Bug Report: Loading custom recipe raises "Unsupported type for 'dataset': str"

1

:bug: Bug description:
When loading a custom recipe the following error is raised:
✘ Couldn't load Python code: src/my_recipe.py
Unsupported type for 'dataset': str

:man_walking:t4:Reproduction steps:
How can we recreate the bug?
This is the definition of the custom recipe:

@prodigy.recipe(
    "my-recipe",
    dataset=("Dataset to save annotations", "positional", None, str),
    source=("Path to JSONL file", "positional", None, Path),
    source2=("Path to second JSONL file", "positional", None, Path),
)
def my_recipe(dataset: str, source: Path, source2: Path):

It is loaded via the command:
python -m prodigy my-recipe "my_annotations" .\data\file1.jsonl .\data\file2.jsonl -F src/my_recipe.py

It was loading successfully with prodigy v1.11.7. The error is raised after upgrading to v1.18.2.

The error disappears if I remove the type annotations from “my_recipe”

:desktop_computer: Environment variables:
Please provide prodigy stats or Python version/OS/Prodigy version:
Version 1.18.2
License Type Prodigy Personal
Platform Windows-11-10.0.22631-SP0
Python Version 3.12.10
spaCy Version 3.7.5

2

Welcome to the forum @jo-fra! :waving_hand:

Thank you very much for the detailed bug report!
In the version 1.14.0 we replaced plac CLI handling library with radicli and it looks like there's a Windows specific bug in radicli that leads to built-in types being converted to strings, which, in turn, leads to "Unsupporter type" error.

We'll issue a patch to radicli to make sure it's more robust with type hints handling across platforms. We'll let you know as soon as Prodigy version with the patch is available.

3

Hi @jo-fra,

We have just released radicli 0.0.26 that should address the issue reported.
If you'd like to try it out, you'd need upgrade radicli in your virtual environment:

python -m pip install --upgrade radicli

Thanks again for the report!

4

@magdaaniol Thank you for the fast response! I can confirm that upgrading radicli resolved the issue.

5

Great! Thanks for the conformation :slight_smile: