Both should be cdevs, not files, so they do not go through the normal filesystem I/O pathway in the kernel and should support select()/poll(). ("ls -l", the first character should be "c" instead of "-" indicating character-mode device nodes.) If ghc is not detecting that, then *that* is indeed an I/O manager issue.

The issue here is that if you look at the source of fdReadBuf, you see that it's a plain system call without any reference to GHC's (relatively new) IO manager.

Best,

Leon