We have code similar to the following in our ktor ...

# ktor

[strindberg]

We have code similar to the following in our ktor services where we receive files from the user in an upload endpoint:

route(UPLOAD_PATH) {
    post {
        call.receiveMultipart().forEachPart { part ->
            if (part is PartData.FileItem) {
                call.respond(
                    uploadService.store(part)
                )
            }
        }
    }
}

The stack is fully in ktor, so what uploadService.store() does contains no blocking code. In a code review it was suggested that this code still could be made more robust with wrapping it in a withContext(<http://Dispatchers.IO|Dispatchers.IO>) block like this:

route(UPLOAD_PATH) {
    post {
        withContext(<http://Dispatchers.IO|Dispatchers.IO>) {
            call.receiveMultipart().forEachPart { part ->
                if (part is PartData.FileItem) {
                    call.respond(
                        uploadService.store(part)
                    )
                }
            }
        }
    }
}

Is this really necessary? My understanding of coroutines tells me that this is not an improvement when the stack is coroutines all the way down, but am I wrong?

[madisp]

if uploadService.store does not block then there's no need for IO indeed

[Chrimaeon]

this question would better fit in #C1CFAFJSK, but if store is a suspend function, then the implementation of store is responsible to use the right dispatcher.