Change GHC support window unit from years to {major versions, LTS, …)

[ffaf1]

See #11530

The gripe some people have with out support window is that it is expressed in years. Options proposed in the thread are:

1. x GHC major versions.
2. x GHC LTS major versions.
3. "cabal-install from the last non-EOL'd Ubuntu LTS (Debian stable, Fedora, whatever is your benchmark) should be able to complete cabal install cabal-install".
4. Keep it as it is (years).

See #11723 on whether to make our window narrower.

[ulysses4ever]

Metacomment: It's really hard to come up with a legitimate number (of years) backed by anything objective, which is the major problem in this discussion. In this setting, most people will lean towards preserving the status quo. That's sort of a bias to inaction, as I call it. We should acknowledge that the status quo doesn't have much backing as well and, so, isn't any better objectively than anything else. Maybe I'm wrong, and there is such backing? That would be good to know.

I was in the status quo camp, as many others, maybe just because of the bias I mentioned above. But lately, I happened to encounter several PRs and discussions where the conservative 5-year boundary is holding back new and noteworthy contributions. (Particularly some PRs from @zlonast). And I started remembering that it does happen once in a while that some old version of GHC on Windows or whatever holds us back. And maybe it will bring some relief for us and, more importantly, the contributors, to cut that window a bit. It will improve CI times as well.

Again, I don't have a rigorous way to define the bound. People quoted things like GHC and HLS, which have much smaller compatitility windows, but it's pretty obvious that those projects are in a completely different boat (GHC is an app that people don't compile most of the time, and HLS uses the GHC API, which makes back-compat very painful).

So, I will throw a random number at you: how about we cut it from 5 to 3?

[ulysses4ever]

Oh, after submitting my pich, I realize it's a wrong issue. I think switching to something other than years was contested in a more principled way than the change of 5 years to N (N<5) years. So, I'd like to discuss the latter somewhere...

[ffaf1]

Sure, I have created #11723, you can comment there.

Of course we are interested in your opinion on this one too.