Library for reading Java class files.
$ dotnet add package IKVM.ByteCodeProvides a Java class file parser, reader and writer implementation used by the IKVM project.
The core of the project contains the various abstractions that are used throughout. These abstractions are loosely modeled on the System.Reflection.Metadata namespace. The design is low allocation.
The IKVM.ByteCode.Decoding namespace contains specialized structures for reading data from the class file specification. The main entry point of this namespace will be ClassFile which can be used to read a Java class file from a variety of sources. ClassFile itself holds a reference to the underlying original memory that was read from and parses the minimal amount possible depending on the operation. ClassFile implements IDisposable for releasing this memory.
ClassFile's companion structures are field-only record types. This allows directly taking a ref to a item in a collection or a field.
Constant values are left unresolved in this structure just like System.Reflection.Metadata token handles. It is the users responsibilty to obtain a ConstantHandle from one of the data structures and refer to the ClassFile.Constants ConstantTable to lookup the ConstantKind, ConstantData or Constant itself.
Each of the field-only record structures can be individually used to parse or encode data. To parse each provides it's own TryMeasure and TryRead methods. TryMeasure does a minimal amount of parsing to determine the size of the data from the initial position. This is helpful due to a Java class file's forward-only nature: you can't know the size of the class without reading each part of the class. TryMeasure thus allows you to measure the given structure without committing to parsing or allocating objects.
TryRead actually reads the data. In both of these cases, is returned if the end of the memory has been reached. Exceptions are thrown for parsing errors involving valid data.
falseEach of the field-only record structures also contains CopyTo and WriteTo. CopyTo processes the structure, along with a IConstantHandleMap, and reemits it to an Encoder. This allows you to copy structures read from one class file into another, either creating the necessary constants in the new class, or reencode it to an output stream using IdentityConstantMap. CopyTo requires acccess to a IConstantHandleMap to read constant data in order to navigate into dynamic components that require constant data to reason about (for instance Attributes). WriteTo is much lighter weight: it simply emits the structure as is without consideration as to the constants. They must already be present in whatever constant table is used with the resulting class file.
Constants are represented in three sets of data structures: *ConstantData, *ConstantHandle and *Constant
*Constant structures represent a full local constant value in native .NET types. For instance, Utf8Constant directly contains the string value. These are used to pass around desired constant values for lookup or insertion. These types are arrainged in a hierarchy implemented through custom operators. For instance one can cast between Constant and Utf8Constant. Casting to Constant encodes the original ConstantKind on the structure, so that casting back to Utf8Constant would fail.
*ConstantHandle structures represent a handle to a location (slot) in a constant table. These are single-field structures, merely used for type safety. Like Constant, casting is allowed between compatible types. These abstract the locality of a constant in a constant table.
*ConstantData structures store the actual constant data that would be present in a constant table. Most constants are simply collections of handles to other constants. However, constants like Utf8ConstantData contain a pointer to the original parsed memory, and is decoded to a .NET string on demand. The base ConstantData structure maps over the original segmented constant data from a class file and parses it only on demand when cast.
Encoding class files or parts of class files in IKVM.ByteCode is similar to System.Reflection.Metadata.Ecma335. A common linkable buffer, BlobBuilder, is provided which is the target of the various builders and encoders. BlobBuilder allows for fast append, and fast enumeration from the beginning, by linked list of segments of memory. BlobBuilders can be linked to the end of existing BlobBuilders. Assembling Java class files then is allocating a BlobBuilder and using one of the various Encoder classes to emit output into it. That output can then be appened together with other segments before being serialized to output.
ClassFormatWriter represents the main operations for emitting class file output data: U1, U2 and U4, as defined by the specification. Each of the encoder structures target a ClassFormatWriter, which can be backed by a Span<byte> which can be allocated in segments from a BlobBuilder. Each Encoder provides a number of methods that have to be invoked in order to write the appropriate components. Encoders can be passed by ref between methods. They maintain minimal state such as a count of emitted elements. The nested structure of a Java class file is encapsulated by passing delegates with the nested encoders.
ClassBuilder makes it easy to encode an entire Java class file to a temporary series of blobs that are then assembled into a final format.