Detecting closed connection race condition

Version
current git master

Platform
Linux (any really)

Description
I'm using hyper with curl.

When we do multiple transfers with curl it tries to reuse a connection used in a previous transfer. Persistent connection style. It has a "pool" of previously used connections.

Any such connection in the pool might be closed or otherwise "die" at any given moment.

When curl wants to send a request on a reused connection there is always a risk that it was closed (by the server) just the millisecond before we send the request over the connection.

A key information piece to detect this state is that curl is not able to read a single byte from the connection. If it could read a byte from it, it was not a race condition.

I struggle to get hyper to tell me this information. When hyper deems the response "illegal" very early in the response I get an error back, but how do I know if it errors before reading a single byte?

HYPERE_UNEXPECTED_EOF seems to be returned for more problems than just this.

No source file, test, or entry point is named. Start by reproducing the connection-reuse race with hyper and curl, then trace how early response errors are surfaced. Done means establishing and testing a reliable distinction between EOF before any byte is read and other unexpected EOF cases.