Should meson and meson-python support Ziggy Pydust?

[jburgy]

Ziggy Pydust uses reflection and compile-time metaprogramming to generate python bindings around Zig code. It produces a python extension module as a shared object. custom_target can be coerced into installing that shared object but the call is finicky:

# progress/meson.build
custom_target(
    input : 'binding.zig',
    output : 'binding.abi3.so',
    command: [py, '-m', 'pydust', 'build', 'progress.binding=@INPUT@'],
    install : true,
    install_dir : py.get_install_dir() / 'progress',
)

(See in the progress.binding=@INPUT@ argument specifically). How much work would it be to make Ziggy Pydust as seamless as Cython in meson? Note I tried to make the command argument less repetitive by using @OUTDIR@ and @BASENAME@ but that didn't work. How many variables will meson substitute in the same command line argument?

Relevant bit of directory layout for context. Actual example is here

blog/
├── meson.build
└── progress
    ├── binding.pyi
    ├── binding.zig
    ├── __init__.py
    └── meson.build

Related to #12458 in the sense that Cython is related to C.

[eli-schwartz]

How much work would it be to make Ziggy Pydust as seamless as Cython in meson?

Probably the first step is to add zig support. Once that is available, it looks like pydust is "just" a zig library smoothing over the API, similar to pybind11 for C++, so you should simply be able to add pydust as a dependency to a

# progress/meson.build
py.extension_module('binding', 'binding.zig',
    dependencies: pydust_dep,
    install: true,
    subdir: 'progress',
)

[dcbaker]

I rebased that branch, which was less painful than I expected. There's still a lot of TODOs there, and zig still doesn't have depfile support.