Proposed patch improves curses tests. Main improvements are:

* Tests now run even if sys.stdout is not terminal.
* The screen is not spoiled during tests.
* Reported the name and arguments of failed function.
* Testing optional functionality is extracted in separated tests. They are reported as skipped if optional functions are not available.

LGTM. I left some comments on Rietveld.

I guess this might help Issue 12669 (running the test on buildbots)? I had assumed a real terminal was required, but it looks like you are substituting a temporary file.

New changeset 55852de7d032 by Serhiy Storchaka in branch '3.5':
Issue #27067: Improved curses tests.
https://hg.python.org/cpython/rev/55852de7d032

New changeset 2917a3ce988e by Serhiy Storchaka in branch 'default':
Issue #27067: Improved curses tests.
https://hg.python.org/cpython/rev/2917a3ce988e

Thank you for your review Berker.

A couple buildbots don’t look happy with this change, e.g.

http://buildbot.python.org/all/builders/x86%20Ubuntu%20Shared%203.5/builds/903/steps/test/logs/stdio

======================================================================
ERROR: test_colors_funcs (test.test_curses.TestCurses)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/srv/buildbot/buildarea/3.5.bolen-ubuntu/build/Lib/test/test_curses.py", line 66, in setUp
    curses.savetty()
_curses.error: savetty() returned ERR

(more failures follow)

My bad. I tested only with stdout redirected, the test fail if both stdout and stderr are redirected.

Proposed patch fixes this.

test_curses_no_atty.patch doesn't work for me using -j1 regrtest option :-/

haypo@smithers$ ./python -m test -u curses -j1 test_curses
Run tests in parallel using 1 child processes
Total duration: 0:00:01
Exception in thread Thread-1:
Traceback (most recent call last):
  File "/home/haypo/prog/python/default/Lib/threading.py", line 916, in _bootstrap_inner
    self.run()
  File "/home/haypo/prog/python/default/Lib/test/libregrtest/runtest_mp.py", line 147, in run
    stop = self._runtest()
  File "/home/haypo/prog/python/default/Lib/test/libregrtest/runtest_mp.py", line 138, in _runtest
    result = json.loads(result)
  File "/home/haypo/prog/python/default/Lib/json/__init__.py", line 319, in loads
    return _default_decoder.decode(s)
  File "/home/haypo/prog/python/default/Lib/json/decoder.py", line 339, in decode
    obj, end = self.raw_decode(s, idx=_w(s, 0).end())
  File "/home/haypo/prog/python/default/Lib/json/decoder.py", line 357, in raw_decode
    raise JSONDecodeError("Expecting value", s, err.value) from None
json.decoder.JSONDecodeError: Expecting value: line 1 column 1 (char 0)

By the way, all buildbots now use -j1 option and so stdout and stderr are no more TTY on all buildbots: see issue #25285.

So test_curses are not fully tested by buildbots anymore.

Maybe we should have at least one buildbot testing test_curses with a TTY?

New changeset 433839643bbb by Serhiy Storchaka in branch '3.5':
Temporary skip curses tests on non-tty (issue #27067).
https://hg.python.org/cpython/rev/433839643bbb

New changeset debe693c62b4 by Serhiy Storchaka in branch 'default':
Temporary skip curses tests on non-tty (issue #27067).
https://hg.python.org/cpython/rev/debe693c62b4

> Maybe we should have at least one buildbot testing test_curses with a TTY?

My non-debug and installed Gentoo bot does run the tests with a PTY, so you
can use its custom builders for experimentation (assuming it can complete a
clone).

Updated patch makes test_curses working with -jN. But it spoils a screen when run with -jN on a terminal (shouldn't be a problem for buildbots).

Zachary Ware added the comment:
>> Maybe we should have at least one buildbot testing test_curses with a TTY?
> My non-debug and installed Gentoo bot does run the tests with a PTY, ...

This one? http://buildbot.python.org/all/builders/x86%20Gentoo%20Non-Debug%20with%20X%203.x/builds/1003/steps/test/logs/stdio

"Run tests in parallel using 1 child processes": test_curses stdout is
also redirected.

I’m not 100% following this, but it is sounding like a terminal may be required for proper testing. Perhaps I could try and help by creating a pseudo-terminal in the Python test code, as I hinted at <https://bugs.python.org/issue12669#msg265663>.

Continued in issue42789 and issue43016.