Jump to content
Setting up an AI agent? ReverseEngineering.wiki skill (works from the command line, no configuration): https://reverseengineering.wiki/skill/rewiki/SKILL.md · setup for every client, including MCP: https://reverseengineering.wiki/llms.txt · REWiki:Connect

Format:U8 archive

From ReverseEngineering.wiki
Format
Magic 55 AA 38 2D
Extensions .arc, .szs, .app
Endianness big
On this wiki Facts · Code · Notes · Topics

Directory archive of Nintendo Wii and GameCube-era games (.arc, .app channel contents, often Yaz0-compressed as .szs): the magic 55 AA 38 2D, a 32-byte big-endian header, a table of 12-byte nodes (file or directory) followed by a string table, and the file data.

Layout

All integers are big-endian.

Header (32 bytes): magic 55 AA 38 2D, node table offset (uint32, normally 0x20), node table size (uint32: nodes plus string table), data offset (uint32), 16 reserved bytes.

Node (12 bytes): kind (1 byte: 0 file, 1 directory), name offset (3 bytes, big-endian, into the string table), offset (uint32), length (uint32).

  • Files: offset is the position of the data (from the start of the archive file in the decomp-toolkit reader's terms: relative to the header data offset field as documented in its source comment), length the size in bytes.
  • Directories: offset is the index of the parent node, length is the index one past the last node inside the directory. The root is node 0, a directory with offset 0, and its length is the total node count; the string table starts right after the nodes.

Names are NUL-terminated strings in the string table. A directory's contents are the nodes between its own index + 1 and its length, in tree order.

Identification

Files start with 55 AA 38 2D. A Yaz0 wrapper (Format:Yaz0) around it is common (.szs); decompress first. Format:RARC archive is the JSystem alternative used by other first-party GameCube games.

Notes

decomp-toolkit's doc comment calls a file node's offset relative to the header's data offset field, but its virtual file system opens the file at [offset, offset + length) of the archive itself (src/vfs/u8_arc.rs lines 28-30, commit 0645e16511), so treat the offset as measured from the start of the archive. See Fact:U8 archive/behavior/File node offset is from the archive start. Unverified: confirm on a real archive with a non-zero gap between the string table and the data offset.

noclip.website (MIT, src/rres/u8.ts at fd973ff240233c63883750a557c62baf6ff2a6c7) agrees: it opens a file at [offset, offset + size) of the archive buffer, finds the string table at the node table offset + 12 * the root's end index, ignores the header's data offset, and treats a directory's children as the nodes up to its end index (see Fact:U8 archive/behavior/noclip U8 reader checks and directory walk). It takes the name offset from the low 24 bits of the node's first word. That settles the base of file offsets on the two readers read; the gap case above remains untested on a real file.

Code & repositories

See linked repositories.

  • https://github.com/encounter/decomp-toolkit @ 0645e16511814920ffb05b25615dd71f1af73e4c · license: MIT OR Apache-2.0 (declared by the contributor for the files used; GitHub reports Apache-2.0 for the repository)
    u8_arc.rs reads and writes U8 archives (u8 list / extract commands)
    Tested on: source read, not run: read source only · checked 2026-10-18; last activity 2026-09-10; Note: the latest release ships binaries; the wiki links only the source, never the binaries
  • https://github.com/magcius/noclip.website @ fd973ff240233c63883750a557c62baf6ff2a6c7 · license: MIT (declared by the contributor for the files used; GitHub reports MIT (read from LICENSE; GitHub's detector did not classify it) for the repository)
    src/gx/gx_enum.ts, gx_texture.ts, src/Common/JSYSTEM/JUTTexture.ts, src/rres/brres.ts and u8.ts: TypeScript parsers for GX textures, BTI, BRRES and U8
    Tested on: source read, not run · checked 2026-10-19; last activity 2026-10-19