gh-154749: Reject a terminal-less screen in curses.set_term()

[fedonman]

curses.set_term() only checked that its argument is a screen that has not been
deleted, so it also accepted a screen from curses.new_prescr(). That screen has no
terminal attached, and once it was made current the next window refresh dereferenced
NULL inside curses and killed the interpreter:

import curses, os
fd = os.openpty()[1]
scr = curses.newterm('xterm', fd, fd)
curses.set_term(curses.new_prescr())
scr.stdscr.refresh()          # Segmentation fault, rc=139

set_term() now raises curses.error when the screen has no standard window, which is
exactly what a new_prescr() screen is. The documentation already described the argument
as a screen returned by newterm(), so this makes the code match, and the docs now
mention the exception.

Both functions are new in 3.16 (gh-90092, GH-151748), so no released version is affected
and there is no NEWS entry.

The same reproducer after the change:

set_term(new_prescr()) -> error: the screen has no terminal
refresh after the rejected switch: ok
set_term(real screen) still works, prev is a screen: True
refresh on the new current screen: ok

Tests:

$ ./python -m test -u all -v test_curses -m 'test_set_term_prescr_screen'
test_set_term_prescr_screen (test.test_curses.ScreenTests.test_set_term_prescr_screen) ... ok
OK

$ ./python -m test -u all test_curses
Total tests: run=164 skipped=3
Result: SUCCESS

• Issue: Segfault after curses.set_term() is given a new_prescr() screen #154749

[read-the-docs-community]

Documentation build overview

📚 cpython-previews | 🛠️ Build #33783780 | 📁 Comparing df7e79a against main (5afbb60)

  🔍 Preview build  

3 files changed

± library/curses.html
± whatsnew/3.15.html
± whatsnew/changelog.html

[serhiy-storchaka]

LGTM. 👍