Repository layout

Table of Contents

• The monorepo
• How configuration is split
• Git hooks

The monorepo

python/                 python library for machine learning
tooling/                shared source — built once per dataset
    include/  src/        C++ analysis library + headers
    mains/                one main_*.cpp per binary
    gpu/                  CUDA timestamp-sort kernel (libgpuaccel.so), dlopen'd at runtime
analysis/<dataset>/
    config/Constants.cpp  per-dataset config overrides (run numbers, thresholds, gates, ...)
    control/              per-reaction TOML control files (gas, beam, target, detector)
    bin/                  built binaries (git-ignored; produced only by the internal Makefile)
    build/                object files + libmusic.a (git-ignored; internal Makefile)
    root_files/           event ROOT files produced by the pipeline
    sim_root_files/       simulated beam/trace ROOT files
    plots/                output figures
flake.nix               Nix dev shells + package derivation, one per dataset
tooling/Makefile        internal build recipe, driven by the flake (never run directly)

How configuration is split

The configuration is split in two. The struct definition and the tooling-wide defaults live in tooling/include/Constants.hpp and tooling/src/Constants.cpp. Each dataset then overrides only the fields it cares about, in analysis/<dataset>/config/Constants.cpp, which is compiled into that dataset's binaries.

The control/ TOML files are the only other per-dataset input. The tooling itself is identical across datasets — which is what makes a single API reference meaningful for all of them.

The TOMLs (gas fill, beam species and energy, target, detector response) are also what generate simulated data for the relevant reaction, via Remix-MUSIC-Sim.

Git hooks

.githooks/pre-commit runs clang-format (LLVM style) on staged C/C++ files. Enable it once with:

git config core.hooksPath .githooks