CHANGELOG

[0.2.3] - 2026-01-11

Use raw ICMP sockets on Linux because Linux does not send all ICMP packets to all open ICMP datagram sockets like macOS does.

Fix IPv6 protocol selection on Linux. It had :"IPV6-ICMP" hard-coded, but that's macOS only.

Fix a race condition in ICMPConn where the connection could terminate prematurely. This occurred when the last registered process terminated while another process was in the middle of registering (between calling ICMPConn.get_or_start_conn/1 and ICMPConn.register/4).

Completely rewrite how ICMP packets are received.
- Previously, every probe would start its own ICMP socket and ignore packets that were meant for other ICMP sockets based on the identifier of the received packet. However, not all ICMP sockets received all packets, so Traceroute would return many more TimeExceeded responses than it should. Now, we start only a single ICMP socket (one for IPv4 and one for IPv6) upon request and receive and send all packets through that singleton socket.
Fix DestinationReached probe count.
- We previously ignored probes that did not receive a "destination reached" response but had the same TTL as probes that did. We now include such probes in the response because although they didn't receive the correct ICMP response, they most likely still reached the destination.

Add min_ttl option to start the traceroute at a higher time-to-live than 1, which makes it easier to skip the first hops (e.g. your local router)
Return Probe.source_domain as string instead of charlist.