search

e2b-dev / code-interpreter

Python & JS/TS SDK for adding code interpreting to your AI app
https://e2b.dev
Apache License 2.0
1.09k stars 76 forks source link

Sandbox.upload_file assumes io.FileIO #36

Open bengladwell opened 3 weeks ago

bengladwell commented 3 weeks ago

The Sandbox.upload_file method has the following signature:

def upload_file(self, file: IO, timeout: Optional[float] = TIMEOUT) -> str:

However, it's not true that file must simply be type IO. Because .name is called on that object:

        filename = path.basename(file.name)

it seems that the method actually expects the object to be of type FileIO, a subtype of IO and the type that you get when you use the builtin open function.

This is unwieldy when dealing with a file that is not stored on disk. In that case, I have access to the file's bytes and I have its filename. But I can't create a FileIO object to pass to upload_file without writing disk (creating a FileIO automatically writes to disk). In the meantime I am creating a subclass of BytesIO that has a name property. So, the duck type satisfies the requirements of upload_file. But I think it should be straightforward to upload a file that is not on disk.

ValentaTomas commented 2 weeks ago

Hey @bengladwell, thank you for creating the issue! This is something we are fixing in the Beta SDK (https://github.com/e2b-dev/E2B/blob/beta/packages/python-sdk/e2b/sandbox_sync/filesystem/filesystem.py#L102) because you are correct that this is very unwieldy.

The Beta releases of the Core SDK and Code Interpreter SDK should have this change already (https://e2b.dev/docs/guide/beta-migration#uploading-data-and-creating-files-in-sandbox).