Fix $PATH ordering in README

The README currently suggests to set your $PATH like so:

export PATH="$HOME/.cabal/bin:$HOME/.ghcup/bin:$PATH"

However, this means the system-installed cabal overrides the ghcup installed cabal. This recently bit me. Instead, I believe the $PATH should be set like so:

export PATH="$HOME/.ghcup/bin:$HOME/.cabal/bin:$PATH"

However, this means the system-installed cabal overrides the ghcup installed cabal. This recently bit me. Instead, I believe the $PATH should be set like so:

This is on purpose! Because a cabal binary installed freshly from hackage on your system is likely more recent and desirable.

closed

My experience was migrating from a stack-managed installation of cabal to ghcup-managed installation. In my case, I was setting up a project that used Cabal 3.x, which Stack does not yet support.

I was getting some incompatibility errors with building my project before I realised that, even though I had run ghcup install-cabal, which cabal was pointing at my Stack-managed installation of Cabal 2.x.

In my case, having the guarantee that "my cabal is always in sync and compatible with my ghcup-installed ghc" would have been more valuable than "my cabal is the one I last installed".

The behaviour that I expect coming from other version managers like NVM is for every ghc version to have its own persistent (i.e. "if I'm running GHC X which by default installs Cabal Y and I upgrade to Cabal Z, the next time I switch back to GHC X, I'm using Cabal Z"), compatible cabal. Is this an intended goal?

ghc and cabal-install are not in sync at all currently. And why would they?

It should be backwards compatible. There is no switching of cabal-install implemented at all. It's purpose is bootstrapping, mainly.

Once you bootstrapped it you want to be able to install any version you want from hackage.

Once you bootstrapped it you want to be able to install any version you want from hackage.

I was under the impression that specific versions of Cabal are only compatible with specific versions of GHC, but it sounds like I might be wrong here. Thanks for clarifying.