[WIP] Fix hiding of errors in SecureChannel.readChunk

[Opsi]

Related to #550

readChunk of secure_channel.go was hiding the errors of Conn.Receive, because of a wrong comparison. len(b) == 0 returns true whenever b=nil.

[magiconair]

Good catch. I would suggest to fix it slightly differently since b != nil is a side effect of err == nil in this case.

if err == io.EOF || err == nil && len(b) == 0 {
    return nil, io.EOF
}

[magiconair]

Hmm, on my local branch the TestAutoReconnection/subscription_failure hangs all of a sudden.

[magiconair]

Hmm, I think this is surfacing a different issue

[magiconair]

have the same issue when I'm applying that patch on v0.3.0

[magiconair]

I can't see it. I'll push this to v0.3.2 since this needs more investigation.

[Opsi]

Sorry I totally overlooked the integration tests. The problem probably has to do with SecureChannel.dispatcher

func (s *SecureChannel) dispatcher() {
    ctx, cancel := context.WithCancel(context.Background())
    defer cancel()

    defer func() {
        close(s.disconnected)
    }()

    for {
        select {
        case <-s.closing:
            return
        default:
            resp := s.receive(ctx) // this is now returning non-io.EOF errors...
                
            if resp.Err != nil {
                select {
                case <-s.closing:
                    return
                default:
                    select {
                    case s.errCh <- resp.Err:
                    default:
                    }
                }
            }
                
            if resp.Err == io.EOF { // therefore this is NOT returning from the dispatcher loop
                return
            }
            ...

[magiconair]

Hmm, when I run the test in isolation then it works every time. But not when I run the entire suite.

[magiconair]

@Opsi can you rebase your branch and fix the integration tests?

[Opsi]

Rebasing is no problem, but I don't if I can fix the integration tests.

1. There seems to be some kind of race condition in uatest/reconnection_test.go TestAutoReconnection/subscription_failure that's (most of the time) causing an endless loop for me on the fourth run (where a "subscription_failure" is simulated)

2. I wasn't able to look into the second error in uatest/timeout_test.go TestClientTimeoutViaContext