FYI, this came about when researching this build failure:

https://ci.appveyor.com/project/harendra-kumar/unicode-transforms/build/1.0.36

CC @harendra-kumar

@snoyberg thanks for the PR. I think there is at least one use of readFile somewhere too. Patches welcome.

@snoyberg A side effect of this change is that we loose newline conversion on Windows.

While I assume it's sensible that the .cabal-file is encoded with UTF-8**, newline conversion is usually done by Git, so we may prefer to use the systems native newline handling when writing / reading files.

** Still, right now I'm puzzled what encoding Cabal uses when reading .cabal files...

** Still, right now I'm puzzled encoding Cabal uses when reading .cabal files...

package.cabal:
    a Unicode UTF-8 text file containing a package description. For details of the syntax of this file, see the section on package descriptions.

https://www.haskell.org/cabal/users-guide/developing-packages.html#creating-a-package

So that is fine.

Just chiming in to say that Cabal files are indeed UTF-8 in practice (with one exception). The currently unreleased Parsec-based parser will use byte strings instead of strings. That explicitly uses UTF-8 behind the scenes.