Functions

Register functions

If running in function execution mode, you must register the functions in your compute module to make them callable from elsewhere in Foundry. This page explains two different methods for manually registering a compute module function.

Register a function from the Compute Modules application

You can manually register a function for a compute module from the Functions page. Select Add function to open the Create function panel:

1. Function name: The name of the function to be invoked. Typically, this should match the name of the function in your compute module.
2. Inputs: The input parameter(s) to be passed as arguments to your compute module function.

Compute module function inputs are packaged into a JSON object; each input that you add corresponds to a property on the input object passed into your function. In the example below, function inputs are on the left, and the JSON object passed to the corresponding function is on the right.

example_function_payload.json

{
    "arg1": "hello",
    "arg2": 2,
    "arg3": "1969-07-20"
}

3. Output: The return type of your function.
4. API name: The API name is the function locator that allows you call your function from other code in Foundry, such as through a TypeScript function. The compute module API format follows the structure com.<namespace>.computemodules.<MyApiName> and must comply to the following naming rules:

• namespace: Must be all lowercase and contain no special characters.
• MyApiName: Must be in camel case and contain no special characters.

Once you define a function, you can switch to the Test tab to try invoking the function, and/or select Save to save the function and make it callable from Foundry.

Register a function using JSON

You can also manually define your function schema by sending an HTTP POST request from within your compute module. Typically, you will only need to do this if you are creating your own client. For information on the HTTP request, review our POST function schema documentation.

This endpoint accepts a JSON array as the payload, where each element in the array corresponds to the specification of a function in your compute module. Our Python SDK ↗ provides a good reference on how to assemble this JSON payload.

Function type reference

Below is a table showing the mapping between function input/output types and how those types are serialized over HTTP to a compute module:

Automatic function schema inference

Compute modules offer a streamlined way to define and register functions, enabling automatic schema inference and integration with Foundry's Compute Module application. This section provides an in-depth look at the automatic registration of functions and advanced usage scenarios, ensuring a smoother development experience.

In your compute module, you can define the schema of a function using a JSON structure directly within your code. This approach offers several benefits:

• Centralized schema definition
• Easy maintenance and updates
• Automatic integration with Foundry

By making a simple POST call when your compute module starts up, the module automatically infers the schema from the endpoint call and makes it available as a function in the Computes Modules application. This allows developers to define endpoint schemas once and easily import them into Foundry.

Example: Add function schema

Consider a simple add function, where inputs are x and y (two integers) and the output is a string. The example below shows how to define the JSON schema for this function:

schemas.json

{
    "functionName": "add",
    "inputs": [
        {
            "name": "x",
            "dataType": {
                "integer": {},
                "type": "integer"
            },
            "required": true,
            "constraints": []
        },
        {
            "name": "y",
            "dataType": {
                "integer": {},
                "type": "integer"
            },
            "required": true,
            "constraints": []
        }
    ],
    "output": {
        "single": {
            "dataType": {
                "string": {},
                "type": "string"
            }
        },
        "type": "single"
    }
}

Once you have defined your JSON schema, send an HTTP POST request in your app.py file to register it with Foundry:

if __name__ == "__main__":
    certPath = os.environ['CONNECTIONS_TO_OTHER_PODS_CA_PATH']
    postSchemaUri = os.environ["POST_SCHEMA_URI"]

    with open('schemas.json', 'r') as file:
        SCHEMAS = json.load(file)

    requests.post(
        postSchemaUri,
        json=SCHEMAS,
        headers={"Module-Auth-Token": moduleAuthToken, "Content-Type": "application/json"},
        verify=certPath
    )

Make sure to handle exceptions and implement proper error logging in a production environment.

Notice that the function adheres to the following constraints:

• The schema definition function must declare the types of all of its inputs and the type of its output, using the supported Python type (see table below).
• The schema definition of each function must declare a functionName that matches the Python function name.

Automatic function discovery with the compute module SDK

The compute module SDK includes functionality for automatic function discovery. It inspects the defined functions and their input/output types, then converts them into FunctionSpecs that can be imported as Foundry Functions without modification.

To ensure this feature works seamlessly, you should understand how type inference works within the SDK and how to correctly define input and output types. Review the following considerations:

• The input class must be a complex type. Foundry Function specifications require the input type of a Function to be a complex type. If your function takes only a single primitive type as input, make sure to wrap that parameter in a complex type to properly infer your function schema.
• Input type definition

✅ TypedDict as input type

# app.py
from typing import TypedDict
from compute_modules.annotations import function

class HelloInput(TypedDict):
    planet: str

@function
def hello(context, event: HelloInput) -> str:
    return "Hello " + event["planet"] + "!"

✅ dataclass as input type

# app.py
from compute_modules.annotations import function
from dataclasses import dataclass
import datetime
import decimal

@dataclass
class TypedInput:
    bytes_value: bytes
    bool_value: bool
    date_value: datetime.date
    decimal_value: decimal.Decimal
    float_value: float
    int_value: int
    str_value: str
    datetime_value: datetime.datetime
    other_date_value: datetime.datetime

@function
def typed_function(context, event: TypedInput) -> str:
    diff = event.other_date_value - event.datetime_value
    return f"The difference between the provided dates is {diff}"

✅ Regular class with both class AND constructor type hints

# app.py
from compute_modules.annotations import function

class GoodExample:
    some_flag: bool
    some_value: int

    def __init__(self, some_flag: bool, some_value: int) -> None:
        self.some_flag = some_flag
        self.some_value = some_value

@function
def typed_function(context, event: GoodExample) -> int:
    return event.some_value

❌ AVOID Python class with no class type hints

# app.py
# This will raise an exception
class BadClassNoTypeHints:
    def __init__(self, arg1: str, arg2: int):
        ...

❌ AVOID Python class with Args in constructor

# app.py
# This will raise an exception
class BadClassArgsInit:
    arg1: str
    arg2: int

    def __init__(self, arg1: str, arg2: int, *args):
        ...

❌ AVOID Python class with Kwargs in constructor

# app.py
# This will raise an exception
class BadClassKwargsInit:
    arg1: str
    arg2: int

    def __init__(self, arg1: str, arg2: int, **kwargs):
        ...

• Streaming output: The compute module python SDK includes support for streaming output if it is any Iterable type (except dict). To enable result streaming, change @function to @function(streaming=True). You can review more details in our [SDK documentation ↗](https://github.com/palantir/python-compute-module?tab=readme-ov-file#advanced-usage-1---streaming-result). To make sure your streaming function is registered correctly, use any Iterabletype as the return type. Then the output will be registered as FoundryArray`.

✅ Regular Iterable as output type

# app.py
# The outputs will be registered as Foundry Array
from compute_modules.annotations import function

@function(streaming=True)
def get_string_list(context, event) -> list[str]:
    return [f'string {i}' for i in range(10)]

@function(streaming=True)
def get_string_set(context, event) -> set[str]:
    return {'string 1', 'string 2', 'string 3'}

✅ Generator as output type

# app.py
# Generator is Iterable. The output will be registered as Foundry Array
from compute_modules.annotations import function
import typing

@function(streaming=True)
def string_generator(context, event) -> typing.Iterable[str]:
    for i in range(10):
        yield f'string {i}'

⚠️ Regular Iterable as output type but streaming not enabled

# app.py
# This is valid. The output will be registered as Foundry Array, but the result will not be streamed
from compute_modules.annotations import function

@function
def get_string_list(context, event) -> list[str]:
    return [f'string {i}' for i in range(10)]

❌ Generator as output type but streaming not enabled

# app.py
# Generator is not JSON serializable as a whole object. Cannot be used in a non-streaming function since it serializes the whole object
# The output type will be registered as Foundry Array, but it will throw when executed
from compute_modules.annotations import function
import typing

@function
def string_generator(context, event) -> typing.Iterable[str]:
    for i in range(10):
        yield f'string {i}'

Register the function

Follow the steps below to register your function:

1. Ensure your compute module is running.
2. Navigate to the Functions tab in the Compute Module application.
3. You should be able to view your function in the list of detected functions.
4. Select the function you want to register to open a pop-up window.
5. In the window, select Import.

Use compute module functions in TypeScript functions

Prerequisites:

• You must register your function in the Compute Module application with an API name.
• You must have the compute module running for live preview to work.
• You must initialize a TypeScript code repository.

Enable resource generation

Before you begin, ensure that resource generation is enabled in your Typescript code repository:

• Open your functions.json file.
• Set the enableResourceGeneration property to true.

Import your compute module function

To import a compute module function in TypeScript, follow the steps below:

1. From the left panel of the Compute Modules application, find and select the Resource imports tab.
2. Select Add, then select Query Functions to display a pop-up window to select an Ontology.
3. Although compute modules are not tied to a specific Ontology, you must select one for the import process. Choose any Ontology that suits your use case.
4. Search for your compute module function's API name.
5. Select the function.
6. Choose Confirm selection.

Rebuild your code workspace

Import and use the function

The example below shows how to import and use a compute module function:

// index.ts
import { Function } from "@foundry/functions-api";

// API Name: com.mycustomnamespace.computemodules.Add
import { add } from "@mycustomnamespace/computemodules";

export class MyFunctions {
    @Function()
    public async myFunction(): Promise<string> {
        return await add({ x: 50, y: 50 });
    }
}

Important considerations

• Project location: Ensure the compute module is in the same Project as your TypeScript code for live preview to work correctly.
• Type consistency: TypeScript enforces strict type checking. Ensure the declared return type matches the actual return type of your compute module function. For example, if you declare a string return type, your registered compute module function must return a string, not a struct type.
• Asynchronous operations: Compute module functions are typically asynchronous. Use async/await syntax for proper handling.