Schema

Overview

Library for serializing C# types to/from binary. Provides a Roslyn generator that automatically implements read/write logic.

Warning: The design of this library is still in flux, so anticipate making changes when upgrading to future versions.

Credits

• @connorhaigh, whose SubstreamSharp library was pulled in for reading substreams.
• @jefffhaynes, whose BinarySerializer attribute library inspired the schema attributes for configuring how binary data is read.
• @Kermalis, whose EndianBinaryIO library inspired Span-based performance improvements.
• @Sergio0694, whose BinaryPack generator inspired the schema source generator used to generate read/write methods.

Usage

Implementing binary schema classes

To write a binary schema class, you must first have it implement the IBinarySerializable or IBinaryDeserializable interfaces (or IBinaryConvertible if you need both).

Then, based on how complicated your schema class is, you can either choose to automatically or manually implement Read()/Write() methods.

Automatically

For most schema classes, you should be able to use the automatic code generator.

All you have to do is annotate the schema class with the [BinarySchema] attribute and mark it as partial; this will flag to the generator that it should implement read/write methods for this class. It will then look into all fields/properties in the schema class, and attempt to implement read/write logic in the same order that the fields/properties appear.

Any nested schema classes will be automatically read/written as expected.

Some types require additional attributes in order to clarify any ambiguity. For example, booleans require a [IntegerFormat(SchemaIntegerType.###)] attribute to know what type of integer to read, which it will then compare to 0.

Any readonly primitives will treated as assertions, which is useful for validating things like magic text or padding.

Manually

For complicated schema classes, such as ones that use decompression logic or pointers, you'll need to implement the read/write logic manually.

Specifically, you'll need to implement both a Read(IBinaryReader br) and Write(IBinaryWriter bw) method. The SchemaBinaryReader and SchemaBinaryWriter classes provide many helpful methods for reading/writing a number of different primitive formats, including basic ones such as byte/int/float, but also more complex/unique ones such as Half (two-byte float) and un16 (unsigned normalized 16-bit float).

Similar to the automatic process, you can nest schema classes and manually read/write them by calling their Read()/Write() methods. This can allow you to automatically generate subsections, so only the most complex logic needs to be manually written.

How to use a binary schema class

To convert a given schema class to or from binary, simply instantiate an SchemaBinaryReader or SchemaBinaryWriter and pass it into the corresponding Read() or Write() methods in the schema class.

Supported Attributes

The following attributes are currently supported in this library when automatically generating code. Some attributes are only used at read or write time—these are prefixed with an R or W respectively.

Warning: These names are not final, so they may change in future versions.

Align

Specifies how a field or property's offset (relative to the start of the stream) should be aligned when reading/writing. If misaligned, the SchemaBinaryReader/SchemaBinaryWriter will automatically insert the remaining bytes of padding. For example, [Align(4)] would force a field/property's starting offset to be a multiple of 4 (0, 4, 8, 12, 16, etc.).

[Align(4)]publicintalignedField;[Align(4)]publicintAlignedProperty{get;set;}

Endianness

Forces a type, field, or property to be read/written with a given endianness (big-endian or little-endian). Tracked via a stack within the SchemaBinaryReader/SchemaBinaryWriter. If unspecified, will use whatever endianness was last specified in the stack (or the system endianness by default).

[BinarySchema][Endianness(Endianness.BigEndian)]publicpartialclassBigEndianType:IBinaryConvertible{
...[Endianness(Endianness.LittleEndian)]publicintLittleEndianProperty{get;set;}
...}

IfBoolean/RIfBoolean

Marks that a nullable field or property will only be read/written if some other boolean field or property is true.

[IntegerFormat(SchemaIntegerType.BYTE)]publicboolHasValue{get;set;}[RIfBoolean(nameof(this.HasValue))]publicint?Value{get;set;}

IChildOf<TParent>

This pseudo-attribute marks a type as a "child" of some "parent" type—that it is contained as one of the members of the "parent type"—and passes the parent down to the child so it can be referenced in Schema logic.

Used by having the child type implement the IChildOf<TParent> interface, where TParent stores the child type in a field/property or as a member of a sequence (array/list):

[BinarySchema]publicpartialclassChildType:IBinaryConvertible,IChildOf<ParentType>{publicParentTypeParent{get;set;}
...}

Below is a simple example where a boolean from the parent is used to decide when to read a value in the child:

[BinarySchema]publicpartialclassParentType:IBinaryConvertible{[IntegerFormat(SchemaIntegerType.BYTE)]publicboolChildHasSomeField{get;set;}publicChildTypeChild{get;}=new();}[BinarySchema]publicpartialclassChildType:IBinaryConvertible,IChildOf<ParentType>{// This is automatically skipped while reading/writing.publicParentTypeParent{get;set;}[Skip]privateboolHasSomeField=>Parent.ChildHasSomeField;[RIfBoolean(nameof(HasSomeField))]publicint?SomeField{get;set;}}

Skip

Designates that a field or property should be skipped while reading/writing.

Note: IChildOf<TParent>.Parent is automatically skipped.

[Skip]publicintskippedField;[Skip]publicintSkippedProperty{get;set;}

This can be used to encapsulate logic within properties, such as in the following examples:

1. Value conversion

[StringLengthSource(4)]publicstringMagic{get;set;}[Skip]publicMagicTypeType=>this.Magic switch{
"IMGE" => MagicType.IMAGE,
"SOND" => MagicType.SOUND,
"TEXT" => MagicType.TEXT,};

2. "Switch" cases

[NullTerminatedString]publicstringMagic{get;set;}[Skip]publicISection?Section=>this.imageSection_ ??this.soundSection_ ??this.textSection_;[Skip]privateboolIsImage_=>this.Magic == "IMAGE";[Skip]privateboolIsSound_=>this.Magic == "SOUND";[Skip]privateboolIsText_=>this.Magic == "TEXT";[RIfBoolean(nameof(this.IsImage))]
private ImageSection?imageSection_{get;set;}[RIfBoolean(nameof(this.IsSound_))]privateSoundSection?soundSection_{get;set;}[RIfBoolean(nameof(this.IsText_))]
private TextSection?textSection_{get;set;}

Numbers/Enums

NumberFormat

TODO

IntegerFormat

TODO

Strings

Note: At the moment, only ASCII is fully supported.

StringLengthSource/RStringLengthSource

Designates the length of a string field or property via one of three cases.

Note: Any trailing null terminators will be ignored at read time.

1. Constant length

If a constant is passed into StringLengthSource, that many characters will be read/written.

[StringLengthSource(8)]publicstringText{get;set;}

2. Preceding value

If a SchemaIntegerType is passed into StringLengthSource, an integer of that type will first be read and used as the length of the string, or the length of the string will first be written before writing the string itself.

[StringLengthSource(SchemaIntegerType.BYTE)]publicstringTextWithByteLength{get;set;}

3. Another field or property

If the name of another field or property is passed into RStringLengthSource, that other value will be used as the length of the string when reading.

publicbyteTextLength{get;set;}[RStringLengthSource(nameof(this.TextLength))]publicstringText{get;set;}

NullTerminatedString

Designates that a string field or property will be read until a null terminator is reached, and written with a null terminator affixed to the end.

[NullTerminatedString]publicstringText{get;set;}

Sequences

Note: "Sequence" is the term used within Schema to refer to an array/list of elements.

SequenceLengthSource/RSequenceLengthSource

TODO

RSequenceUntilEndOfStreamAttribute

TODO

Pointers/Memory

TODO

WPointerTo

TODO

WSizeOfMemberInBytes

TODO