Add ImageTexture unit tests

[brennennen]

Added ImageTexture unit tests.

I wasn't entirely sure on how to go about testing the draw methods and some feedback around that area would be appreciated. Thanks.

Link back to #43440

[Sauermann]

You can use the viewport texture for this and access the images pixels.
Please have a look at https://docs.godotengine.org/en/latest/classes/class_viewport.html#class-viewport-method-get-texture

For reference you find a usage of this in colorpicker:

godot/scene/gui/color_picker.cpp

Lines 1599 to 1602 in d335281

	Ref<Image> img = picker_texture_rect->get_texture()->get_image();
	if (img.is_valid() && !img->is_empty()) {
	Vector2 ofs = mev->get_position();
	picker_color = img->get_pixel(ofs.x, ofs.y);

[Calinou]

You can use the viewport texture for this and access the images pixels.

Will this work in headless mode without a GPU? CI doesn't have access to any GPU.

[Sauermann]

Will this work in headless mode without a GPU? CI doesn't have access to any GPU.

I didn't think about this - I'm not yet familiar with the rendering server.

After looking in detail at the PR, I'm no longer sure, if my previously mentioned method can work, since no nodes are added to the scene tree.

[AThousandShips]

Can confirm it if the style errors are fixed and the CI can run

[kleonc]

Note this wouldn't draw anything, the canvas would also need to be attached to a viewport.

But if tests run in headless mode (as Calinou mentioned in #88044 (comment)) then this test case seems useless to me. 🤔 Again, I'm not that familiar with the tests / not sure what's desired here hence I won't be approving/dismissing it. cc @akien-mga
(the other test cases look fine to me though)

[akien-mga]

The tests typically run as headless, but they should also work if ran with an actual rendering server and display server.

So it's worth making sure that it does something useful in non-headless mode, and if it can't work in headless, it can be made optional with some if (DisplayServer::get_singleton()->get_name() == "headless") checks.

Now I didn't check, but does headless drawing to a texture in memory actually work? If it does, that's also worth testing.

[brennennen]

Thanks for the feedback.

I'm not sure how to execute tests headless vs non-headless. When you add the "--test" flag, the editor binary doesn't actually start up, it just executes the tests and exits like the console binary. Looking at the startup code, it looks like if the "--test" argument is present, the main enters "TEST_MAIN_PARAM_OVERRIDE" and returns/exits from that before opening a windows context or running the actual proper engine main.

Things I've tried:

• DisplayServer::get_singleton()->get_name() returns "mock"
• DisplayServer::get_singleton()->window_can_draw() returns false
• Adding proper viewport initialization code (see below), errors out on "RenderDummy" calls to access a null pointer viewport texture.
• Accessing "RSG::viewport->get_total_objects_drawn()" and similar metrics functions, they all always return 0.

Unless anyone has any other ideas, I'm on board to just delete the draw test case.

TEST_CASE("[SceneTree][ImageTexture] draw") {
	print_line(DisplayServer::get_singleton()->get_name()); // prints "mock"

	RID scenario = RS::get_singleton()->scenario_create();
	RID viewport = RS::get_singleton()->viewport_create();
	RS::get_singleton()->viewport_set_update_mode(viewport, RS::VIEWPORT_UPDATE_DISABLED);
	RS::get_singleton()->viewport_set_scenario(viewport, scenario);
	RS::get_singleton()->viewport_set_size(viewport, 128, 128);
	RS::get_singleton()->viewport_set_transparent_background(viewport, true);
	RS::get_singleton()->viewport_set_active(viewport, true);

	Ref<Image> image = Image::load_from_file(TestUtils::get_data_path("images/icon.png"));
	Ref<ImageTexture> image_texture = ImageTexture::create_from_image(image);
	RID canvas = RenderingServer::get_singleton()->canvas_create();
	RID canvas_item = RenderingServer::get_singleton()->canvas_item_create();
	RenderingServer::get_singleton()->canvas_item_set_parent(canvas_item, canvas);
	image_texture->draw(canvas_item, Point2(0.0, 0.0));
	
	print_line("get_total_objects_drawn: ", RSG::viewport->get_total_objects_drawn()); // prints 0
	print_line("get_total_primitives_drawn: ", RSG::viewport->get_total_primitives_drawn()); // prints 0
	print_line("get_total_draw_calls_used: ", RSG::viewport->get_total_draw_calls_used()); // prints 0
	
	RS::get_singleton()->viewport_attach_canvas(viewport, canvas);
	RID viewport_texture = RS::get_singleton()->viewport_get_texture(viewport);
	Ref<Image> img = RS::get_singleton()->texture_2d_get(viewport_texture); // ERROR: Parameter "t" is null.
// at: RendererDummy::TextureStorage::texture_2d_get (C:\repos\godot\servers\rendering\dummy\storage\texture_storage.h:106)
 	if (img.is_valid() && !img->is_empty()) { 
		// never reached
	} else {
		print_line("viewport texture is null/empty"); // executed every time
	}
}

[kleonc]

Unless anyone has any other ideas, I'm on board to just delete the draw test case.

Yeah, makes sense to just remove it. If desired (and decided what's a proper way to test it) it could be always added separately in another PR.
(Sorry for the late response; I've seen your comment before but I think I've missed this specific sentence. 🤔)

[brennennen]

All good, my comment was a bit of a wall of text. I went ahead and deleted the test case. I'll re-request a review.

[kleonc]

I went ahead and deleted the test case. I'll re-request a review.

I've already approved the PR after the removal. So that was a pointless re-request. 🙃

[akien-mga]

Thanks!