[Sean-Der]

This trend has been bothering me a lot lately.

In standards bodies you will have a 'True Believer' about a particular topic. They will push it for years and years, and eventually get their way. The idea isn't bad, and would be great if included in the spec from day one. Unfortunately adding it to what we have today causes massive breakage/incompatibility.

Maybe I am just being resistant to good change. It just is frustrating because in most cases the 'True Believer' isn't going to go worry about the real-world impact of the change.

This phenomena seems to be hurting a lot of programming languages as well. So much harder to say no, and idealistic people are always going to find a way :)

[woodson]

I like to see the other side (that of the true believer) as well. If all (or most) can agree on that it is the right thing to do and the current way of doing it is wrong, why oppose the change for 20 years saying it will cause breakage? If the change had been made at the beginning, any breakage would have been dealt with long ago. (I’m specifically not talking about change for the sake of making changes.)

[m463]

I think, like python 2->3, the chaos and confusion of trying to pull off a "clean slate reboot" is very rarely successful or worth it.

[Dagger2]

There wasn't really any way to avoid what v6 did though. v4 can't handle addresses that are bigger than 32 bits, and that's the end of it.

v6 also has pretty much every backwards compatibility mechanism that can work with v4. It's hard to see how it could've done any better, and nobody I've ever talked to has managed to come up with anything that a) would work and b) isn't already a thing v6 does or can do.

I've seen plenty of proposals that don't satisfy those two conditions (like, "just add an octet" or "just make the numbers go up to 999")...