Try unsetting LANG and passing git-annex unicode filenames.
joey@gnu:~/tmp/aa>git annex add ./Üa
add add add add git-annex: <stdout>: commitAndReleaseBuffer: invalid
argument (Invalid or incomplete multibyte or wide character)
Interestingly, I can get the same crash in the de_DE.UTF-8 locale with certian input filenames, while in en_US.UTF-8, it's ok. The workaround below avoided the problem in de_DE.UTF-8. --Joey
Put in the utf-8 forcing workaround for now. done --Joey
underlying haskell problem and workaround
The same problem can be seen with a simple haskell program:
import System.Environment
import Codec.Binary.UTF8.String
main = do
args <- getArgs
putStrLn $ decodeString $ args !! 0
joey@gnu:~/src/git-annex>LANG= runghc ~/foo.hs Ü
foo.hs: <stdout>: hPutChar: invalid argument (Invalid or incomplete multibyte or wide character)
(The call to decodeString
is necessary to make the input
unicode string be displayed properly in a utf8 locale, but
does not contribute to this problem.)
I guess that haskell is setting the IO encoding to latin1, which is documented to error out on characters > 255.
So this program doesn't have the problem -- but may output garbage on non-utf-8 capable terminals:
import System.IO
main = do
hSetEncoding stdout utf8
args <- getArgs
putStrLn $ decodeString $ args !! 0