[Python-Dev] Unicode support of the curses module in Python 3.3

[Victor Stinner]

Hi,

I changed many functions of the curses module in Python 3.3 to improve
its Unicode support:
 - new functions: curses.unget_wch() and window.get_wch()
 - new attribute: window.encoding
 - the default encoding is now the locale encoding instead of UTF-8
 - use the C functions *_wch() and *wstr() when available instead of
*ch() and *str() functions. For example, the Python function addstr()
calls waddwstr() and addch(str) calls wadd_wch() (addch(int) and
addch(bytes) are still calling waddch())

Most new features related to Unicode now depends if the Python curses
module is linked to the C libncursesw library or not... and the Python
module is not linked to libncursesw if the libreadline library is
linked to libncurses module. How the readline library is linked to
libncurses/libncursesw is not a new problem but it may become more
annoying than before. I hope that most Linux distro are/will link
readline to libncursesw.

For example, if the Python curses module is not linked to libncursesw,
get_wch() is not available and addch("é") raises an OverflowError if
the locale encoding is UTF-8 (because "é".encode("utf-8") is longer
than 1 byte).

I introduced two bugs: get_wch() didn't support keycodes (like
curses.KEY_UP) and addch() didn't work anymore with special characters
like curses.ACS_HLINE. These issues are referenced as #15785 and
#14223 in the bug tracker, and I pushed fixes: c58789634d22 and
27b5bd5f0e4c. I hope that Georg will accept them in Python 3.3 final!

I didn't find these bugs myself because I only used dummy scripts to
test my changes. Does anyone know "real world" applications using the
curses module and supporting Python 3? Can you please test them with
non-ASCII characters and the last development version of Python 3.3?

So please try to test the curses module before Python 3.3 final with
your favorite application!

Victor