Skip to content

oracles.md

Latest commit

 

History

History
104 lines (81 loc) · 3.76 KB

oracles.md

File metadata and controls

104 lines (81 loc) · 3.76 KB
Raw

Building the oracles

The parity suite compares against C-built oracle binaries. They live in References/ (git-ignored, GPL-3.0 — built from GPL reference material, used only as local test oracles, never shipped).

!!! note "Credits" The RAHasher CLI we test against is written by LeXofLeviafan (RALibretro lineage). His project is the reference oracle for this suite — byte-identical output between our port and his binaries is asserted on every run.

Which oracle is used

The harness resolves, in order:

  1. RASHARP_ORACLE env var;
  2. References\rcheevos-12.4.0\bin64\RAHasher.exe — built from rcheevos 12.4.0 (the Part II source of truth);
  3. References\RAHasher-1.8.3\bin64\RAHasher.exe — built from the pinned 1.8.3 sources;
  4. References\RAHasher.exe — any other 1.8.3 binary (a legacy build that only accepts numeric console ids; the harness adapts).

Prerequisites (Windows)

  • MSYS2 (/c/msys64) with make, sh, and the MinGW-w64 toolchain (gcc/g++ 16.x).
  • miniz and libchdr sources under References/RAHasher-1.8.3/src/ (cloned; libchdr pinned to v0.3.0 with its vendored deps lzma-25.01/zstd-1.5.7/miniz-3.1.1).

The TMP problem (why the wrapper scripts exist)

The MSYS2 sh clears TMP/TEMP for child processes, so MinGW gcc falls back to C:\WINDOWS and fails with "Cannot create temporary file". The ccw-gcc.sh / ccw-g++.sh wrappers set TMP=C:\msys64\tmp before exec-ing the real compiler:

#!/bin/sh
export TMP='C:\msys64\tmp'
export TEMP='C:\msys64\tmp'
exec /c/msys64/mingw64/bin/gcc.exe "$@"

Building the 1.8.3 oracle

cd References/RAHasher-1.8.3
export PATH="/c/msys64/usr/bin:/c/msys64/mingw64/bin:$PATH"
make -f Makefile.RAHasher HAVE_CHD=1 CC=./ccw-gcc.sh CXX=./ccw-g++.sh
# → bin64/RAHasher.exe

Notes on the reference-tree adaptations (build-only, never shipped):

  • src/rcheevos/ is the 40d916d snapshot; src/miniz is 3.1.0 (master's API drifted — 3.1.0 restores mz_alloc_func/tinfl_decompressor); src/Util.cpp includes <miniz.h> before <miniz_zip.h> (the header split);
  • Makefile.common's CHD_OBJS were adapted to libchdr v0.3.0's dep layout (lzma-25.01/src/LzmaDec.c, zstd-1.5.7/zstddeclib.c, codec files split per-module) and codec_zlib's miniz include is shimmed to the CLI's miniz to avoid duplicate symbols;
  • src/RA_BuildVer.h is a handwritten 1.8.3 header (the git-generated one needs the RAInterface submodule).

Building the 12.4.0 oracle

cd References/rcheevos-12.4.0
export PATH="/c/msys64/usr/bin:/c/msys64/mingw64/bin:$PATH"
make -f Makefile.oracle CC=./ccw-gcc.sh CXX=./ccw-g++.sh
# → bin64/RAHasher.exe

Makefile.oracle compiles the RAHasher 1.8.3 CLI stack (RAHasher.cpp, Util.cpp, Hash3DS.cpp, HashCHD.cpp, Logger.cpp, sha256.c) against rcheevos 12.4.0's src/rhash, reusing the same miniz 3.1.0 and libchdr v0.3.0 trees for zip pre-load and CHD support.

Why CHD needs libchdr

rc_hash's default cdreader has no CHD support — the CHD track reader is HashCHD.cpp (from the RAHasher tree, GPL reference), which plugs into rc_hash_cdreader_t and reads hunks via libchdr. Both oracles therefore link libchdr v0.3.0 + vendored deps (zstd single-file decoder, LZMA decoder, miniz for the zlib codec).

Verification

After building, smoke-test:

bin64/RAHasher.exe GB <any-file>        # hash + exit 0
bin64/RAHasher.exe '?' <any-file>       # iterate mode
bin64/RAHasher.exe PS1 <game>.cue       # disc
bin64/RAHasher.exe 12 <game>.chd        # CHD (numeric id also works)
bin64/RAHasher.exe -s <sysdir> 62 <file>.cia   # 3DS

Then run dotnet test --filter FullyQualifiedName~Parity — all 90 cases must pass byte-identically.