Suggestion: use stderr for I/O, or make it configurable

[semicontinuity]

Hello, here is a suggestion: it would be nice if picotui used stderr for I/O, and not stdin+stdout.
Why? that would free stdio and stdout for other application-specific things: the app can get its input from stdin and write result to stdout, unix-way, and it would allow to use picotui-base app in unix pipes:
echo "input-data" | picotui-app | consumer

I'm doing exactly that in my picotui-based app, and it works great.
Here's the code I had to add to make it work (thanks to python, no need to modify picotui code, can just patch it):

from picotui.screen import Screen

#################################################################################
# Start patching picotui to use STDERR (FD 2)
#################################################################################
FD_IN = 2
FD_OUT = 2


import os


def wr(*args):
    s = args[-1]
    # TODO: When Python is 3.5, update this to use only bytes
    if isinstance(s, str):
        s = bytes(s, "utf-8")
    os.write(FD_OUT, s)


from picotui.basewidget import Widget
from picotui.defs import KEYMAP as _KEYMAP


def get_input(self):
    if self.kbuf:
        key = self.kbuf[0:1]
        self.kbuf = self.kbuf[1:]
    else:
        key = os.read(FD_IN, 32)
        if key[0] != 0x1b:
            key = key.decode()
            self.kbuf = key[1:].encode()
            key = key[0:1].encode()
    key = _KEYMAP.get(key, key)

    if isinstance(key, bytes) and key.startswith(b"\x1b[M") and len(key) == 6:
        row = key[5] - 33
        col = key[4] - 33
        return [col, row]

    return key


def screen_size(*args):
    import select
    wr(b"\x1b[18t")
    res = select.select([FD_IN], [], [], 0.2)[0]
    if not res:
        return (80, 24)
    resp = os.read(FD_IN, 32)
    assert resp.startswith(b"\x1b[8;") and resp[-1:] == b"t"
    vals = resp[:-1].split(b";")
    return (int(vals[2]), int(vals[1]))


def init_tty(*args):
    import tty, termios
    global ttyattr
    ttyattr = termios.tcgetattr(FD_IN)
    tty.setraw(FD_IN)


def deinit_tty(*args):
    import termios
    termios.tcsetattr(FD_IN, termios.TCSANOW, ttyattr)


def patch_picotui():
    Screen.wr = wr
    Screen.init_tty = init_tty
    Screen.deinit_tty = deinit_tty
    Screen.screen_size = screen_size
    Widget.get_input = get_input

#################################################################################
# End patching picotui
#################################################################################

.. that's it, just introduced constants instead of hard-coded FD 0 and 1

[pfalcon]

Thanks for the feedback and suggestion. I'll keep it in mind, but arguments for doing it like that would be that it's customary to do it at all. Do other TUI applications do it like that? Does curses lib do it like that by default?

As a random quick test, I tried Midnight Commander with:

mc >/dev/null

That doesn't produce any output, so it clearly does TUI operations on stdout.

[semicontinuity]

I agree, by very nature TUI must interact with console. Most likely, there are no conventions on FD usage for TUI.
In shell, FD 0 and 1 can participate in pipe, and 2 cannot be piped, and supposed to be attached to "debug console".
Anyway, if you would just introduce constants FD_IN=0 and FD_OUT=1, so that it behaves the same way and give the possibility to modify them and attach to different FD - that would be really nice

[pfalcon]

Anyway, if you would just introduce constants FD_IN=0 and FD_OUT=1

I'll keep that in mind. I have a queue of things I'd like to do, including to that part of the code, so it may need to wait some time, but I'm not against it in principle.