Building

Table of Contents

• Dependencies
• Building the binaries
• Editing
• Building this documentation

Dependencies

Everything is pinned with Nix; flakes must be enabled. The flake provides ROOT (CUDA build), a C++ toolchain, tomlplusplus, and the shared Analysis-Utilities library.

A CUDA-capable GPU is used for the timestamp sort. Without the CUDA toolchain — either because the build skipped it, or because isLaptop = true in flake.nix — it falls back to CPU.

Building the binaries

nix build               # builds music-tooling-37Cl (the default package)
nix build .#87Rb        # the other dataset
./result/bin/...        # binaries land under ./result/bin

./result is a symlink to the built package's store path.

Warning

There is no make step you run yourself. tooling/Makefile is an internal recipe that the flake drives as make -C tooling. It deliberately does not sit at the repository root, so make cannot be run from there; invoking it by hand would drop binaries into analysis/<dataset>/bin/ instead of keeping them at the repository root.

Editing

nix develop .#37Cl      # default shell is 37Cl; use .#87Rb for the other

This gives the compiler toolchain and include paths so clangd can index the code. It does not build anything.

Note

A built binary still needs the dev-shell environment to find its runtime libraries — libanalysis-utils.so among them — so run it from inside nix develop .#<dataset> rather than a bare shell.

Building this documentation

doxygen and graphviz are already in the dev shells:

nix develop .#37Cl
./scripts/build_docs.sh

Output lands in docs/html/index.html (gitignored), with any parse warnings in docs/doxygen-warnings.log. nix build .#docs produces the same site as a store path, which is what the published site is served from.

The reference is dataset-independent: it documents tooling/, which every dataset shares, so there is one docs output rather than one per dataset.