GHC 9.4 on Windows switches from a GCC-based toolchain to a Clang-based one (see [here](https://gitlab.haskell.org/ghc/ghc/-/issues/21019)). One consequence of switching to Clang is that `System.EntropyWindows` will fail to preprocess. For the full details, see the commentary on the GHC issue tracker at [GHC#21668](https://gitlab.haskell.org/ghc/ghc/-/issues/21668). The short version is that this comment in `System.EntropyWindows`: ```hs {- C example for windows rng - taken from a blog, can't recall which one but thank you! #include <Windows.h> #include <Wincrypt.h> ... // // DISCLAIMER: Don't forget to check your error codes!! // I am not checking as to make the example simple... // HCRYPTPROV hCryptCtx = NULL; BYTE randomArray[128]; CryptAcquireContext(&hCryptCtx, NULL, MS_DEF_PROV, PROV_RSA_FULL, CRYPT_VERIFYCONTEXT); CryptGenRandom(hCryptCtx, 128, randomArray); CryptReleaseContext(hCryptCtx, 0); -} ``` Will be left alone by GCC's preprocessor with `-traditional-cpp` enabled (which is a flag that GHC passes to the C preprocessor), but Clang's preprocessor will expand the `#include`s with `-traditional-cpp` enabled, leading to pandemonium. This is arguably a bug in Clang, and indeed, there is a FIXME about this particular preprocessor quirk in the Clang test suite [here](https://github.com/llvm-mirror/clang/blob/aa231e4be75ac4759c236b755c57876f76e3cf05/test/Preprocessor/traditional-cpp.c#L56-L58). That being said, it seems unlikely that this will be fixed any time soon. I've adopted a more expedient fix of switching from `{- ... -}` comments to `--` comments, which prevents the `#include`s from being expanded on both GCC's and Clang's preprocessors.