Repository navigation
Datagrams sent to an IP literal, and a v6only switch for UDP sockets - #9
Open
neutrino38 wants to merge 6 commits into
Open
neutrino38 wants to merge 6 commits into
neutrino38 wants to merge 6 commits into
Conversation
A URI authority and a Host header carry an IPv6 literal inside square brackets (RFC 3986 §3.2.2), and that is the form a caller holding an address taken from a URI has in hand. parse/1 answered nil on it, so every such caller had to strip the brackets itself, and the one who forgot dialed the wrong address family with no diagnostic. Only a v6 address is written that way: "[127.0.0.1]" stays nil, and so does a bracket that never closes.
connect/3 turned every binary into a charlist and handed it to :gen_tcp.connect/4, which resolves it as a name in the socket's family. "::1" therefore came back :nxdomain on a socket nobody had declared :inet6, while a tuple carries its own family. Parse first, and fall back to a charlist only for what really is a host name. The @SPEC already announced :inet.ip_address(), which the code could not handle. It now says Socket.Address.t(), which it does.
connect/3 carried the same charlist conversion as Socket.TCP.connect/3, with the same effect: the literal went to the resolver instead of to the socket. Its test reads a failure rather than a success — no TLS server answers on the other end — because :nxdomain is what the resolution path returns and a handshake timeout is not.
Socket.Datagram.send/3 takes a destination address in every form Socket.Address.t() covers, and a literal goes to :gen_udp as a tuple. Two things follow. The bracketed form a URI or a Host header carries, "[::1]", now reaches its peer. And a destination given as a literal costs no name resolution, on any datagram.
open/2 takes `v6only: true | false` and maps it to the socket's ipv6_v6only. It is the last IPv6 setting the wrapper could not spell, and an option it does not know raises rather than passes through. It matters beyond the socket: an IPv6 socket that also accepts IPv4 receives from those peers under ::ffff:a.b.c.d, and every protocol message that writes a peer address then carries that form. Only the socket can be told otherwise.
This was referenced Nov 2, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Base. This branch sits on the TCP/SSL literal PR #8,
which itself sits on #7.
Take the three in that order.
Two changes, one commit each.
Socket.Datagram.send/3takes a destination in every formSocket.Address.t()covers: a tuple, a literal such as"::1", the bracketed"[::1]"that a URI authority or aHostheader carries, or a host name. Aliteral is sent as a tuple. Only a real host name is resolved.
Socket.UDP.open/2takesv6only: true | false, which sets the socket'sipv6_v6only. Oneopen/2call now expresses the family, the bind address andthis switch.
Tests
Two tests in
test/ipv6_test.exsare addedElixir 1.18.3 / OTP 26:
mix format --check-formattedpasses,mix testis 16tests / 0 failures over four runs, and
mix compile --warnings-as-errorsfailswith exactly the nine warnings
masteralready has, at the same sites.