Fix Unicode output in cabal-testsuite

cabal-testsuite/src/Test/Cabal/Monad.hs

@@ -402,8 +402,8 @@ runTestM mode m =
  testSkipSetupTests = argSkipSetupTests (testCommonArgs args),
  testHaveCabalShared = runnerWithSharedLib senv,
  testEnvironment =
- -- Try to avoid Unicode output
- [ ("LC_ALL", Just "C")
+ -- Use UTF-8 output on all platforms.
+ [ ("LC_ALL", Just "en_US.UTF-8")
  -- Hermetic builds (knot-tied)
  , ("HOME", Just (testHomeDir env))
  -- Set CABAL_DIR in addition to HOME, since HOME has no

cabal-testsuite/src/Test/Cabal/Run.hs

@@ -54,6 +54,29 @@ runAction _verbosity mb_cwd env_overrides path0 args input action = do
  mb_env <- getEffectiveEnvironment env_overrides
  putStrLn $ "+ " ++ showCommandForUser path args
  (readh, writeh) <- createPipe
+
+ -- `System.Process.createPipe` calls (through many intermediaries)
+ -- `GHC.IO.Handle.FD.fdToHandle`, whose documentation says:
+ --
+ -- > Makes a binary Handle. This is for historical reasons; it should
+ -- > probably be a text Handle with the default encoding and newline
+ -- > translation instead.
+ --
+ -- The documentation for `System.IO.hSetBinaryMode` says:
+ --
+ -- > This has the same effect as calling `hSetEncoding` with `char8`, together
+ -- > with `hSetNewlineMode` with `noNewlineTranslation`.
+ --
+ -- But this is a lie, and Unicode written to or read from binary handles is
+ -- always encoded or decoded as Latin-1, which is always the wrong choice.
+ --
+ -- Therefore, we explicitly set the output to UTF-8 to keep it consistent
+ -- between platforms and correct on all modern computers.
+ --
+ -- See: https://gitlab.haskell.org/ghc/ghc/-/issues/25307
+ hSetEncoding readh utf8
+ hSetEncoding writeh utf8
+
  hSetBuffering readh LineBuffering
  hSetBuffering writeh LineBuffering
  let drain = do