Remove typed-paths interface and use FilePath for everything

[patrickt]

Though FilePath is very flawed, especially on Windows, typed paths were a failed experiment. They did not actually catch any bugs; the best thing they did was expose some very weird behavior in certain functions, ones that we were unable to change (without huge renovations) lest command-line invocations break. There is a new FilePath type coming, one based on ShortByteString and officially blessed by a GHC working group, and we should just switch back to FilePath so that we can transition to that when possible.

[hasufell]

My package hpath, which will transition to AbstractFilePath when it rolls out, is a fork of the path package and adds the distinction of absolute/relative only: https://github.com/hasufell/hpath

Right now it only works on bytestring and unix. I have an experimental branch that adds windows support and uses AbstractFilePath: https://github.com/hasufell/hpath/tree/abstract-filepath

It also contains a rather big module with its own IO functions implemented from scratch: https://github.com/hasufell/hpath/blob/abstract-filepath/hpath-io/src/HPath/IO.hs

If you're interested in that approach, I'm open to discuss API. GHCup will switch back to it once all the transitions are ironed out, but it doesn't have any other user afaik.

---

Edit: I also happen to be the filepath maintainer and the implementor of AFPP.

If you want to test it, check out:

• https://github.com/hasufell/file-io for reading/writing files (use the cabal.project contained in that repo, which uses unix/win32 patches)
• clone https://github.com/haskell/filepath/pull/103 and build the haddocks to get an idea of the upcoming API

docs:

• https://hasufell-stuff.eu-central-1.linodeobjects.com/filepath/index.html
• https://hasufell-stuff.eu-central-1.linodeobjects.com/file-io/index.html

[robrix]

@hasufell: This is very cool; thank you for sharing!