Support non-BMP characters (or, surrogate pairs) on Windows

[minoki]

Currently, haskeline does not properly handle surrogate pairs on Windows. This leads to issues like

• GHCi panics on Windows when terminal set to UTF-8 and four-byte characters are used
• Tab-completing in a directory with Unicode heiroglyph crashes ghci.

---

This PR consists of three parts:

• Writing to console

The String must be UTF-16 encoded when calling WriteConsoleW. Otherwise, the program crashes at toEnum here.

Note that my patch doesn't care WriteConsoleW's buffer limit. Actually, I tried running

main = runInputT defaultSettings $ outputStrLn $ replicate 20000 '\x1F986'

on my Windows machine (Win10 Pro 1903), but WriteConsoleW seems to have succeeded. If problem arises on older Windows, the patch may need to be reconsidered.

• Reading from console

Windows sends two input events for a non-BMP character: lead surrogate, trail surrogate. So we need to decode them.

• wcwidth

Since haskeline calls wcwidth on the prompt string, wcwidth must also be able to handle non-BMP characters. Otherwise, a program like

main = runInputT defaultSettings $ do
    _ <- getInputLine "\x1F986"
    return ()

crashes at toEnum here.

The fix is just changing wchar_t/CWchar to int/CInt when interfacing with the C counterpart (haskeline_mk_wcwidth).

Other C functions in h_wcwidth.c (haskeline_mk_cwswidth, haskeline_mk_wcwidth_cjk, haskeline_mk_wcswidth_cjk) are not modified, because they seem to be unused.

[judah]

Thank you for this fix! I don't have an easily-accessible Windows machine to try it on myself, but the changes all look very reasonable.