Even though the library request HTTP/1.1, there are some servers (cough api.github.com) that returns HTTP/1.0. This is immediately "rejected" and infinite loops the event handler:

ClientEvent{ .invalid = Invalid{ .buffer = HTTP/1.0, .message = expected HTTP/1.1, .state = ParserState.initial } }

ClientEvent{ .invalid = Invalid{ .buffer = 403, .message = expected HTTP/1.1, .state = ParserState.initial } }

ClientEvent{ .invalid = Invalid{ .buffer = Forbidden
Cache-Control:, .message = expected HTTP/1.1, .state = ParserState.initial } }

// -- snip --

ClientEvent{ .invalid = Invalid{ .buffer = causes.
, .message = expected HTTP/1.1, .state = ParserState.initial } }

ClientEvent{ .closed = void }

ClientEvent{ .closed = void }

ClientEvent{ .closed = void }
// -- snip --

I think hzzp should support 1.0 parsing since afaik it's a strict subset of 1.1. Plus we should fix the bug where it infinite loops — maybe invalid should be an error?

Hzzp does not currently support parsing trailing headers on chunked messages. This is part of the HTTP/1.1 and should be parsed.

Todo

• Figure out how the trailing headers will be emitted. (should .header be emitted after .head_complete?)

(comment from it):

Design criteria:

• the same header field is allowed more than once
• must be able to fetch separate occurrences (important for some headers e.g. Set-Cookie)
• optionally available as comma separated list
• http2 adds flag to headers that they should never be indexed
• header order should be recoverable

Headers are implemented as an array of entries.
An index of field name => array indices is kept.