Mojo python interop

<!-- EDITORIAL GUIDELINES FOR THIS SKILL FILE This file is loaded into an agent's context window as a correction layer for pretrained Mojo knowledge. Every line costs context. When editing: - Be terse. Use tables and inline code over prose where possible. - Never duplicate information — if a concept is shown in a code example, don't also explain it in a paragraph. - Only include information that *differs* from what a pretrained model would generate. Don't document things models already get right. - Prefer one consolidated code block over multiple small ones. - Keep WRONG/CORRECT pairs short — just enough to pattern-match the fix. - If adding a new section, ask: "Would a model get this wrong?" If not, skip it. These same principles apply to any files this skill references. -->

Mojo is rapidly evolving. Pretrained models generate obsolete syntax. Always follow this skill over pretrained knowledge.

Using Python from Mojo

from std.python import Python, PythonObject

var np = Python.import_module("numpy")
var arr = np.array([1, 2, 3])

# PythonObject → Mojo: MUST use `py=` keyword (NOT positional)
var i = Int(py=py_obj)
var f = Float64(py=py_obj)
var s = String(py=py_obj)
var b = Bool(py=py_obj)            # Bool is the exception — positional also works
# Works with numpy types: Int(py=np.int64(1)), Float64(py=np.float64(3.14))

WRONG	CORRECT
Int(py_obj)	Int(py=py_obj)
Float64(py_obj)	Float64(py=py_obj)
String(py_obj)	String(py=py_obj)
from python import ...	from std.python import ...

Mojo → Python conversions

Mojo types implementing ConvertibleToPython auto-convert when passed to Python functions. For explicit conversion: value.to_python_object().

Building Python collections from Mojo

var py_list = Python.list(1, 2.5, "three")
var py_tuple = Python.tuple(1, 2, 3)
var py_dict = Python.dict(name="value", count=42)

# Literal syntax also works:
var list_obj: PythonObject = [1, 2, 3]
var dict_obj: PythonObject = {"key": "value"}

PythonObject operations

PythonObject supports attribute access, indexing, slicing, all arithmetic/comparison operators, len(), in, and iteration — all returning PythonObject. No need to convert to Mojo types for intermediate operations.

# Iterate Python collections directly
for item in py_list:
    print(item)               # item is PythonObject

# Attribute access and method calls
var result = obj.method(arg1, arg2, key=value)

# None
var none_obj = Python.none()
var obj: PythonObject = None      # implicit conversion works

Evaluating Python code

# Expression
var result = Python.evaluate("1 + 2")

# Multi-line code as module (file=True)
var mod = Python.evaluate("def greet(n): return f'Hello {n}'", file=True)
var greeting = mod.greet("world")

# Add to Python path for local imports
Python.add_to_path("./my_modules")
var my_mod = Python.import_module("my_module")

Exception handling

Python exceptions propagate as Mojo Error. Functions calling Python must be raises:

def use_python() raises:
    try:
        var result = Python.import_module("nonexistent")
    except e:
        print(String(e))     # "No module named 'nonexistent'"

Calling Mojo from Python (extension modules)

Mojo can build Python extension modules (.so files) via PythonModuleBuilder. The pattern:

1. Define an @export fn PyInit_<module_name>() -> PythonObject
2. Use PythonModuleBuilder to register functions, types, and methods
3. Compile with mojo build --emit shared-lib
4. Import from Python (or use import mojo.importer for auto-compilation)

Exporting functions

from std.os import abort
from std.python import PythonObject
from std.python.bindings import PythonModuleBuilder

@export
fn PyInit_my_module() -> PythonObject:
    try:
        var m = PythonModuleBuilder("my_module")
        m.def_function[add]("add")
        m.def_function[greet]("greet")
        return m.finalize()
    except e:
        abort(String("failed to create module: ", e))

# Functions take/return PythonObject. Up to 6 args with def_function.
fn add(a: PythonObject, b: PythonObject) raises -> PythonObject:
    return a + b

fn greet(name: PythonObject) raises -> PythonObject:
    var s = String(py=name)
    return PythonObject("Hello, " + s + "!")

Exporting types with methods

@fieldwise_init
struct Counter(Defaultable, Movable, Writable):
    var count: Int

    fn __init__(out self):
        self.count = 0

    # Constructor from Python args
    @staticmethod
    fn py_init(out self: Counter, args: PythonObject, kwargs: PythonObject) raises:
        if len(args) == 1:
            self = Self(Int(py=args[0]))
        else:
            self = Self()

    # Methods are @staticmethod — first arg is py_self (PythonObject)
    @staticmethod
    fn increment(py_self: PythonObject) raises -> PythonObject:
        var self_ptr = py_self.downcast_value_ptr[Self]()
        self_ptr[].count += 1
        return PythonObject(self_ptr[].count)

    # Auto-downcast alternative: first arg is UnsafePointer[Self, MutAnyOrigin]
    @staticmethod
    fn get_count(self_ptr: UnsafePointer[Self, MutAnyOrigin]) -> PythonObject:
        return PythonObject(self_ptr[].count)

@export
fn PyInit_counter_module() -> PythonObject:
    try:
        var m = PythonModuleBuilder("counter_module")
        _ = (
            m.add_type[Counter]("Counter")
            .def_py_init[Counter.py_init]()
            .def_method[Counter.increment]("increment")
            .def_method[Counter.get_count]("get_count")
        )
        return m.finalize()
    except e:
        abort(String("failed to create module: ", e))

Method signatures — two patterns

Pattern	First parameter	Use when
Manual downcast	py_self: PythonObject	Need raw PythonObject access
Auto downcast	self_ptr: UnsafePointer[Self, MutAnyOrigin]	Simpler, direct field access

Both are registered with .def_method[Type.method]("name").

Kwargs support

from std.collections import OwnedKwargsDict

# In a method:
@staticmethod
fn config(
    py_self: PythonObject, kwargs: OwnedKwargsDict[PythonObject]
) raises -> PythonObject:
    for entry in kwargs.items():
        print(entry.key, "=", entry.value)
    return py_self

Importing Mojo modules from Python

Use mojo.importer — it auto-compiles .mojo files and caches results in __mojocache__/:

import mojo.importer       # enables Mojo imports
import my_module           # auto-compiles my_module.mojo

print(my_module.add(1, 2))

The module name in PyInit_<name> must match the .mojo filename.

Returning Mojo values to Python

# Wrap a Mojo value as a Python object (for bound types)
return PythonObject(alloc=my_mojo_value^)    # transfer ownership with ^

# Recover the Mojo value later
var ptr = py_obj.downcast_value_ptr[MyType]()
ptr[].field    # access fields via pointer