Type translations

In order to communicate between Python and Javascript, we “translate” objects between the two languages. Depending on the type of the object we either translate the object by implicitly converting it or by proxying it. By “converting” an object we mean producing a new object in the target language which is the equivalent of the object from the source language, for example converting a Python string to the equivalent a Javascript string. By “proxying” an object we mean producing a special object in the target language that forwards requests to the source language. When we proxy a Javascript object into Python, the result is a JsProxy object. When we proxy a Python object into Javascript, the result is a PyProxy object. A proxied object can be explicitly converted using the explicit conversion methods JsProxy.to_py and PyProxy.toJs.

Python to Javascript translations occur:

  • when returning the final expression from a pyodide.runPython call,

  • when using pyodide.globals.get('key'),

  • when passing arguments to a Javascript function called from Python,

  • when returning the results of a Python function called from Javascript,

  • when accessing an attribute of a PyProxy

Javascript to Python translations occur:

  • when using the from js import ... syntax

  • passing arguments to a Python function called from Javascript

  • returning the result of a Javascript function called from Python

  • when accessing an attribute of a JsProxy

Memory Leaks and Python to Javascript translations

Any time a Python to Javascript translation occurs, it may create a PyProxy. To avoid memory leaks, you must store the PyProxy and destroy it when you are done with it. See Best practices for avoiding memory leaks for more info.

Round trip conversions

Translating an object from Python to Javascript and then back to Python is guaranteed to give an object that is equal to the original object (with the exception of nan because nan != nan). Furthermore, if the object is proxied into Javascript, then translation back unwraps the proxy, and the result of the round trip conversion is the original object (in the sense that they live at the same memory address).

Translating an object from Javascript to Python and then back to Javascript gives an object that is === to the original object. Furthermore, if the object is proxied into Python, then translation back unwraps the proxy, and the result of the round trip conversion is the original object (in the sense that they live at the same memory address). There are a few exceptions:

  1. NaN is converted to NaN after a round trip but NaN !== NaN,

  2. null is converted to undefined after a round trip, and

  3. a BigInt will be converted to a Number after a round trip unless its absolute value is greater than Number.MAX_SAFE_INTEGER (i.e., 2^53).

Implicit conversions

We implicitly convert immutable types but not mutable types. This ensures that mutable Python objects can be modified from Javascript and vice-versa. Python has immutable types such as tuple and bytes that have no equivalent in Javascript. In order to ensure that round trip translations yield an object of the same type as the original object, we proxy tuple and bytes objects.

Python to Javascript

The following immutable types are implicitly converted from Javascript to Python:




Number or BigInt*









  • An int is converted to a Number if the int is between -2^{53} and 2^{53} inclusive, otherwise it is converted to a BigInt. (If the browser does not support BigInt then a Number will be used instead. In this case, conversion of large integers from Python to Javascript is lossy.)

Javascript to Python

The following immutable types are implicitly converted from Python to Javascript:




int or float as appropriate*











  • A number is converted to an int if it is between -2^{53} and 2^{53} inclusive and its fractional part is zero. Otherwise it is converted to a float.


Any of the types not listed above are shared between languages using proxies that allow methods and some operations to be called on the object from the other language.

Proxying from Javascript into Python

When most Javascript objects are translated into Python a JsProxy is returned. The following operations are currently supported on a JsProxy.







proxy.foo = bar

x.foo = bar

del proxy.foo

delete x.foo

hasattr(proxy, "foo")

"foo" in x






new X(...)


x.length or x.size

foo in proxy




proxy[foo] = bar

x.set(foo, bar)

del proxy[foo]


proxy1 == proxy2

x === y


typeof x





await proxy

await x



Some other code snippets:

for v in proxy:
    # do something

is equivalent to:

for(let v of x){
    // do something

The dir method has been overloaded to return all keys on the prototype chain of x, so dir(x) roughly translates to:

function dir(x){
    let result = [];
    do {
    } while (x = Object.getPrototypeOf(x));
    return result;

As a special case, Javascript Array, HTMLCollection, and NodeList are container types, but instead of using array.get(7) to get the 7th element, Javascript uses array["7"]. For these cases, we translate:





proxy[idx] = val


idx in proxy

idx in array

del proxy[idx]


Proxying from Python into Javascript

When most Python objects are translated to Javascript a PyProxy is produced. See also the API docs for PyProxy.

Fewer operations can be overloaded in Javascript than in Python so some operations are more cumbersome on a PyProxy than on a JsProxy. The following operations are supported:



foo in proxy

hasattr(x, 'foo')



proxy.foo = bar

x.foo = bar

delete proxy.foo

del x.foo










foo in x



proxy.set(foo, bar)

x[foo] = bar


del x[foo]







await proxy

await x



Memory Leaks and PyProxy

When proxying a Python object into Javascript, there is no way for Javascript to automatically garbage collect the Proxy. The PyProxy must be manually destroyed when passed to Javascript, or the proxied Python object will leak. To do this, call PyProxy.destroy() on the PyProxy, after which Javascript will no longer have access to the Python object. If no references to the Python object exist in Python either, then the Python garbage collector can eventually collect it.

let foo = pyodide.globals.get('foo');
foo(); // throws Error: Object has already been destroyed

Memory Leaks and PyProxy method calls

Every time you access a Python method on a PyProxy, it creates a new temporary PyProxy of a Python bound method. If you do not capture this temporary and destroy it, you will leak the Python object. See Best practices for avoiding memory leaks.

Explicit Conversion of Proxies

Python to Javascript

Explicit conversion of a PyProxy into a native Javascript object is done with the PyProxy.toJs method. By default, the toJs method does a recursive “deep” conversion, to do a shallow conversion use proxy.toJs(1). The toJs method performs the following explicit conversions:



list, tuple






a buffer*


  • Examples of buffers include bytes objects and numpy arrays.

In Javascript, Map and Set keys are compared using object identity unless the key is an immutable type (meaning a string, a number, a bigint, a boolean, undefined, or null). On the other hand, in Python, dict and set keys are compared using deep equality. If a key is encountered in a dict or set that would have different semantics in Javascript than in Python, then a ConversionError will be thrown.

See Converting Python Buffer objects to Javascript for the behavior of toJs on buffers.

Memory Leaks and toJs

The toJs method can create many proxies at arbitrary depth. It is your responsibility to manually destroy these proxies if you wish to avoid memory leaks, but we provide no way to manage this.

This is a flaw in the current design of the toJs API, we hope to improve the situation in the future.

To ensure that no PyProxy is leaked, the following code suffices:

function destroyToJsResult(x){
        for(let k of x){

Javascript to Python

Explicit conversion of a JsProxy into a native Python object is done with the JsProxy.to_py method. By default, the to_py method does a recursive “deep” conversion, to do a shallow conversion use proxy.to_py(1) The to_py method performs the following explicit conversions:











** to_py will only convert an object into a dictionary if its constructor is Object, otherwise the object will be left alone. Example:

class Test {};
window.x = { "a" : 7, "b" : 2};
window.y = { "a" : 7, "b" : 2};
Object.setPrototypeOf(y, Test.prototype);
    from js import x, y
    # x is converted to a dictionary
    assert x.to_py() == { "a" : 7, "b" : 2}
    # y is not a "Plain Old JavaScript Object", it's an instance of type Test so it's not converted
    assert y.to_py() == y

In Javascript, Map and Set keys are compared using object identity unless the key is an immutable type (meaning a string, a number, a bigint, a boolean, undefined, or null). On the other hand, in Python, dict and set keys are compared using deep equality. If a key is encountered in a Map or Set that would have different semantics in Python than in Javascript, then a ConversionError will be thrown. Also, in Javascript, true !== 1 and false !== 0, but in Python, True == 1 and False == 0. This has the result that a Javascript map can use true and 1 as distinct keys but a Python dict cannot. If the Javascript map contains both true and 1 a ConversionError will be thrown.


Converting Javascript Typed Arrays to Python

Javascript ArrayBuffers and ArrayBuffer views (Int8Array and friends) are proxied into Python. Python can’t directly access arrays if they are outside of the wasm heap so it’s impossible to directly use these proxied buffers as Python buffers. You can convert such a proxy to a Python memoryview using the to_py api. This makes it easy to correctly convert the array to a Numpy array using numpy.asarray:

self.jsarray = new Float32Array([1,2,3, 4, 5, 6]);
    from js import jsarray
    array = jsarray.to_py()
    import numpy as np
    numpy_array = np.asarray(array).reshape((2,3))

After manipulating numpy_array you can assign the value back to jsarray using JsProxy.assign:

    numpy_array[1,1] = 77
console.log(jsarray); // [1, 2, 3, 4, 77, 6]

The JsProxy.assign and JsProxy.assign_to methods can be used to assign a Javascript buffer from / to a Python buffer which is appropriately sized and contiguous. The assignment methods will only work if the data types match, the total length of the buffers match, and the Python buffer is contiguous.

These APIs are currently experimental, hopefully we will improve them in the future.

Converting Python Buffer objects to Javascript

Python objects supporting the Python Buffer protocol are proxied into Javascript. The data inside the buffer can be accessed via the PyProxy.toJs method or the PyProxy.getBuffer method. The toJs API copies the buffer into Javascript, whereas the getBuffer method allows low level access to the WASM memory backing the buffer. The getBuffer API is more powerful but requires care to use correctly. For simple use cases the toJs API should be prefered.

If the buffer is zero or one-dimensional, then toJs will in most cases convert it to a single TypedArray. However, in the case that the format of the buffer is 's', we will convert the buffer to a string and if the format is '?' we will convert it to an Array of booleans.

If the dimension is greater than one, we will convert it to a nested Javascript array, with the innermost dimension handled in the same way we would handle a 1d array.

An example of a case where you would not want to use the toJs method is when the buffer is bitmapped image data. If for instance you have a 3d buffer shaped 1920 x 1080 x 4, then toJs will be extremely slow. In this case you could use PyProxy.getBuffer. On the other hand, if you have a 3d buffer shaped 1920 x 4 x 1080, the performance of toJs will most likely be satisfactory. Typically the innermost dimension won’t matter for performance.

The PyProxy.getBuffer method can be used to retrieve a reference to a Javascript typed array that points to the data backing the Python object, combined with other metadata about the buffer format. The metadata is suitable for use with a Javascript ndarray library if one is present. For instance, if you load the Javascript ndarray package, you can do:

let proxy = pyodide.globals.get("some_numpy_ndarray");
let buffer = proxy.getBuffer();
try {
        // We can't stop you from changing a readonly buffer, but it can cause undefined behavior.
        throw new Error("Uh-oh, we were planning to change the buffer");
    let array = new ndarray(buffer.data, buffer.shape, buffer.strides, buffer.offset);
    // manipulate array here
    // changes will be reflected in the Python ndarray!
} finally {
    buffer.release(); // Release the memory when we're done

Importing Objects

It is possible to access objects in one languge from the global scope in the other language. It is also possible to create custom namespaces and access objects on the custom namespaces.

Importing Python objects into Javascript

A Python object in the __main__ global scope can imported into Javascript using the pyodide.globals.get method. Given the name of the Python object to import, it returns the object translated to Javascript.

let sys = pyodide.globals.get('sys');

As always, if the result is a PyProxy and you care about not leaking the Python object, you must destroy it when you are done. It’s also possible to set values in the Python global scope with pyodide.globals.set or remove them with pyodide.globals.delete:

pyodide.globals.set("x", 2);
pyodide.runPython("print(x)"); // Prints 2

If you execute code with a custom globals dictionary, you can use a similar approach:

let my_py_namespace = pyodide.globals.get("dict")();
pyodide.runPython("x=2", my_py_namespace);
let x = my_py_namespace.get("x");

Importing Javascript objects into Python

Javascript objects in the globalThis global scope can be imported into Python using the js module.

When importing a name from the js module, the js module looks up Javascript attributes of the globalThis scope and translates the Javascript objects into Python.

import js
js.document.title = 'New window title'
from js.document.location import reload as reload_page

You can also assign to Javascript global variables in this way:

pyodide.runPython("js.x = 2");
console.log(window.x); // 2

You can create your own custom Javascript modules using pyodide.registerJsModule and they will behave like the js module except with a custom scope:

let my_js_namespace = { x : 3 };
pyodide.registerJsModule("my_js_namespace", my_js_namespace);
    from my_js_namespace import x
    print(x) # 3
    my_js_namespace.y = 7
console.log(my_js_namespace.y); // 7

Translating Errors

All entrypoints and exit points from Python code are wrapped in Javascript try blocks. At the boundary between Python and Javascript, errors are caught, converted between languages, and rethrown.

Javascript errors are wrapped in a JsException. Python exceptions are converted to a PythonError. At present if an exception crosses between Python and Javascript several times, the resulting error message won’t be as useful as one might hope.

In order to reduce memory leaks, the PythonError has a formatted traceback, but no reference to the original Python exception. The original exception has references to the stack frame and leaking it will leak all the local variables from that stack frame. The actual Python exception will be stored in sys.last_value so if you need access to it (for instance to produce a traceback with certain functions filtered out), use that.

Avoid Stack Frames

If you make a PyProxy of sys.last_value, you should be especially careful to destroy() it when you are done with it or you may leak a large amount of memory if you don’t.

The easiest way is to only handle the exception in Python:

def reformat_exception():
    from traceback import format_exception
    # Format a modified exception here
    # this just prints it normally but you could for instance filter some frames
    return "".join(
        traceback.format_exception(sys.last_type, sys.last_value, sys.last_traceback)
let reformat_exception = pyodide.globals.get("reformat_exception");
try {
} catch(e){
    // replace error message
    e.message = reformat_exception();
    throw e;

Best practices for avoiding memory leaks

If the browser supports FinalizationRegistry then a PyProxy that is not part of a Javascript/Python reference cycle will eventually be collected, but it is unpredictable when it will be collected. In practice it typically takes a long time. Furthermore, the Javascript garbage collector does not have any information about whether Python is experiencing memory pressure. So it’s best to aim to avoid leaks.

When using a PyProxy, note that accessing a field of the PyProxy is likely to yield more PyProxy objects that also need to be destroyed. A particular gotcha occurs with method calls:


This leaks pyproxy! Insteaad:

let some_func = pyproxy.some_func;

To be absolutely foolproof we can do it in a finally block:

let some_func;
try {
    some_func = pyproxy.some_func;
} finlly {
    // To be extra sure we do it in a finally block.

Obviously it’s not a whole lot of fun writing code like this. We hope to improve the design to make managing PyProxy lifecycles more ergonomic in the future.

Here are some tips for how to do that when calling functions in one language from another.

There are four cases to consider here:

  • calling a Python function from a Javascript function you wrote,

  • calling a Python function from an existing Javascript callback,

  • calling a Javascript function from Python code you wrote, or

  • calling a Javascript function you wrote from an existing Python callback.

If you want to pass an existing Javascript function as a callback to an existing Python function, you will need to define a wrapper around the Javascript callback. That wrapper can then use approaches described here. Similarly with the reverse direction.

Calling Python functions from Javascript

In this case we just need to pay attention to the return value (and to the function itself if you care about not leaking it).

pyodide.runPython("from itertools import accumulate");
let accumulate = pyodide.globals.get("accumulate");
let pyresult = accumulate([1,5,1,7]);
let result = [...pyresult];
console.log(result); // [1, 6, 7, 14]

Calling Javascript functions from Python

If the arguments will be implicitly converted, nothing needs to be done. Otherwise, there are different solutions depending on the circumstance.

  1. Call pyodide.to_js on the argument before passing it if is a list, dict, set, or buffer.

  2. For anything, you can use pyodide.create_proxy. Suppose obj is some arbitrary Python object that you want to pass to a Javascript function.

obj = [1, 2, 3]
jsobj = pyodide.create_proxy(obj)
jsobj.destroy() # reclaim memory

Note that as long as obj wouldn’t be implicitly translated, the Javascript function will recieve an identical object regardless of whether you call it directly (i.e., jsfunc(obj)) or as jsfunc(create_proxy(obj)).

create_proxy is particularly helpful with addEventListener:

def callback():
proxy = pyodide.create_proxy(callback)
from js import document
document.body.addEventListener("click", proxy)
# do other stuff, keep hold of proxy
document.body.removeEventListener("click", proxy)
proxy.destroy() # reclaim memory
  1. If the argument is a function to be called once (for example, the argument to Promise.new) you can use pyodide.create_once_callable:

from pyodide import create_once_callable
def executor(resolve, reject):
    # Do something
p = Promise.new(create_once_callable(executor))
  1. If you are using the promise methods PyProxy.then, PyProxy.catch, or PyProxy.finally, these have magic wrappers around them so no intervention is needed to prevent memory leaks.

  2. If the last argument of the Javascript function is an object you can use keyword arguments, so the following:

from js import fetch
from pyodide import to_js
resp = await fetch('example.com/some_api',
    method= "POST",
    body= '{ "some" : "json" }',
    credentials= "same-origin",
    headers= to_js({ "Content-Type": "application/json" }),

is equivalent to the Javascript code

let resp = await fetch('example.com/some_api',{
    method : "POST",
    body : '{ "some" : "json" }',
    credentials : "same-origin",
    headers : { "Content-Type": "application/json" },

Using a Javascript callback with an existing Python function

If you want to pass a Javascript callback to an existing Python function, you should destroy the argument when you are done. This can be a bit tedious to get correct due to PyProxy usage constraints.

function callback(arg){
    let res_method = arg.result;
    let res = res_method();
    window.result = res.toJs();
let fut = pyodide.runPython(`
    from asyncio import ensure_future
    async def temp():
        return [1, 2, 3]

Using a Python callback with an existing Javascript function

If it’s only going to be called once:

from pyodide import create_once_callable
from js import setTimeout
def my_callback():
setTimeout(create_once_callable(my_callback), 1000)

If it’s going to be called many times:

from pyodide import create_proxy
from js import document
def my_callback():
proxy = document.create_proxy(my_callback)
document.body.addEventListener("click", proxy)
# ...
# make sure to hold on to proxy
document.body.removeEventListener("click", proxy)

Be careful with the return values. You might want to use to_js on the result:

from pyodide import to_js
def my_callback():
    result = [1, 2, 3]
    return to_js(result)