- add unicode-derived 'Keystroke' class, which is yielded by the
generator method _resolve_multibyte, adding additional properties,
.is_sequence (bool), .code (int), and .name (str).
- detect encoding and initialize incrementaldecoder in __init__, used
by _resolve_multibyte.
- inherit curses key capability names, values, and create multibyte
sequence pattern lookup in __init__
- add _resolve_keycode method, returns key capability name for an
integer.
- add _resolve_multibyte(text) method, given a string of multibyte
input, 'text', returns instances of Keystroke class, which may be
checked for .is_sequence, .code, and .name.

Add wrap method to Terminal, with accompanying AnsiWrapper class, a
derived version of textwrap.TextWrapper, with a modified version of
_wrap_chunks that is sequence-safe.

A module-global function, 'ansiwrap' is exposed. Sequences are detected
using a range of byte matching and regular expressions. This sequence
padding code been tested thoroughly on ANSI Art from the "Dark Domains"
DVD by "ACiD Productions". Example usage:

>> import blessings
>> t = blessings.Terminal()
>>> for num, byte in enumerate('Pony Express, Choo! CHooooooOOoooOOo!'):
...    seq += t.color(num % 7) + t.bold + byte + t.normal
>>> import textwrap
>>> print u'\n'.join(textwrap.wrap(seq, 15))

(as you can see, textwrap fails horribly !!)

>>> print u'\n'.join(t.wrap(seq, 15))
Pony Express,
Choo!
CHooooooOOoooOOo!

winning !!

Continuing the example from the previous commit, it is now possible:

>>> print t.center(seq)
                     Pony Express, Choo! CHooooooOOoooOOo!

Instantate AnsiString from Terminal, so AnsiString contains these
padding methods, requires less instantiation.

also readability from pexpect

* add _conanical and _echo terminal state helpers and related functions.
* remove another magic bit, '27', by explicitly setting to curses.ascii.ESC
* fix for incremental decoder
* add warning for failed decodes in _resolve_multibyte
* placeholder for inkey() with docstrings of expected behavior

use tty.LFLAGS instead of magic attrs[3] bit. Also rename 'attrs' references to 'mode' to match python standard module naming conv.

in the usual style of mine, fully written but not yet run, this intermittent commit has complete docstrings for the interface in mind.

I haven't a win32 platform, but diligent effort was made to support it.
This commit also includes a possible fix for windows systems to determine the terminal height and width without calls to fcntl.

- remove unused defaultdict import
- refactor input stream key capability from __init__ to
  _init_keystrokes()
- remove redundant input key capability definitions
- store _state_echo and _state_canonical; necessary for win32 systems
- refactor term.stream to term.o_stream to accompany term.i_stream
- refactor documented @Property, do_style
- refactor _resolve_multibyte to _resolve_msb, consistent reference
  'MSB' can be either 'escape sequence' for application keys, or utf-8 bytes
  as part of a hamsterface emoticon.

getch() so far works in non-conanical mode, as unimpressive as it is.

various issues with multibyte sequence decoding, currently looks like
the loop timing and buffering at issue.

This is what the curses wrapper is missing; it is very obnoxious when developing curses when your program crashes, you have to type 'reset' after each crash

That's fine, we simply check the state and always restore it on __del__. If python throws an exception and kicks you out while the terminal is in cbreak mode, a warning is issued to let you know it was restored after doing so.

This will appear after your traceback. So it won't be any suprised as to what happened.

also sprinly various print debugs for MSB, there is a timing issue related to the 2nd byte of a '\x1b[A' sequence failing the kbhit() check. I think i've seen this before, has to do with flushing something correctly at a particular time? will look tomorrow.

remove @property.setter, not python 2.5 compatible.

do_styling remains a property, but read-only !

That is, for win32 systems, if the new height & width returns 0, then the old behavior of trying to use termios is used..

so that the tests don't have to be updated
also a fix for o_fd/decriptor

was trying to figure out why this test failed for so long, but I was running it with 'nose 2>&1 | less' -- no wonder, output was not a tty...

For the longest time, the 2nd byte in a multibyte sequence was blocking on select. I discovered using i_stream.read(1) does not interact with the VMIN and VTIME set for cbreak mode on terminals connected to the stream. a call using os.read(i_stream.fileno(), 1) resolved this.

- is_a_tty class attribute changed to property. This brevities the
  documentation and interface, certainly enforcing the concept that
  this property is read-only!
- the input stream is *always* stdin.
- helped comment the whys and whats of __init__, about fd's, curses, etc.
- remove list comprehension, fix bug with _keycodes being empty
- only detect input encoding if stream is connected to a tty, otherwise
  it is not necessary to detect encoding of multibyte input, as we can't
  support interactive keyboard input on non-ttys.
- try to cleanup _height_and_width win32 support.
- however; removed the windows getch() and kbhit() until it can be tried ..
- removed the debug prints since inkey appears to work now
- removed the excessive delve into termios, using only tty.setcbreak()
  and restoring original terminal settings in finally: clause

doubles as a video game

also snuck a kbhit() timing fix, woops

caught \033#8, my favorite 'fill the screen' sequence. not that anyone would have noticed.

passed w/flying colors ..

(take a byte out of i/o crime!)

this is really just a test of timing performance, revived an old alogorithm i made as a kid and wrapped it into a simple game, called "newton's nightmare".

add a new planet? every 50 points.

really should track the tracers and keep only the 10th last.

leave only a limited tail length using a stack of unique screen positions
die animation !

also add SUP/SDOWN/SLEFT/SRIGHT to game for testing

velocity was always in Down-right direction, now can be any.
also change the 'your score' to be term.height -3, and use term.clear_eol

this game is pretty fun !

do not allow negative numbers to pass to select

pexpect gives us a real pty; will travis-ci allows it?

make unicode decoding more internal, but externally exposed and documented
through the encoding parameter. further improving pexpect test cases.

the default mode is a comprimise of the most compatible and most capable and documented well.

go ahead, draw a box ...

also brought a few changes to WILL_MOVE, WONT_MOVE
also assert TypeError for non-ints for things like term.move(1.0)

my nose for python 2.5 doesn't seem to work

plus a test case, and disable the float exception testing,
as it fails on older python where only a DeprecationWarning is emitted