RogueForge

Redux

Starting over

For twenty years RogueForge restored Rogue from whatever source survived. In 2026 the Rogue Restoration Project went back to first principles.

The first restoration

RogueForge began with UltraRogue. Herb Chong sent his source in 1992, and over the next few years it was brought up to ANSI C and given new code to save and restore a game. XRogue followed in 1996, from Robert Pietkivitch, and Super-Rogue in 2004, from Robert Kindelberger. A SourceForge project opened in 2000, then the rogueforge.net site, then a Trac and Subversion server that carried the work for two decades.

Each of those restorations started from source somebody had kept. That was the only way to do it, and it produced games people could play. It also left questions nobody could answer. The Rogue 3.6 sources that survive came from half a dozen sites and dates, between January 1982 and June 1984, and they disagree. Which one is the game Berkeley released? Were any of them compiled unchanged? For Rogue 4.22 and 5.3 there was almost nothing to start from.

What survived

Two kinds of evidence survive, and the restart uses both.

The first is source. Three trees are likely to be close to the versions they claim to be: Rogue 3.6, from the several site copies; Rogue 5.2, from the University of Illinois FTP archive; and Rogue 5.4, as Ken Arnold later released it. None is guaranteed to be exactly what Berkeley compiled, but each is a strong starting point, and the versions between them can be worked out by moving forward or back from the nearest one.

The second is the programs themselves, with provenance better than any source tree's. Copies of Rogue turn up in Berkeley's BSD distribution trees, preserved on the CSRG archive CD-ROMs and at The Unix Heritage Society:

  • Rogue 3.4 in 2.8BSD, built on 21 May 1981 for the PDP-11.
  • Rogue 3.6 in 4.1BSD for the VAX, and twice more in 2.9BSD for the PDP-11.
  • Rogue 4.22 in 4.1aBSD, March 1982.
  • Rogue 5.2 on the last 4.1BSD tape, July 1982.
  • Rogue 5.3 in 4.2BSD, as both a distributed object file and the installed game.
  • Rogue 5.4 in the 4.3BSD source tree, as a stripped object file, main.obj, with only its machine-dependent code as source.

Every one of these is a fixed point. Start from the nearest source tree, change it until the period compiler turns it into exactly this file, and you have the source as it was, whatever the surviving copies say.

The new standard

That became the rule for the restart: a version counts as restored when its rebuilt source compiles, with the original toolchain, to a file whose checksum matches the one on the tape. Reading the disassembly and agreeing with it is not enough. As the 3.4 notes put it, reading disassembly convinces; only recompiling proves.

The standard is strict on purpose. In March 2026 a first pass over Rogue 3.4 marked all 145 of its functions verified by reading. When the source was finally compiled in June, the byte-for-byte comparison found dozens of places where it was wrong. Every result on this site comes from the second kind of check.

The work, in brief

  • March 2026. Ghidra and radare2 are taught enough PDP-11 and VAX to decompile every version. The Rogue 5.4 main.obj is the first to come close, on 2 April.
  • June and July. Rogue 3.4 matches on 3 July, after the 2.8BSD compiler is rebuilt and proven against itself. Rogue 3.6 follows on both machines, then 5.3, 5.2 and 4.22.
  • November. The rebuilt sources become the base for new Linux ports, with fixes for the bugs a modern system exposes.

The method describes each step, and the findings collect what the binaries revealed about the code their authors wrote.

What we don't do

  • We are not writing a new roguelike. Where a version's gameplay differs from its parent, as UltraRogue's does, that comes from its original author, not from us.
  • We don't round results up. If a rebuild matches everything but a symbol table, the page for that version says so.
  • Fixes for modern systems go in the ports, as separate changes, never in the rebuilt source.