Deploy packages

st build --deploy {cpp,python} builds the library and then assembles a self-contained, copy-and-run package under package/ at the repository root. The package bundles what a downstream project needs (the compiled library, the headers or binding, a default config, and a working example), so it can be lifted out of this repository and dropped into another project without a build tree or environment variables. The directory is regenerated from scratch on every deploy and is git-ignored.

The deploy is platform-specific: it copies the artefacts for the OS it runs on. Run it on Windows to get a Windows package (.dll + .lib), and on Linux to get a Linux package (.so). There is no cross-compilation: to ship both platforms, run the deploy once on each.

st build --deploy cpp        # C/C++ package
st build --deploy python     # Python package
st build --clean --deploy cpp   # clean rebuild, then deploy

Deploy reuses the same artefacts a plain st build produces; it never needs a test target. If the shared library (or, on Windows, its import library) is missing, the deploy fails with an explicit message telling you to build first.

--deploy cpp layout

Targets a C or C++ consumer through the C ABI (SushiTrack/sushitrack_c.h). That header includes SushiTrack/sushitrack_export.hpp, which holds preprocessor macros only, and nothing else from the library, so a C consumer needs no Eigen and no C++ header. The package still carries every header of include/SushiTrack/.

sushitrack_c.h includes that file and does not repeat the macro, so SUSHITRACK_API is defined in exactly one place. A second copy could drift from the one sushitrack_export.hpp carries.

package/
├── include/
│   └── SushiTrack/                   # every public header; a C consumer includes SushiTrack/sushitrack_c.h
├── lib/
│   └── sushitrack.lib                # Windows import library (Linux: libsushitrack.so)
├── bin/
│   └── sushitrack.dll                # Windows runtime library (Linux: omitted; the .so in lib/ is both)
├── sushitrack.json                   # default runtime config; edit to retune without recompiling
├── example/
│   ├── example.c                    # minimal create → update loop → destroy
│   └── CMakeLists.txt               # links lib/, copies the DLL beside the exe (Windows), sets RPATH (Linux)
└── README.md

Build and run the bundled example:

cd package/example
cmake -B build
cmake --build build
./build/sushitrack_example            # Windows: the DLL is auto-copied next to the exe by CMake

To integrate into your own project: add package/include to the include path, include SushiTrack/sushitrack_c.h, link against package/lib/sushitrack.lib (Windows) or package/lib/libsushitrack.so (Linux), and ensure the runtime library is resolvable at load time (DLL on PATH/beside the exe on Windows; RPATH or LD_LIBRARY_PATH on Linux). Create the tracker with sushitrack_create_from_config("sushitrack.json") to pick up the bundled config, and edit that JSON to retune the algorithm without recompiling.

--deploy python layout

Targets a Python consumer. The native library is bundled inside the package directory, beside the binding module, so the package is portable with no build tree and no SUSHITRACK_LIB variable.

package/
├── sushitrack/
│   ├── __init__.py                  # re-exports the public API (from .sushitrack import *)
│   ├── sushitrack.py                 # the ctypes binding (copied from bindings/python/)
│   └── sushitrack.dll                # bundled native library (Linux: libsushitrack.so)
├── sushitrack.json                   # default runtime config
├── example.py                       # minimal update loop using the bundled package
└── README.md

Run the bundled example:

cd package
python example.py

To integrate into your own project, copy the sushitrack/ directory into your source tree (or add package/ to PYTHONPATH) and use it as a normal package:

from sushitrack import Tracker, TrackState
tracker = Tracker(frame_rate=30, kalman_type=2)
tracks  = tracker.update([(x, y, w, h, score, label)], dt=1/30)

The binding looks beside its own module for the library before anywhere else, which is why the package needs no configuration. The full search order is in bindings/python/README.md.