colfer

Colfer

Colfer is a binary serialization format optimized for speed and size.

The project's compiler colf(1) generates source code from schema definitions to marshal and unmarshall data structures.

This is free and unencumbered software released into the public domain. The format is inspired by Protocol Buffers.

Language Support

• C, ISO/IEC 9899:2011 compliant a.k.a. C11, C++ compatible
• Go, a.k.a. golang
• Java, Android compatible
• JavaScript, a.k.a. ECMAScript, NodeJS compatible

Features

• Simple and straightforward in use
• No dependencies other than the core library
• Both faster and smaller than the competition
• Robust against malicious input
• Maximum of 127 fields per data structure
• No support for enumerations
• Framed; suitable for concatenation/streaming

TODO's

• Rust and Python support
• Protocol revision

Use

Download a prebuilt compiler or run go get -u github.com/pascaldekloe/colfer/cmd/colf to make one yourself. Without arguments the command prints its manual.

NAME
	colf — compile Colfer schemas

SYNOPSIS
	colf [-h]
	colf [-vf] [-b directory] [-p package] \
		[-s expression] [-l expression] C [file ...]
	colf [-vf] [-b directory] [-p package] \
		[-s expression] [-l expression] Go [file ...]
	colf [-vf] [-b directory] [-p package] \
		[-x class] [-i interfaces] [-c file] \
		[-s expression] [-l expression] Java [file ...]
	colf [-vf] [-b directory] [-p package] \
		[-s expression] [-l expression] JavaScript [file ...]

DESCRIPTION
	Generates source code from a model definition for one language.
	The file operands specify schema input. Directories are scanned
	for files with the colf extension. When no files are given, then
	the current working directory is used.
	A package definition may be spread over several schema files.
	The directory hierarchy of the input is not relevant for the
	generated code.

OPTIONS
  -b directory
    	Use a base directory for the generated code. (default ".")
  -c file
    	Insert a code snippet from a file.
  -f	Normalize the format of all input schemas on the fly.
  -i interfaces
    	Make all generated classes implement one or more interfaces.
    	Use commas as a list separator.
  -l expression
    	Set the default upper limit for the number of elements in a
    	list. The expression is applied to the target language under
    	the name ColferListMax. (default "64 * 1024")
  -p package
    	Compile to a package prefix.
  -s expression
    	Set the default upper limit for serial byte sizes. The
    	expression is applied to the target language under the name
    	ColferSizeMax. (default "16 * 1024 * 1024")
  -v	Enable verbose reporting to standard error.
  -x class
    	Make all generated classes extend a super class.

EXIT STATUS
	The command exits 0 on succes, 1 on compilation failure and 2
	when invoked without arguments.

EXAMPLES
	Compile ./io.colf with compact limits as C:

		colf -b src -s 2048 -l 96 C io.colf

	Compile ./*.colf with a common parent as Java:

		colf -p com.example.model -x com.example.io.IOBean Java

BUGS
	Report bugs at <https://github.com/pascaldekloe/colfer/issues>.

	Text validation is not part of the marshalling and unmarshalling
	process. C and Go just pass any malformed UTF-8 characters. Java
	and JavaScript replace unmappable content with the '?' character
	(ASCII 63).

SEE ALSO
	protoc(1), flatc(1)

It is recommended to commit the generated source code into the respective version control to preserve build consistency and minimise the need for compiler installations. Alternatively, you may use the Maven plugin.

<plugin>
	<groupId>net.quies.colfer</groupId>
	<artifactId>colfer-maven-plugin</artifactId>
	<version>1.11.2</version>
	<configuration>
		<packagePrefix>com/example</packagePrefix>
	</configuration>
</plugin>

Schema

Data structures are defined in .colf files. The format is quite self-explanatory.

// Package demo offers a demonstration.
// These comment lines will end up in the generated code.
package demo

// Course is the grounds where the game of golf is played.
type course struct {
	ID    uint64
	name  text
	holes []hole
	image binary
	tags  []text
}

type hole struct {
	// Lat is the latitude of the cup.
	lat float64
	// Lon is the longitude of the cup.
	lon float64
	// Par is the difficulty index.
	par uint8
	// Water marks the presence of water.
	water bool
	// Sand marks the presence of sand.
	sand bool
}

See what the generated code looks like in C, Go, Java or JavaScript.

The following table shows how Colfer data types are applied per language.

• † signed representation of unsigned data, i.e. may overflow to negative.
• ‡ range limited to [1 - 2⁵³, 2⁵³ - 1]
• †† timezone not preserved

Lists may contain floating points, text, binaries or data structures.

Security

Colfer is suited for untrusted data sources such as network I/O or bulk streams. Marshalling and unmarshalling comes with built-in size protection to ensure predictable memory consumption. The format prevents memory bombs by design.

The marshaller may not produce malformed output, regardless of the data input. In no event may the unmarshaller read outside the boundaries of a serial. Fuzz testing did not reveal any volnurabilities yet. Computing power is welcome.

Compatibility

Name changes do not affect the serialization format. Deprecated fields should be renamed to clearly discourage their use. For backwards compatibility new fields must be added to the end of colfer structs. Thus the number of fields can be seen as the schema version.

Performance

Colfer aims to be the fastest and the smallest format without compromising on reliability. See the benchmark wiki for a comparison. Suboptimal performance is treated like a bug.