Lol, the fact that they just went with IPv4 supports the theory that they know what they're doing.

I've read multiple tutorials on IPv6 and it's so complicated that I still can't grasp how it all works in my head. For some context, I've personally written an Ethernet IPv4 network stack with DHCP from scratch. I also worked on Ethernet switch firmware, coincidentally for SpaceX...

I guess I don't have as much low level experience as you do, but in my eyes IPv6 is literally just IPv4 with more bits. Due to the increased number of possible addresses, IPv6 is used with some different (better) concepts than IPv4. For example having (very likely) unique local networks (ULAs) is amazing when you happen to merge previously independent networks together or do some VPN stuff. You pretty much never have to calculate subnets because you'll likely never run out of addresses. And even if you somehow manage to do the impossible and fill up a /48 ULA (at which point you would have died from NATing billions of entire IPv4 address spaces with multiple layers to somehow make it all one network on IPv4), it takes just two seconds to generate a new one.

Also getting an entire public prefix assigned instead of just a single address is amazing too. On my public servers I let my ssh service listen on a randomly generated static IPv6 address from my prefix, which is not used for anything else. On that same machine I can have dozens of other public services on their own IP each (they can even all run on the same port because they're using different addresses), and when someone then decides to scan the IP of my website for interesting ports, they won't find anything.

To be honest, if someone doesn't understand IPv6, I doubt they really understand IPv4 either. Many issues with IPv6 today come from the fact that everyone and everything is so used to V4 concepts, that most products focus on IPv4 first and then half-ass IPv6 support (if they implement it at all).

The thing about IP and therefore with IPv6 is it's not just an address format - it's a set of protocols and practices so a lot of it changes.

Address asignment in an IPv6 world is more complicated - you can do DHCPv6, or your can do stateless auto-config. Which should you choose and when? Many other aspects of the stack require more knowledge.

And THEN you ALSO need to understand how v4 and v6 interacts so you can provide suitable compatibility - the various different methods of tunnelling and negotiation.

Using SLAAC is the way to go in most cases. DHCPv6 should only really be used when you have a network that is for some reason smaller than a /64.

In my opinion you shouldn't really bother with compatibility. Just do dual stack and maybe NAT IPv4 if necessary.

For my personal use cases IPv6 just proved to be much more versatile than IPv4. It did take some getting used to since it's approached differently, but learning it is very worth it and also helps using IPv4 better

Look at all the additional optional bits DCHP offers in terms of being able to communicate additional information to clients to configure themselves. Two big ones that stand out are timeservers and network booting parameters.

SLAAC is only a drop in replacement in the (typical) home.

That's just the addressing/subnetting though. Ipv6 and 4 for that matter also have a lot more going on in the packet handling and specification. and Ipv6's packet handling stuff is way more complicated than Ipv4. I was reading through just the ICMP parts and the difference is significant. I imagine the rest of it is the same.

> it's so complicated that I still can't grasp how it all works

I have seen too many people that have the same excuse. The advise I gave them is to forgot all the things learnt from IPv4 and study IPv6 as a totally new IP stack.

The problem is people look at it from IPv4 view, which is a currently a mess with IPv4 exhaustion.

Going through the trouble of implementing CGNAT and skipping IPv6 seems like a partially degraded service for the end user though. It's becoming common practice for ISPs to slap all the users between two layers of NAT and calling it a day. You can't run essentially anything on your network (not even an SSH server) without having to have a server on standby and reverse tunnels. I know there are not enough IPv4s but at least let me having the option to pay for one. Or, you know, give me an IPv6. Otherwise it's practically a networked content consumption service at this point. It's 2021, the "IPv6 is complicated" excuse must stop at some point.

Honestly, it's not that complicated to set up a L2TP or VPN tunnel to a server with routable inbound IP traffic.

Most people who would know how to set up port forwarding could probably set it up. I think even some routers support it built in now.

That is additional latency, especially if you can't find a VPS near your ISP.

Fair point but starlink ground stations are all near major PoPs where you won't struggle to find one.

IPv6 really isn't that complicated compared to CGNAT!

Of course, today IPv6 only can't be considered full internet access, so providers usually need both – and given that CGNAT gets consumers 99% of what they want/are willing to pay for, providers sometimes can't be bothered to also implement IPv6.

I used to have GCNAT + IPv6 on my provider, but the v6 part of their CPE was worse than useless – no option to open ports since "port forwarding is not needed on IPv6" (nevermind the mandatory firewall that doesn't allow inbound traffic unless completely disabled).

It can’t? I at one point messed up my firewall rules and didn’t notice that IPv4 wasn’t accessible for a few days. Nobody complained. I eventually noticed something was amiss and fixed it but really most places seems to be running dual stack. I am starting to run some IPV6 only services.

It would be great if offering public IPv6 only services was viable, but according to Google statistics (https://www.google.com/intl/en/ipv6/statistics.html) you're still cutting off about two thirds of potential users.