SHUTDOWN_COMPLETE event happening before all received data could be processed

[CarnaViire]

I am investigating failing tests in System.Net.Quic library that uses msquic. The tests are very basic, there is a peer sending some data to an endpoint (let me call it my app) in a unidirectional stream and then indicating a graceful shutdown. After that, as soon as my app's msquic acks this shutdown, SHUTDOWN_COMPLETE event appears on both sides. On my app side though, this event doesn't seem to pay attention to pending RECEIVE data (the one that got buffered by msquic but is yet to be read by the app). It may be important that the app reads the data asynchronously (i.e. returns QUIC_STATUS_PENDING, and then calls to StreamReceiveComplete and StreamReceiveSetEnabled when finished)

There are two scenarios in which I got SHUTDOWN_COMPLETE event in the middle of processing the data.

1. When the reads are complete

(peer sends e.g. 10K and I receive in a buffer of the same size)

I sometimes will get SHUTDOWN_COMPLETE right away after the last RECEIVE -- even before I'm able to call StreamReceiveComplete, and in this case StreamReceiveComplete will fail. Events will go as follows:

[PEER] data sent
[PEER] received event SEND_COMPLETE
        [APP] received event RECEIVE
[PEER] shutdown sent
        [APP] read started
[PEER] received event SEND_SHUTDOWN_COMPLETE
[PEER] received event SHUTDOWN_COMPLETE
        [APP] received event SHUTDOWN_COMPLETE
        [APP] read finished, will call StreamReceiveComplete
[PEER] connection closed
        [APP][ERROR] message: Exception occurred during connection callback: The channel has been closed.
        [APP][ERROR] System.Net.Quic.QuicException : Could not complete receive call. Error Code: INVALID_STATE

2. When reads are partial

(peer sends e.g. 10K and I read data in an 1K buffer several times)

In this case, I don't get any errors, but I lose data. E.g. I will read 3K and then I'll get a SHUTDOWN_COMPLETE event. After that there is no way to get the rest of data, even though it was already buffered inside msquic.

Expected behavior

Is it possible for PEER_SEND_SHUTDOWN/SHUTDOWN_COMPLETE to behave in line with RECEIVE events (so I'll get it after I notify msquic that previous receive is completed)?

cc @nibanks