Repository navigation
Missing feature: sendfile #45
Description
On Unix, os.sendfile is present and just needs a simple (but slightly finicky) wrapper.
On Windows, os.sendfile doesn't exist, but we can wrap TransmitFile - shouldn't be too hard.
Reacted by Gabe Joseph and Alexey Shokov
Activity
njsmith commentedon Jan 19, 2018
Asyncio support is coming in 3.7: https://bugs.python.org/issue32410
sseg commentedon Apr 14, 2018
I started having a look at this but I'm not sure how you intend the API. Do you think that
trio.socket.sendfileshould acceptAsyncIOWrapper, plainfileobjects, or both?njsmith commentedon Apr 14, 2018
Oh shoot, that's an excellent question. I was just thinking about wrapping
sendfilein the obvious naive way, like we wrapsendor whatever, but I forgot that this may cause synchronous disk access. And it's a bit tricky, because we could route it through a thread like we do with other disk access, but in this case that might add enough overhead to defeat the purpose of usingsendfilein the first place. It's also unclear whether a thread is really necessary on Windows – maybeTransmitFiledoes async disk I/o natively?It would help to have more information. I guess one way forward would be to start by making a toy little benchmark that uses
sendfilein the naive way on Linux, and compare it toread+send, to see what benefit we can even hope to get?pipehappy1 commentedon Jan 17, 2020
Do we have any update on this? Having sendfile in the API would be nice.