Hi, I am experimenting with the `<http://kotlinx.io|kotlinx.io>` library in Ktor 3, but having this ...

# ktor

[Arjan van Wieringen]

Hi, I am experimenting with the <http://kotlinx.io|kotlinx.io> library in Ktor 3, but having this issue:

call.respondSource(
   source = asset.output.source, // <--- this is a Source
   contentType = asset.output.contentType
)

This returns an EMPTY response.

call.respondBytes(
    bytes = asset.output.source.readByteArray(),
    contentType = asset.output.contentType
)

This doesn't. This hardly makes sense to me.

When I set a breakpoint in the first example before the respondSource and run the following expression: asset.output.source.readText() I ge the full text (once, of course).

[Arjan van Wieringen]

Okay, so in tests it works, but not in a real application. I filed a bug including a very minimal reproducer:
https://youtrack.jetbrains.com/issue/KTOR-7683/call.respondSource-returns-empty-response-but-passes-in-tests

It is very worrying to me that the tests behave differently than running an actual browser.

[Bruce Hamilton]

Hey, it looks like respondSource relies on Source.transferFrom which expects the input to be fully buffered into memory. I'm not sure if this was a mistake in Ktor's use of kotlinx-io or an issue in kotlinx-io itself. I added a workaround in the ticket using Buffer , which is not ideal, but we ought to fix the problem for the next patch release.

[Arjan van Wieringen]

Great. But why would it work in a test then?

[Bruce Hamilton]

there's gotta be something in the pipeline that lets the inputstream buffer it's content before hitting the transfer

[Arjan van Wieringen]

That makes sense. The testApplication is completely different Engine than Netty ofcourse.