can crash with DoesNotExistError on file that exists (encoding issue)

[joeyh]

addWatch can crash if fed a filename that contains a character that is not valid in the current locale. For exampe, in the C locale, the ü character is not valid.

This is due to the complex way that GHC handles invalid characters in filenames. It uses a FileSystemEncoding which represents those characters using special high unicode surrigate characters. Filenames that are read in (eg by getDirectoryContents) have this encoding applied. Filenames that are passed to C code have this encoding reversed, to get back to the original byte string that matches the file on disk. GHC has been doing this for a year or two (7.6 iirc).

addWatch does not apply the FileSystemEncoding when calling C code. Note that since addWatch first calls getFileStatus on the filename, the caller cannot just arrange for the encoding to be applied before calling addWatch.

I have a patch, which I'll send in a pull request. This is affecting multiple users, so I hope it can get fixed quickly.

[joeyh]

My patch was applied, so this bug should probably be closed..