Hi.

ustring::to_utf8() creates a

  MiniIconv ic("UTF-8", "UTF-16");

assuming that iconv(3) uses the native byte order for "UTF-16". On OS X w/ Intel CPUs (I installed poppler through MacPorts, but this issue is unrelated, see below) this fails, as a quick

  $ echo -n 7 | iconv -t utf-16 |  hexdump -C
  00000000  fe ff 00 37                                       |...7|

reveals: it's UTF-16BE.

This breaks page-labels for me, which instead of "78" (UTF-8) return the (hex) values

  e3 9c 80 e3 a0 80

which is 0x3700 0x3800.


A fix might be to not "decode" GooString's UTF-16BE to native byte order in

  detail::unicode_GooString_to_ustring(GooString *str)

or use a source encoding based on the BYTE_ORDER macro instead of just "UTF-16BE" or to check the BOM-character output by iconv(3) (which e.g.

  ustring::from_utf8(const char *str, int len)

currently skips).