EF-BF-BE is the unicode "byte order mark" (BOM) encoded in UTF-8. It was born as a way to distinguish big- and little-endian UTF-16. Since it's not really a character, Iconv tries to strip it when converting to a UnicodeString, but it is failing to do so in this case.

Now, under Mac OS X I get the expected result, under Linux I get yours. The reason is that my Mac is big-endian, so Iconv produces big-endian UTF-16, while Linux produces little-endian UTF-16. Since the default encoding of UTF-16 is big-endian, the Mac happens to get the right thing, while Linux messes up the encoding. So later on the "pipe peekFor: $<16rFEFF>" statement to strip the BOM does not work.

The attached patch fixes this by making EncodedString look for a BOM when retrieving the encoding, rather than when setting it.

fixed in patch-612, which is the same patch I posted plus this testcase

  str := EncodedString fromString: (String new: 2) encoding: 'UTF-16'.
  str valueAt: 1 put: 254; valueAt: 2 put: 255.
  self assert: str numberOfCharacters = 0.
  str valueAt: 1 put: 255; valueAt: 2 put: 254.
  self assert: str numberOfCharacters = 0

Thanks!