Wouldn't be better to restore sys.ps1 in test_idle if it is changed here?

A program that needs sys.ps1/2 in particular states should set the state it needs. Test_code_module needs them both unset and set to a non-default string. Running test_code_module after test_idle exposed the same bug that was exposed previously by running autotest in a Python shell. The current patch prevents the autotest test_code_module failure that I saw previously. See the issue for more.

The full test suite passes on my Win10 machine.

I think this issue should be fixed on both sides. Tests that change module globals should restore them after testing. And test_code_module should not depend on the current value of sys.ps1, but test all possible situations. The test.support.swap_attr() context manager can help here.

Tests routine mutate sys.modules and add attributes to other modules without 'restoring' them. For instance, test_tk adds at least .font, .test, and .ttk to tkinter. sys.ps1 is at least as expected as these.

However, on reading the IDLE code, I decided it would be improved by adding a PyShell() attribute set from sys.ps1 when it exists, and never modifying sys. (The code module could do the same.) I made this a separate issue (bpo-31858) and PR (#4143). I plan to push both tomorrow.

I suggest to not change sys.ps1 permanently, but restore the old values after testing.

Actually test_code_module already contains necessary code for restoring the sys module. It only needs to initialize ps1 and ps2 properly when they already were set.

After the update merge, which included the change to IDLE not setting sys.ps1, test_code_module continues to pass the real test, which is running with import test.autotest.

Thanks @terryjreedy for the PR 🌮🎉.. I'm working now to backport this PR to: 3.6.
🐍🍒⛏🤖

GH-4156 is a backport of this pull request to the 3.6 branch.