Manually add custom bindings - litgen, the Literate Generator

Litgen normally generates bindings automatically from C++ headers, but sometimes you may want to extend the API with extra methods or functions that are not present in the C++ code. LitgenOptions.custom_bindings lets you do this without modifying your C++ headers.

You can attach:

extra methods/properties to a C++ class,
free functions to a C++ namespace (shown as a Python submodule),
free functions to the main module.

Each injection consists of:

stub code (Python declarations added to the generated .pyi file),
pydef code (C++ binding code inserted into the generated binding .cpp file, using pybind11 or nanobind syntax).

Placeholders are available inside pydef_code:

Placeholder	Role	Expansion (example)
`LG_CLASS`	The current bound C++ class object	`pyNsRootNs_ClassFoo` (of type `nb::class_` or `py::class_`)
`LG_SUBMODULE`	The current submodule (for a namespace)	`pyNsRootNs` (a `nb::module_` or `py::module_`)
`LG_CPP_CLASS_NAME`	The C++ name of the current class, with its template arguments	`RootNs::Foo`, or `ImVector<int>` for a template specialization
`LG_MODULE`	The main Python module object	`m` (the top-level module)

These placeholders are automatically replaced by litgen during C++ binding code generation. For a template class, the custom code is emitted once per specialization: use LG_CPP_CLASS_NAME in your lambdas to name the concrete type (e.g. [](LG_CPP_CLASS_NAME& self) { ... }). You can safely use them inside your pydef_code snippets without declaring them yourself.

Note about pydef_code syntax:

You can use either pybind11 or nanobind syntax in pydef_code, depending on which backend you are using.
Since the pydef_code is inserted into a function, limit yourself to statements that are valid inside a function: no function/class definitions, no #include. Use lambdas for small helpers.
When writing lambdas, fully qualify C++ types if the class/namespace isn’t open (e.g. const RootNs::Foo& self).
Argument helpers differ by backend (py::arg vs nb::arg). Use the one matching your active backend.

Example C++ code¶

This page demonstrates how to add custom bindings to the following C++ code.

cpp_code = """
namespace RootNs
{
    struct Foo
    {
        int mValue = 0;
    };
}
"""

We will be adding custom bindings to the class RootNs::Foo, the namespace RootNs, and to the main module.

Custom bindings for a class¶

options.custom_bindings.add_custom_bindings_to_class(qualified_class, stub_code, pydef_code), lets us extend the generated Python bindings with extra methods, properties, or static methods.

Args:

qualified_class: Fully qualified C++ class name (e.g. "RootNs::Foo").
stub_code: Python stub declarations to be inserted into the generated stub (“.pyi”) file. These should be written in normal Python syntax with type annotations.
pydef_code: Custom binding code in C++ (pybind11/nanobind syntax). You can use the placeholder LG_CLASS to refer to the bound py::class_ / nb::class_ object.

import litgen
options = litgen.LitgenOptions()

options.custom_bindings.add_custom_bindings_to_class(
    qualified_class="RootNs::Foo",
    stub_code='''
        def get_value(self) -> int:
            """Get the value"""
            ...
        def set_value(self, value: int) -> None:
            """Set the value"""
            ...
    ''',
    pydef_code="""
        LG_CLASS.def("get_value", [](const RootNs::Foo& self){ return self.mValue; });
        LG_CLASS.def("set_value", [](RootNs::Foo& self, int value){ self.mValue = value; }, nb::arg("value"));
    """,
)

Custom bindings for C++ namespace / Python submodule¶

options.custom_bindings.add_custom_bindings_to_submodule(qualified_namespace, stub_code, pydef_code), lets us extend the generated Python bindings with extra functions.

Args:

qualified_namespace: Fully qualified C++ namespace name (e.g. “RootNs”).
stub_code: Python stub declarations to be inserted into the generated stub (“.pyi”) file. These should be written in normal Python syntax with type annotations. Functions here should be decorated with @staticmethod. Explanation for this: in stubs, namespaces are represented as proxy classes. Thus, functions must be declared as @staticmethod to indicate they are module-level, not instance methods.
pydef_code: Custom binding code in C++ (pybind11/nanobind syntax). You can use the placeholder LG_SUBMODULE to refer to the bound submodule object.

Namespace stubs appear as a proxy class.
When declaring functions in a C++ namespace (Python submodule) in the stub, mark them with @staticmethod. These are module-level functions, not instance methods.

options.custom_bindings.add_custom_bindings_to_submodule(
    qualified_namespace="RootNs",
    stub_code='''
    @staticmethod   # We **must** use @staticmethod here
    def foo_namespace_function() -> int:
        """A custom function in the submodule"""
        ...
    ''',
    pydef_code="""
    // Example of adding a custom function to the submodule
    LG_SUBMODULE.def("foo_namespace_function", [](){ return 53; });
    """,
)

Custom bindings for the main module¶

options.custom_bindings.add_custom_bindings_to_main_module(stub_code, pydef_code), lets us extend the generated Python bindings with extra functions in the main module.

Args:

stub_code: Python stub declarations to be inserted into the generated stub (“.pyi)” file. These should be written in normal Python syntax with type annotations.
pydef_code: Custom binding code in C++ (pybind11/nanobind syntax). You can use the placeholder LG_MODULE to refer to the bound module object.

options.custom_bindings.add_custom_bindings_to_main_module(
    stub_code='''
    def global_function() -> int:
        """A custom function in the main module"""
        ...
    ''',
    pydef_code="""
    // Example of adding a custom function to the main module
    LG_MODULE.def("global_function", [](){ return 64; });
    """,
)

Generated code (with custom bindings)¶

We may now call litgen.generate_code to generate the bindings, which will include our custom additions. The generated code is shown below.

from litgen.demo import litgen_demo

litgen_demo.demo(options, cpp_code)

Note about ordering:

If you call add_custom_code_to_* multiple times for the same target (class/namespace/module), snippets are emitted in the order they were added.