FujiNet Proposes One Self-Describing Protocol for All Its Platforms

On 6 October 2025 the FujiNet project published FEP-004: FujiNet Protocol, a draft by FozzTexx (updated to version 1.1 on 5 November 2025). It proposes a new packet format to replace the different command formats each FujiNet platform has grown. It also gives the clearest look yet at where the hardware is heading.

The problem

FujiNet started on the Atari 8-bit SIO bus and has since been ported to the Apple II, Coleco ADAM, Commodore, CoCo and RS-232 machines. FozzTexx's motivation section explains that each platform implements the protocol slightly differently. Command handlers read their own parameters in custom ways, so a bug fix or feature added for one machine rarely reaches the others without being ported, tested and maintained separately. The root cause is that the old packets have fixed headers with context-dependent payloads: you can't decode a packet without knowing in advance what that particular command expects.

A universal FujiNet

The proposal ties this to a new architecture modelled on the RS-232 FujiNet. Future devices are expected to use a Raspberry Pi RP2040 or RP2350 as the physical bus interface, with an ESP32, or a Linux, Windows or macOS host, handling device emulation and networking. Because the bus side and the device logic may be on different processors, packets must be fast to serialise and parse.

What FEP-004 proposes

  • A fixed header: target device, command ID, 16-bit data length, a simple checksum and a data descriptor byte.
  • Self-describing parameters: after surveying every FujiNet command, the author found only eight combinations of the old "aux" arguments. A 3-bit descriptor value covers them all, from no fields up to four bytes, two 16-bit words or one 32-bit value. Extra descriptor bytes can be chained, and an optional binary blob follows.
  • Little-endian throughout. The draft says plainly that "little-endian is superior to big-endian", which suits the 6502.
  • SLIP framing (0xC0 END, 0xDB ESC) to mark packet boundaries. This replaces the out-of-band command pin of the old protocol, which doesn't exist on networks or USB virtual serial ports.

The aim is that packets can be decoded generically into C/C++ structures, with common handling in a base class and device-specific code in subclasses. It also makes protocol-inspection tools possible and lets new fields be added without breaking old code. The design is still meant to be implementable directly on a 6502, 8080 or other 8-bit CPU for simple builds without the ARM chip.

Still open

The draft lists unresolved questions. How should replies to the legacy computer be framed? How can a machine with no serial buffer or interrupts be told that data is waiting, without polling? How should the host talk to the RP2nnn itself, for resets and firmware uploads? And how should FujiNet behave when no RP2nnn is present?

Why it matters

For Atari users, FujiNet is the most important modern peripheral, and this proposal sets out how its next hardware generation and firmware might be organised. The FujiNet Enhancement Proposal series (FEP-001 on URL handling, February 2025; FEP-002 on lobby leaderboards, May 2025; FEP-003 on NetSIO, April 2025) also shows the project now writing its designs down as specifications that anyone can read and comment on, on the FujiNet wiki.

Credits