Issue:

In an IPv6-only setup, raw IPv6 packets were being sent with an incorrect checksum, whereas they worked correctly in a dual-stack configuration (both IPv4 and IPv6).

Analysis:

• The checksum is calculated in raw.c when the pcb->chksum_reqd flag is set.
• This flag is enabled for all IPv6 raw pcbs in api_msg.c, but only when NETCONNTYPE_ISIPV6(msg->conn->type) evaluates to true.
• The issue arises because the macro NETCONNTYPE_ISIPV6() checks if the NETCONN_TYPE_IPV6 flag is set in the netconn type. This flag wasn’t being set when creating the socket, as the DOMAIN_TO_NETCONN_TYPE() macro returned the original type without adding the IPv6 flag.

Fix:

• Made enum netconn_type reflect protocol family also in IPv6-only configs, updating the macro DOMAIN_TO_NETCONN_TYPE() to set the NETCONN_TYPE_IPV6 flag.

Alternative Approach:

• Another potential solution would be to modify the enum netconn_type to reflect both the connection type and protocol family (as of now) in a dual-stack setup, while reflect connection type only in a single-stack setup (IPv4 only or IPv6 only).
  (i.e. setting NETCONNTYPE_ISIPV6() to true in IPv6 only mode)

PS: saw that you enabled CI runs for new PRs, so maybe you'll accept something from Github, too.
PPS: the main reason for choosing this way of fixing is that the alternative would require semantic changes to the API, but open to suggestions.