Pierce Lopez

Something is sending a plain-http request to a port expecting https. Look in nginx and oauth2_proxy logs for a hint as to where exactly this is happening, then you might...

This change matches a HEAD request (which had Upgrade headers or for which the user returned 2 from the `on_headers_complete` callback). A HEAD _response_ may have a `Content-Length` but does...

see #348 and my own #321

this should be taken care of already by ```c http_parser_url_init(struct http_parser_url *u) { memset(u, 0, sizeof(*u)); } ```

The user of the library should call that before calling `http_parser_parse_url()`. Or, alternatively, the `struct http_parser_url` might be a member of a larger struct that was `memset()` all at once,...

Looks like a bug in the test to me. I'm not a maintainer though, just trying to help clarify the situation.

it's worth noting that this costs an additional 8 bytes per `struct http_parser`

you may be able to do something like: ```c static const uint8_t method_lengths[] = { #define XX(num, name, string) uint8_t(sizeof(#string) - 1), HTTP_METHOD_MAP(XX) #undef XX } ```

(but there may be a slightly less expensive way to just error-out if the request line overall is too big ...)

A new algorithm, "Errol3" (and presumably minor update "Errol4") has been developed, which is apparently half the speed of "Grisu3", but always produces the optimal decimal string output (Grisu3 only...