Might be worth considering replacing the union of Union[io.TextIOBase, io.BufferedIOBase] with io.IOBase.

I don't know about this - as pyright already complains, this breaks a few places that expect bytes but now receive bytes | str. Making disnake.File generic over the self.fp type could fix typing issues but still doesn't really help enforce it at runtime.

For the io.StringIO case in particular: aiohttp effectively handles that using stream.read().encode() (code), arguably something that's easy to do in user code as well.

I don't know about this - as pyright already complains, this breaks a few places that expect bytes but now receive bytes | str. Making disnake.File generic over the self.fp type could fix typing issues but still doesn't really help enforce it at runtime.

For the io.StringIO case in particular: aiohttp effectively handles that using stream.read().encode() (code), arguably something that's easy to do in user code as well.

So you suggest not to change the typing, and instead have the user do something like disnake.File(BytesIO(text_file.read().encode()))?

So you suggest not to change the typing, and instead have the user do something like disnake.File(BytesIO(text_file.read().encode()))?

I can't think of many scenarios where one can get a string stream but not a bytes stream, in the simplest case it's a matter of changing open(..., "r") to open(..., "rb").
From the looks of it, this is primarily a typing issue and supporting string streams in disnake.File isn't trivial, and there's a tradeoff to be made between library complexity and usability :/
Perhaps @onerandomusername or @EQUENOS can chip in, there might be other ideas for resolving this.

I can't think of many scenarios where one can get a string stream but not a bytes stream

My primary use case is creating an in-memory text file from a string with io.StringIO; It's useful when message length surpasses 2000 char limit

Just commenting for the record, I'd also plus 1 this feature as something I use within bots to save arbitrary writes to disk

@shiftinv I forgot about this PR; how about we special-case io.StringIO as acceptable fp to File, and encode the contents to BytesIO

Since this topic came up again yesterday, I looked into this more; somehow missed the last comment here, sorry.
I've partially implemented both approaches (generic file/encode bytes->str) here, and the Generic one is pretty jank, to say the least.

how about we special-case io.StringIO as acceptable fp to File, and encode the contents to BytesIO

Yeah, this seems like the better option. aiohttp ends up doing almost the same thing internally if you pass StringIO directly, so this should be fine?

Do we have anything further on this PR? Seems like a usable resolution is fairly close