When python is in interactive mode, curses does not react to resize
events properly.

test.py:

import curses

def run():
 stdscr=curses.initscr()
 key=0
 while(key!=ord('q')):
  key=stdscr.getch()
  stdscr.addstr(0,0,str(stdscr.getmaxyx())+' '+str(key))
  stdscr.refresh()
 curses.endwin()

run()

When this is run directly, everything is ok. When it is called via
execfile() from the interactive prompt, it shows the right screen size
after the first keypress, but behaves oddly after the resize.

IMHO, the following happens:
For some reason, env. variables LINES and COLUMNS are set but they are
not reflected in the os.environ structure nor they respond to screen
size changes. If these variables are set then the ncurses library (see
man pages) uses their values instead of getting the term size via
ioctl.
The ncurses library receives a SIGWINCH and sees that LINES
and COLUMNS are set. However, their values are same as the
screen dimensions before the resize, so it is perplexed why there is
a SIGWINCH if the screen did not change and it just ungetchs an ERR
and ncurses internal structures are not changed appropriately.

From the resizeterm man page:
"If the environment variables LINES or COLUMNS are set, this overrides
the library's use of the window size obtained from the operating
system. Thus, even if a SIGWINCH is received, no screen size change may
be recorded. In that case, no KEY_RESIZE is queued for the next call to
getch; an ERR will be returned instead."

Executing

import os
os.environ['LINES']="blah"
del os.environ['LINES']
os.environ['COLUMNS']="blah"
del os.environ['COLUMNS']

solves the problem for me.
Perhaps the problem has sth to do with python using readline
in interactive mode???
PB

Could somebody witht a nix box try this out, I'd do it myself if I had such a beast.

Well, I can run it, but I have no idea what it is telling me or how it demonstrates a bug. Pavel, is this still a problem in 2.6/2.7?

Just to confirm: curses SIGWINCH handling is still (2.7/3.2) broken after importing readline.

Readline seems to set the LINES/COLUMNS environment vars, and this confuses curses, even if curses is imported after readline.

Clearing the LINES/COLUMNS vars after import readline fixes the issue:
os.environ['LINES'] = os.environ['COLUMNS'] = ''
or
os.unsetenv('LINES'); os.unsetenv('COLUMNS')
(or other variations).

I spent a couple hours tearing my hair out over this, before I found this report.

It may be possible for Python to work around this readline behavior by saving LINES/COLUMNS and restoring them after initializing readline.

Or maybe this should just be documented somewhere.

bpo-3948 is almost certainly a duplicate.

For the record, this is my interpretation of Pavel’s demo:

If you press a key (other than lowercase “q”) or resize the terminal, it displays the terminal dimensions in the top-left corner. Without Readline, the dimensions are updated when the terminal resizes.

When Readline is used (interactive mode, or simply “import readline”), the reported dimensions are static, which I understand is the bug.

Documentating this quirk may be a good start.