame/`
+
+## Frame Structure
+
+Each method invocation creates a `Frame` containing:
+
+| Component | Description |
+|-----------|-------------|
+| Program Counter (PC) | i64 tracking current bytecode instruction |
+| Operand Stack | Generic Vec-backed stack for intermediate values |
+| Local Variables | Indexed slots, handles wide values (long/double occupy 2 slots) |
+| Constant Pool | Arc reference to the class constant pool |
+| Bytecode | Instructions for the method |
+
+### OperandStack (`operand_stack.rs`)
+
+- Generic Vec-backed stack with push/pop/peek operations
+- `pop_n(n)` returns values in push order (not pop order) for method arguments
+- Supports underflow detection
+
+### LocalVariables (`local_vars.rs`)
+
+- Vec-backed, indexed by slot
+- Handles wide values (long, double) that occupy 2 slots with padding
+- `from_args()` automatically spaces wide values correctly
+- Prevents access to padding slots with runtime panic
+
+## Execution Loop
+
+```rust
+loop {
+    let (offset, op) = self.next().unwrap();
+    self.pc = offset as i64;
+    let result = self.execute_instruction(op.clone());
+    match result {
+        Ok(ExecutionResult::Advance(offset)) => self.pc += offset as i64,
+        Ok(_) => self.pc += 1,
+        Err(x) => return error with stack trace,
+    }
+}
+```
+
+## Opcode Dispatch
+
+**Location**: `frame.rs` `execute_instruction()` (lines 199-1516)
+
+- Single match statement over `Ops` enum variants (defined in `instructions.rs`)
+- 200+ opcodes: constants, loads/stores, math, stack ops, branches, references, method invocation
+- Uses helper macros (`load!`, `store!`, `binary_op!`, `shift_op!`, etc.) for common patterns
+- Each opcode returns one of:
+  - `ExecutionResult::Continue` (auto-increment PC by 1)
+  - `ExecutionResult::Advance(offset)` (jump)
+  - `ExecutionResult::Return(())` or `ExecutionResult::ReturnValue(Value)` (exit frame)
+  - `VmError` on failure
+
+## Opcode Encoding (`instructions.rs`)
+
+- Uses `deku` derive for binary deserialization
+- Each opcode has a u8 ID (0x00-0xFF)
+- Some opcodes carry operands (e.g., `iload(u8)`, `goto(i16)`)
+- Wide instruction prefix (0xC4) for accessing local slots > 255
+
+## Method Invocation
+
+**Location**: `thread.rs` (lines 308-338)
+
+### Invocation Types
+
+1. **`invoke()`**: Resolve method by class and descriptor, execute it
+2. **`invoke_virtual()`**: Virtual dispatch - find method on actual runtime class
+3. **`invoke_native()`**: Call native JNI method via FFI
+
+### Bytecode Invocation Instructions
+
+- **`invokevirtual`**: Virtual method dispatch - pop receiver + arguments, get actual class, call `thread.invoke_virtual()`
+- **`invokespecial`**: Non-virtual (constructors, private, super) - pop receiver + arguments, call `thread.invoke()` with static resolution
+- **`invokestatic`**: Static methods - pop arguments only (no receiver), call `thread.invoke()`
+- **`invokeinterface`**: Interface method dispatch - similar to `invokevirtual` with interface resolution
+
+### Frame Creation & Execution
+
+```rust
+fn execute_method(&self, class: &Arc<RuntimeClass>, method: &MethodData, args: Vec<Value>) {
+    let mut frame = Frame::new(
+        class.clone(),
+        method_ref,
+        code_attr,  // Contains max_stack, max_locals, bytecode
+        args,       // Initialize local vars with parameters
+        ...
+    );
+    self.frame_stack.lock().push(frame.clone());
+    frame.execute()  // Bytecode interpretation loop
+    self.frame_stack.lock().pop();
+}
+```
+
+## Supported Instructions
+
+### Constants
+`aconst_null`, `iconst_*`, `lconst_*`, `fconst_*`, `dconst_*`, `bipush`, `sipush`, `ldc`, `ldc_w`, `ldc2_w`
+
+### Load/Store
+`iload`, `lload`, `fload`, `dload`, `aload`, `istore`, `lstore`, `fstore`, `dstore`, `astore` (including `_0-3` variants)
+
+### Array Operations
+`iaload`, `laload`, `faload`, `daload`, `aaload`, `baload`, `caload`, `saload`, `iastore`, `lastore`, `fastore`, `dastore`, `aastore`, `bastore`, `castore`, `sastore`, `arraylength`
+
+### Stack Manipulation
+`pop`, `pop2`, `dup`, `dup_x1`, `dup_x2`, `dup2`, `dup2_x1`, `dup2_x2`, `swap`
+
+### Arithmetic
+All int/long/float/double add, sub, mul, div, rem, neg operations
+
+### Bitwise
+`ishl`, `lshl`, `ishr`, `lshr`, `iushr`, `lushr`, `iand`, `land`, `ior`, `lor`, `ixor`, `lxor`
+
+### Type Conversions
+All primitive type conversions (`i2l`, `l2i`, `f2d`, etc.)
+
+### Comparisons
+`lcmp`, `fcmpl`, `fcmpg`, `dcmpl`, `dcmpg`
+
+### Control Flow
+`ifeq`, `ifne`, `iflt`, `ifge`, `ifgt`, `ifle`, `if_icmp*`, `if_acmp*`, `goto`, `ifnull`, `ifnonnull`, `tableswitch`, `lookupswitch`
+
+### Object Operations
+`new`, `newarray`, `anewarray`, `multianewarray`, `checkcast`, `instanceof`
+
+### Field Access
+`getstatic`, `putstatic`, `getfield`, `putfield`
+
+### Method Invocation
+`invokevirtual`, `invokespecial`, `invokestatic`, `invokeinterface`
+
+### Returns
+`ireturn`, `lreturn`, `freturn`, `dreturn`, `areturn`, `return`
+
+### Synchronization
+`monitorenter`, `monitorexit`
--- a/docs/jni.md
+++ b/docs/jni.md
@ -0,0 +1,155 @@
+# JNI Implementation
+
+**Location**: `crates/core/src/native/jni.rs`
+
+## JNIEnv Structure
+
+The JNIEnv is created as a function table (`JNINativeInterface_` from the `jni` crate) with 250+ function pointers:
+
+```rust
+pub fn create_jni_function_table(thread: *const VmThread) -> JNIEnv {
+	Box::into_raw(Box::new(JNINativeInterface_ {
+		reserved0: thread as *mut _, // Stores pointer to VmThread for context
+		reserved1: std::ptr::null_mut(),
+		reserved2: std::ptr::null_mut(),
+		reserved3: std::ptr::null_mut(),
+		GetVersion: Some(jni_get_version),
+		FindClass: Some(find_class),
+		RegisterNatives: Some(register_natives),
+		GetMethodID: Some(get_method_id),
+// ... 240+ more function pointers
+	}))
+}
+```
+
+**Key Feature:** The `reserved0` field stores a pointer to the `VmThread`, allowing each JNI function to access thread
+context via `get_thread(env)`.
+
+## VmThread's JNIEnv Storage
+
+**Location**: `crates/core/src/thread.rs`
+
+Each thread owns its JNIEnv:
+
+```rust
+pub struct VmThread {
+	pub id: ThreadId,
+	pub vm: Arc<Vm>,
+	pub loader: Arc<Mutex<ClassLoader>>,
+	pub jni_env: JNIEnv,  // Stored per-thread
+	// ... other fields
+}
+
+// Created during VmThread initialization:
+let jni_env = create_jni_function_table(weak_self.as_ptr() as * mut VmThread);
+```
+
+## Native Method Invocation
+
+**Location**: `crates/core/src/thread.rs` (lines 340-428)
+
+The flow is:
+
+1. **Method Detection:** When a method has `ACC_NATIVE` flag, `invoke_native()` is called
+2. **Symbol Resolution:** Generates JNI symbol name (e.g., `Java_java_lang_String_intern`)
+3. **Lookup:** Searches registered native methods or loaded native libraries via `find_native_method()`
+4. **FFI Call:** Uses `libffi` to call the native function with constructed arguments
+
+```rust
+pub fn invoke_native(&self, method: &MethodRef, args: Vec<Value>) -> MethodCallResult {
+	let symbol_name = generate_jni_method_name(method, false);
+
+	// Find the function pointer
+	let p = self.vm.find_native_method(&symbol_name)
+				.ok_or(VmError::NativeError(...))?;
+
+	// Build Call Interface (CIF) for FFI
+	let cp = CodePtr::from_ptr(p);
+	let built_args = build_args(args, &mut storage, &self.jni_env as *mut JNIEnv);
+	let cif = method.build_cif();
+
+	// Invoke with type-specific call
+	match &method.desc.return_type {
+		None => {
+			cif.call::<()>(cp, built_args.as_ref());
+			Ok(None)
+		}
+		Some(FieldType::Base(BaseType::Int)) => {
+			let v = cif.call::<jint>(cp, built_args.as_ref());
+			Ok(Some(v.into()))
+		}
+		// ... handle other return types
+	}
+}
+```
+
+## Argument Marshalling
+
+**Location**: `crates/core/src/thread.rs` (lines 509-548)
+
+Native functions receive:
+
+1. `JNIEnv*` - pointer to the function table
+2. `jclass` or `jobject` (receiver) - always an ID (u32)
+3. Parameters - converted from VM `Value` types to JNI types
+
+```rust
+fn build_args(mut params: VecDeque<Value>, storage: &mut Vec<Box<dyn Any>>,
+			  jnienv: *mut JNIEnv) -> Vec<Arg> {
+	storage.push(Box::new(jnienv));                    // Slot 0: JNIEnv*
+	let receiver_id = params.pop_front().map(...);
+	storage.push(Box::new(receiver_id as jobject));    // Slot 1: this/class
+
+	for value in params {
+		match value {
+			Value::Primitive(Primitive::Int(x)) => storage.push(Box::new(x)),
+			Value::Reference(Some(ref_kind)) => {
+				storage.push(Box::new(ref_kind.id() as jobject))  // References as IDs
+			}
+			// ... other types
+		}
+	}
+	storage.iter().map(|boxed| arg(&**boxed)).collect()
+}
+```
+
+## Native Method Registration
+
+**Location**: `crates/core/src/native/jni.rs` (lines 381-442)
+
+Java code calls `RegisterNatives()` JNI function, which stores pointers:
+
+```rust
+unsafe extern "system" fn register_natives(
+	env: *mut JNIEnv,
+	clazz: jclass,
+	methods: *const JNINativeMethod,
+	n_methods: jint,
+) -> jint {
+	let thread = &*get_thread(env);
+
+	for i in 0..n_methods as usize {
+		let native_method = &*methods.add(i);
+		let full_name = generate_jni_short_name(&class_name, name);
+
+		thread.vm.native_methods.insert(full_name, native_method.fnPtr);
+	}
+	JNI_OK
+}
+```
+
+## Implemented JNI Functions
+
+| Category          | Functions                                                    |
+|-------------------|--------------------------------------------------------------|
+| Version           | `GetVersion`                                                 |
+| Class Operations  | `FindClass`, `GetSuperclass`, `IsAssignableFrom`             |
+| Exceptions        | `Throw`, `ThrowNew`, `ExceptionOccurred`, `ExceptionClear`   |
+| References        | `NewGlobalRef`, `DeleteGlobalRef`, `NewLocalRef`             |
+| Object Operations | `AllocObject`, `NewObject`, `GetObjectClass`, `IsInstanceOf` |
+| Field Access      | `GetFieldID`, `Get/Set<Type>Field`, `GetStaticFieldID`       |
+| Method Invocation | `GetMethodID`, `Call<Type>Method`, `CallStatic<Type>Method`  |
+| String Operations | `NewString`, `GetStringLength`, `GetStringChars`             |
+| Array Operations  | `NewArray`, `GetArrayLength`, `Get/Set<Type>ArrayRegion`     |
+| Registration      | `RegisterNatives`, `UnregisterNatives`                       |
+| Monitors          | `MonitorEnter`, `MonitorExit`                                |
--- a/docs/native-ffi.md
+++ b/docs/native-ffi.md
@ -0,0 +1,125 @@
+# Native Methods and FFI System
+
+**Location**: `crates/core/src/thread.rs`, `crates/core/src/native/`
+
+## Library Loading
+
+Native libraries are loaded dynamically using `libloading`:
+
+- **Location**: `crates/core/src/main.rs`
+- **Supported platforms**: Windows (.dll), Linux (.so)
+- **Libraries loaded**:
+  - `roast_vm` - VM-specific native methods
+  - `jvm` - Java virtual machine standard library
+  - `java` - Java standard library
+
+Libraries are registered with the VM via `Vm::load_native_library()` and stored in `native_libraries: Arc<RwLock<Vec<(String, Library)>>>`.
+
+## Native Method Registration
+
+Native methods are registered through JNI's `RegisterNatives` function:
+
+- **Location**: `crates/core/src/native/jni.rs`
+- **Process**:
+  1. Java code calls `RegisterNatives()` with method names, signatures, and function pointers
+  2. VM generates JNI-formatted symbol names using `generate_jni_method_name()`
+  3. Function pointers are stored in `Vm::native_methods: DashMap<String, *const c_void>`
+  4. Filters prevent registration of certain methods (e.g., Thread operations)
+
+## Native Method Dispatch
+
+When a native method is invoked:
+
+- **Detection**: `MethodData::ACC_NATIVE` flag identifies native methods
+- **Location**: `crates/core/src/thread.rs`
+- **Flow**:
+  1. `execute_method()` checks if method has `ACC_NATIVE` flag
+  2. If static, adds class reference to args; otherwise adds instance reference
+  3. Calls `invoke_native()` with method reference and arguments
+
+## FFI System - libffi Integration
+
+libffi is used to call native functions with correct calling conventions:
+
+### Call Interface Building
+
+```rust
+fn build_cif(&self) -> Cif {
+    let mut args = vec![
+        Type::pointer(), // JNIEnv*
+        Type::pointer(), // jclass/jobject
+    ];
+    for v in self.desc.parameters {
+        args.push(v.into())
+    }
+    let return_type = ...;
+    Builder::new().args(args).res(return_type).into_cif()
+}
+```
+
+- Constructs a Call Interface (Cif) from method signature
+- Maps Java types to FFI types
+- Always adds JNIEnv* and jclass/jobject as first two parameters
+
+### Argument Marshalling
+
+```rust
+fn build_args(params, storage, jnienv) -> Vec<Arg>
+```
+
+- Marshals Java values to native format
+- Converts references to object IDs (u32)
+- Boxes primitives for FFI passing
+- Stores all values in a temporary vector
+
+### Function Invocation
+
+```rust
+let cp = CodePtr::from_ptr(p);
+cif.call::<ReturnType>(cp, built_args.as_ref());
+```
+
+- Converts function pointer to CodePtr
+- Calls through libffi with correct return type handling
+
+## Type Mapping
+
+| Java Type | FFI Type |
+|-----------|----------|
+| byte | i8 |
+| char | u16 |
+| short | i16 |
+| int | i32 |
+| long | i64 |
+| float | f32 |
+| double | f64 |
+| boolean | i8 |
+| Object/Array | pointer |
+
+## JNI Function Table
+
+A complete JNI function table is created and passed to native code:
+
+- **Location**: `crates/core/src/native/jni.rs`
+- **Implementation**:
+  - 250+ JNI functions defined as unsafe extern "system" functions
+  - Covers class operations, method invocation, field access, array operations, string handling
+  - Many functions are stubs returning `todo!()` for unimplemented features
+
+## Unsafe Support (sun.misc.Unsafe)
+
+Low-level unsafe operations are tracked:
+
+- **Location**: `crates/core/src/native/unsafe.rs`
+- **Features**:
+  - Field offset registry mapping to class/field pairs
+  - Off-heap memory allocation tracking
+  - Base offset constant: `0x1_0000_0000`
+
+## Key Implementation Characteristics
+
+- **Thread-safe**: All structures use `DashMap`, `RwLock`, or `Mutex`
+- **JNI environment**: Created per-thread, stored in `VmThread::jni_env`
+- **Symbol lookup**: Two-pass search (without params, then with params)
+- **Error handling**: Returns `VmError::NativeError` if symbols not found
+- **Tracking**: Maintains statistics on library resolution counts
--- a/docs/object-management.md
+++ b/docs/object-management.md
@ -0,0 +1,95 @@
+# Object Management
+
+**Location**: `crates/core/src/objects/`
+
+## Object Representation
+
+Java objects are represented through a multi-layered abstraction:
+
+```rust
+pub struct Object {
+    pub id: u32,                           // Unique identifier
+    pub class: Arc<RuntimeClass>,          // Runtime class reference
+    pub fields: DashMap<String, Value>,    // Concurrent field storage
+}
+
+pub type ObjectReference = Arc<Mutex<Object>>;
+```
+
+- **Objects**: Contain a unique ID (u32), runtime class reference, and field storage (DashMap)
+- **ObjectReference**: `Arc<Mutex<Object>>` - reference-counted, thread-safe smart pointer
+- **Value wrapper**: The `Value` enum encapsulates both primitives and references for operand stack/local variable storage
+
+## Array Management
+
+Arrays are type-safe with separate variants for primitives and objects:
+
+```rust
+pub enum ArrayReference {
+    Int(Arc<Mutex<Array<jint>>>),
+    Byte(Arc<Mutex<Array<jbyte>>>),
+    Short(Arc<Mutex<Array<jshort>>>),
+    Long(Arc<Mutex<Array<jlong>>>),
+    Float(Arc<Mutex<Array<jfloat>>>),
+    Double(Arc<Mutex<Array<jdouble>>>),
+    Char(Arc<Mutex<Array<jchar>>>),
+    Boolean(Arc<Mutex<Array<jboolean>>>),
+    Object(Arc<Mutex<Array<Option<ReferenceKind>>>>),
+}
+```
+
+- **Primitive arrays**: Int, Byte, Short, Long, Float, Double, Char, Boolean
+- **Object arrays**: Can hold references to other objects
+- **Array structure**: Each array wraps a boxed slice `Box<[T]>` with id, class, and backing storage
+- **Thread-safe**: All arrays use `Arc<Mutex<Array<T>>>` for concurrent access
+
+## Allocation Strategy
+
+Allocation is centralized in **ObjectManager**:
+
+- **Object allocation**: `new_object()` generates unique IDs via atomic counter and stores references in HashMap
+- **Array allocation**: Separate methods for primitive arrays (`new_primitive_array()`) and object arrays (`new_object_array()`)
+- **String interning**: `new_string()` creates UTF-16 encoded strings with automatic interning via string pool
+- **Memory tracking**: `bytes_in_use()` calculates total heap usage across all objects/arrays
+
+All allocated objects are registered in `objects: HashMap<u32, ReferenceKind>` for global access.
+
+## Reference Handling
+
+Two-level reference system:
+
+1. **ReferenceKind enum**: Distinguishes between `ObjectReference` and `ArrayReference`
+2. **Reference type alias**: `Option<ReferenceKind>` (None = null)
+3. **Conversion methods**: Safe conversions with `try_into_object_reference()` and `try_into_array_reference()`
+
+## Memory Management
+
+**No explicit garbage collection** - relies on Rust's reference counting:
+
+- Arc ensures objects live as long as references exist
+- Mutex provides thread-safe field/element access
+- Shallow cloning for `clone()` operations (copies references, not objects)
+- Array copy operations handle both primitive and object types with bounds checking
+
+## Object Synchronization
+
+Monitor-based concurrency for synchronized operations:
+
+```rust
+pub struct Monitor {
+    owner: Option<ThreadId>,
+    entry_count: u32,
+    condition: Condvar,
+    mutex: Mutex<()>,
+}
+```
+
+- **Operations**: `monitor_enter()`, `monitor_exit()`, `wait()`, `notify_one()`, `notify_all()`
+- **Wait semantics**: Full support for Java-style wait/notify with timeout
+- **Reentrant**: Same thread can enter multiple times (tracked by entry_count)
+
+## Special Features
+
+- **String handling**: UTF-16 LE encoding with automatic String object creation
+- **Reflection support**: Methods to create Constructor, Method, MethodHandle objects
+- **Class mirrors**: Every class has an associated mirror object (java/lang/Class)
--- a/docs/roast-vm-sys.md
+++ b/docs/roast-vm-sys.md
@ -0,0 +1,87 @@
+# roast-vm-sys Crate
+
+**Location**: `crates/roast-vm-sys/`
+
+A cdylib crate that exports native method implementations callable from Java via JNI.
+
+## Overview
+
+**roast-vm-sys** is a JNI wrapper crate that exposes the roast-vm-core runtime to Java. It's compiled as a C dynamic library (cdylib) named `roast_vm`.
+
+## Exported Native Methods
+
+The crate exports 40+ JNI native functions via `#[no_mangle] extern "system"` declarations.
+
+### By Module
+
+| Module | Functions |
+|--------|-----------|
+| `thread.rs` | `Thread.currentThread()`, `Thread.start0()`, `Thread.setPriority0()` |
+| `object.rs` | `Object.hashCode()`, `Object.clone()`, `Object.notify()`, `Object.notifyAll()`, `Object.wait()` |
+| `class.rs` | `Class.forName0()`, `Class.getPrimitiveClass()`, `Class.getDeclaredConstructors0()` |
+| `reflection.rs` | `Reflection.getCallerClass()` |
+| `reflect/array.rs` | `Array.newArray()` |
+| `string.rs` | `String.intern()` |
+| `system.rs` | `System.arraycopy()`, `System.nanoTime()` |
+| `runtime.rs` | `Runtime.maxMemory()`, `Runtime.availableProcessors()` |
+| `misc_unsafe.rs` | `Unsafe` field offsets, volatile read/write, memory allocation |
+| `file_output_stream.rs` | `FileOutputStream.writeBytes()` |
+| `system_props.rs` | `vmProperties()` - VM identity (version 0.1.0, vendor "infernap12") |
+| `CDS.rs` | Class Data Sharing stubs |
+| `signal.rs` | `Signal.handle0()` stub |
+| `scoped_memory_access.rs` | `ScopedMemoryAccess` registration |
+
+## Bridge Pattern
+
+Each native function follows this pattern:
+
+1. **Extract VmThread** from `JNIEnv.reserved0` using `get_thread()` helper
+2. **Resolve References** - Convert JNI handles (jobject) to internal references:
+   - `resolve_object()` - gets ObjectReference
+   - `resolve_array()` - gets ArrayReference
+   - `resolve_reference()` - gets generic ReferenceKind
+3. **Perform Operation** via core VM APIs
+4. **Return Result** in JNI-compatible format
+
+## Example Native Implementation
+
+```rust
+#[unsafe(no_mangle)]
+pub extern "system" fn Java_org_example_MockIO_print(
+    env: JNIEnv,
+    _jclass: JClass,
+    input: JString,
+) {
+    unsafe {
+        let input: String = env.get_string_unchecked(&input)
+            .expect("Couldn't get java string!")
+            .into();
+        std::io::stdout().write_all(input.as_bytes()).ok();
+    }
+}
+```
+
+## File Structure
+
+17 modules organized by Java class:
+
+- `lib.rs` - Core helpers, test functions (`MockIO.print()`, `Main.getTime()`)
+- `runtime.rs`, `thread.rs`, `class.rs` - Core VM operations
+- `object.rs`, `string.rs`, `reflection.rs`, `reflect/` - Object/class introspection
+- `system.rs`, `file_output_stream.rs` - System I/O
+- `misc_unsafe.rs` - Unsafe memory operations (largest implementation, ~626 lines)
+- `CDS.rs`, `signal.rs`, `system_props.rs`, `scoped_memory_access.rs` - Stubs/properties
+
+## GC Interaction
+
+Native methods access the garbage collector via:
+- `thread.gc.read()` / `thread.gc.write()` for object access
+- Object creation, cloning, and array operations go through GC
+- Field access via `thread.gc` or direct field references
+
+## Error Handling
+
+Mixed approach:
+- Some methods panic on errors
+- Some return null/default values
+- TODO comments indicate incomplete exception throwing