When I run test suites I find something interesting. Even if I don't enable gui resource there are still windows flashing. The two tests are test_idle and test_tk. I think the root cause is that they both use test.support.import_module with no reason and then root.update(in _is_gui_available) is called and windows flash. Comment out root.update windows are gone. Is this a must or can we find some way to suppress this?

For your reference, the root.update is introduced in bpo-22770.

The order of the checks in support.requires{,_resource} should be reversed: check if the resource is enabled, then check for gui availability if the resource in question is 'gui'.

I tried that, no help to this thread, but I think that is better.

root.update itself draws something on the screen.

Ahh, sorry. I misunderstand your message. Just ignore my last one. Really sorry for making noise.

Your suggestion makes sense I think.

To my surprise, "python -m test test_idle" gives a single flash in Windows. bpo-22770 modified /Lib/test/support/init.py and added the creation and packing of a Label. Those two lines are gone in repository 3.6 but the flash is still there, I added root.withdraw() at line 469 and the flash is gone.

    try:
        from tkinter import Tk
        root = Tk()
        root.withdraw()
        root.update()
        root.destroy()

I added a Label packing back and the flash is still gone here.

"use test.support.import_module with no reason " ???
I used import_module in test_idle because I was told to and had to to prevent inappropriate test failures. Please explain what you mean.

Some time ago I tried adding 'root.withdraw' to idle_test files that used tkinter to stop the widget flashing. This never worked completely. Now I know why -- it was the test.support import. Knowing this now, I opened bpo-27922 to add 'root.withdraw' where else needed. With support.__init__ patched, idle tests in 3.6 are currently flash-free, with -ugui, and 2.7 and 3.5 should be also. So unless there is objection by tomorrow, I will patch all three versions. I want the import flash suppressed even when -ugui is passed.

they both use test.support.import_module with no reason

Sorry, this is a typo. I meant to say "they both use test.support.requires". There should be *no* "no reason" and it's "requires" not "import_module". Sorry to confuse you. Staying late in night fools my brain.

I think it's nice to suppress the flash totally in support. Even gui is enabled a single support.requires may seem somewhat confusing to flash a window. Can we also apply Zachary's suggestion? If gui is not enable, there is no need to check _is_gui_available.

I completely agree that we should not check that gui works unless and until it is requested. Can you write a patch to do the switch?

My point was that even when I request gui, I still don't want to see the flash. I do TDD for test files as well as app files and I may run tests 10s of times in a day. I will be happy to have that be a bit less stressful.

It's my pleasure.

My point was that even when I request gui, I still don't want to see the flash.

I also mean that in my last message, in this sentence " Even gui is enabled a single support.requires may seem somewhat confusing to flash a window". But it seems my poor English doesn't make it clear. :(

bpo-27918.patch does the switch and I add the withdrawn you suggest.

Xiang’s second patch looks okay to me. The flashes occasionally annoy me with X windows if I am doing something else and have the Python tests running in the background.

I understand deferring the _is_gui_available() call will fix the main problem, and the withdraw() call is just an added nicety to avoid unnecessary windows when you actually enable -ugui.

New changeset de9e410e78d8 by Terry Jan Reedy in branch '2.7':
Issue bpo-27918# test.resource.is_gui_available no longer flashes tk window.
https://hg.python.org/cpython/rev/de9e410e78d8

New changeset 756c27efe193 by Terry Jan Reedy in branch '3.5':
Issue bpo-27918# test.resource.is_gui_available no longer flashes tk window.
https://hg.python.org/cpython/rev/756c27efe193

This issue only addresses and fixes the 1 flash per process from test.support.is_gui_available. I separately added root.withdraw to IDLE tests in bpo-27922, to stop their flashes. Doing the same for tkinter tests would be a third issue. I believe it would be easy since tkinter has a support module with mechanisms to set up and tear down root windows. (I am thinking of something similar for IDLE.) I leave this to Serhiy.