Client-Side Logics

Errors and client side

Problem the first:

the store entry is no longer authoritative on the reply status. EBITTEST (E_ABORT) is no longer a valid test outside of client_side_reply.c. Problem the second: resources are wasted if we delay in cleaning up. Problem the third we can't depend on a connection close to clean up.

Nice thing the first:

Any step in the stream can callback with data representing an error. Nice thing the second: once you stop requesting reads from upstream, upstream can be stopped too.

Solution #1:

Error has a callback mechanism to hand over a membuf with the error content. The failing node pushes that back as the reply. Can this be generalised to reduce duplicate efforts? A: Possibly. For now, only one location uses this. How to deal with pre-stream errors? Tell client_side_reply that we want an error page before any stream calls occur. Then we simply read as normal.

Persistent connection logic:

requests (httpClientRequest structs) get added to the connection list, with the current one being chr

The request is immediately kicked off, and data flows through to clientSocketRecipient.

If the data that arrives at clientSocketRecipient is not for the current request, clientSocketRecipient simply returns, without requesting more data, or sending it.

ConnStateData::kick() will then detect the presence of data in the next ClientHttpRequest, and will send it, restablishing the data flow.