ArrowStreamReader.GetSchema (async) does synchronous IO when stream is empty

[spanglerco]

Describe the bug, including details regarding any error messages, version, and platform.

When given an empty stream, calls to ArrowStreamReader.GetSchema() first calls ArrowStreamReaderImplementation.GetSchemaAsync() and then retrieves the Schema property. When the stream is empty, GetSchemaAsync returns early without setting _schema.

Then, the Schema property getter, seeing that _schema is still null, calls ArrowStreamReaderImplementation.GetSchema() (the synchronous version). This ends up performing a synchronous read on the (still empty) stream.

In particular, this is an issue when the stream is an HTTP request body from Kestrel (ASP.NET), which blocks synchronous IO by default. The result is an InvalidOperationException.

I'm willing to provide a PR if someone else doesn't get to it. My initial reaction would be to change ArrowReaderImplementation.GetSchemaAsync() to return a ValueTask<Schema>, but that's not the only possible solution.

[CurtHagenlocher]

What would you say is the correct behavior here? To return a null Schema to the external caller when ArrowStreamReaderImplementation.ReadMessageLengthAsync returns zero?

[spanglerco]

That's what it currently does (after the synchronous IO), which seems reasonable to me. It used to throw an exception until apache/arrow#47098.