(Imported from Trac #692, reported by @ezyang on 2010-05-24)
They seem innocuous (usually hiding in source code comments, lest they fail to compile) but they make uniformly loading *.hs files into memory a pain, since you have to heuristically determine what encoding a file that is not Unicode-enabled is. We should ban them: it's UTF-8 or nothin'.
(Imported comment by @dcoutts on 2010-05-24)
GHC since version 6.10 (if I recall correctly) loads .hs files as UTF8. Why does it not fail when the package is compiled? We should not need an additional specific check.
(Imported comment by @ezyang on 2010-05-25)
I believe that GHC will ignore invalid UTF8 sequences that are contained in comments.
This is not a Cabal issue.