this gets assigned objects like stdout or stderr, which break logging when frozen; this should therefore be a case of a known exception, considering that the main goal of this is to make options shareable across ractors

so there's only 1 object shape

fix: when an ipv6 address gets added to the tcp addresses list, ip index must be updated to continue pointing to the same address

so that recalcuation on adding new addresses works the same

fix: on connection coalescing, do not merge ssl session if the main connection already has one ssl session established

ssl context object gets frozen after initialization and cannot therefore be changed

Fixes https://github.com/HoneyryderChuck/httpx/issues/101

when addresses from the list expire, always assume that one of them is the current one, and point to the last element of the list (or none)

usually, a connection connects when there are available IPs to try.
in a persistent connection scenario, the same holds. however, in a
fiber-based environment, fiber 1 may wait for request 1, while fiber 2
may, while attempting request 2, receive a GOAWAY frame, and initiate
termination/reconnection/resend while switching on the DNS request
phase,
whereas when the context is switched back to fiber 1, request 1 has
already
failed due to what happened in fiber 2 and should be reattempted,
however the
connection is still waiting on the DNS answer, so the reattempt should
just be enqueued.

This fixes by relying on the assumption that an idle connection with no
addresses in the select loop is a connection that has been sent back to
the name resolution loop, so the flow needs to ignore it and go back to
the request handling loop, which will forward the request-to-retry and
wait on the same resolution.

a case was reported where a goaway error was received with a cancel error code; this code seems to have been misused by the peer server, since it is a stream error code, however the client should react to it by just closing it anyway, as the http2 parser already moved on to half-closed anyway, so not accepting new streams anymore

fixes github.com/HoneyryderChuck/httpx/issues/102

fixes to socket addresses expiration rebalancing

See merge request !407

in case there was misalignment when doing HEv2, this brings ipv6 address forward in case the socket started with ipv4 addresses and got ipv6 addresses meanwhile

this cascades down terrible since we're using entries as proxies for strings...

this requires calling , which calls the  syscall, which may be blocked at boot time in certain constrained environments

this is moved into a module function, which memoizes the result; it's done in a non-thread safe way, but since the result is expected to be the same, it should be ok to race

native resolver: do not rely on resolver supported ip families if options for ip_families is explicitly used

option from user takes precedence

fix: do not check if multi-homed network at boot time, instead do it a runtime

See merge request !408