MeshRenderer render into a RenderTexture

[TedGriggs]

Is this possible? I could not find an example of this.

Calling visit() on a MeshRenderer requires a renderer, a transform, and some flags. Where do I get these from?

The best I got working was this by putting the MeshRenderer as a child of a Node:

auto size = Director::getInstance()->getVisibleSize();

auto sceneRT = RenderTexture::create(
    size.width,
    size.height,
    rhi::PixelFormat::RGBA8
);
sceneRT->setPosition(Vec2(size.width / 2, size.height / 2));
sceneRT->setAnchorPoint(Vec2::ANCHOR_MIDDLE);
sceneRT->retain();

auto node=Node::create();
node->setPosition(size/2);
node->retain();

auto model = MeshRenderer::create("boss1.obj");
model->setTexture("boss.png");
model->setPosition3D(Vec3(0,0,0));
model->setScale(4);
node->addChild(model);

sceneRT->begin();
sceneRT->clear(0, 0, 0, 0);  // NOTE: removing this stops the model from being drawn at all
node->visit();
sceneRT->end();

auto sprite= Sprite::createWithTexture(sceneRT->getSprite()->getTexture());
sprite->setPosition(size/2);
sprite->setFlippedY(true); // important!
scene->addChild(sprite);

However, the model flickers. If I don't call clear() or put it outside the begin() and end() then nothing is rendered.

I can render onto the screen scene fine.

I am using Linux.

[rh101]

Call beginWithClear() rather than begin() and clear(), since clear() calls end().

Try this to see if it works:

sceneRT->beginWithClear(0, 0, 0, 0);
node->visit();
sceneRT->end();

Otherwise, if you want to call clear() and begin() separately for some reason, then this is the correct order:

sceneRT->clear(0, 0, 0, 0);  // Must be before begin()
sceneRT->begin();
node->visit();
sceneRT->end();

Also, some nodes may require you to call onEnter() prior to visit(), in case there is logic in there that needs to be run.

[rh101]

I found the EVENT_AFTER_SET_NEXT_SCENE by looking into the code to see how it worked and why it was crashing. There is isn't an example of this in the cpp-tests, by the way.

I have looked at the examples, and using 2D Sprites, RenderTexture works, but using MeshRenderers it does not.

OK, since that is the case, then one will need to be added to cpp-tests.

I an not sure why you added those last two paragraphs where 1) I clearly did look into the code, and 2) you didn't answer the question on why this isn't working even after I tried what you said. I don't know if this was your intent, but it was a bit patronizing.

I apologize if it came across that way, but I suggested you look into cpp-tests since developers new to Axmol often don't realize the project is there. If the test project is lacking test cases, then that is something that should be addressed.

Now I'm a little confused why you're stating that it did not work, when in the previous post you stated this:

I had to put in a callback for when the first scene was done via EVENT_AFTER_SET_NEXT_SCENE. However, this seems like a hack, since nothing in the engine seemed to need this extra step.

That implied that you got it working, or did I misunderstand, and you meant that this was the only way to avoid the crash, but the RenderTexture output is still incorrect?

[rh101]

OK, so I wasn't imagining things; there should be a test in Axmol, but it's missing. This issue was fixed for Cocos2d-x v3 back in 2016 in this PR, and the same code made it into Cocos2d-x v4, but I'm not entirely sure why it's not in Axmol.

Axmol also seems to be missing methods in the API, such as Mesh::setForce2DQueue() and MeshRenderer::setForce2DQueue() (MeshRenderer was previously named Sprite3D). Axmol was forked from Cocos2d-x v4 back in 2019, so it all should be there, but again, I'm not sure why it is not the same.

[rh101]

Add this line and it should work:

model->getMaterial()->setForce2DQueue(true);

Also, you may need to set the depth value when calling beginWithClear, but it seems to work without it.

The following works:

auto size  = Director::getInstance()->getVisibleSize();

auto sceneRT = RenderTexture::create(size.width, size.height, PixelFormat::RGBA8);
sceneRT->setPosition(Vec2(size.width / 2, size.height / 2));
sceneRT->setAnchorPoint(Vec2::ANCHOR_MIDDLE);

auto node = Node::create();
node->setPosition(size / 2);

auto model = MeshRenderer::create("boss1.obj");
model->setTexture("boss.png");
model->setRotation3D(Vec3(0, 0, 0));
model->setScale(4);
model->getMaterial()->setForce2DQueue(true); // This fixes the issue
node->addChild(model);

sceneRT->beginWithClear(1, 0, 0, 1, 1); // set depthValue too, which also sets ClearFlag::DEPTH. Not sure if it's required though
node->visit();
sceneRT->end();

auto sprite = Sprite::createWithTexture(sceneRT->getSprite()->getTexture());
sprite->setPosition(size / 2);
sprite->setFlippedY(true);
this->addChild(sprite);

[rh101]

It seems that the specific test was commented out in Cocos2d-x v4 when Apple Metal support was added to the engine, and never re-implemented, prior to the Axmol fork being created. Eventually the test was deleted in Axmol in this PR when Sprite3D was renamed to MeshRenderer, perhaps because of the comment in the code regarding that specific test.

That explains why it's missing from Axmol. It may be best to add a similar test back into cpp-tests at some point.

[TedGriggs]

Thank you very much for looking into this! I resolved the null scene issue with the event handler, but meant that still didn't draw anything with the previous proposed fix. It seems you figured this out and I appreciate you determining what was going on. I will add the setForce2DQueue() call, and it sounds like that will get it working!

With Cocos2d-x gone, Axmol is the only multi-platform game engine that caters to C++, and I really hope it gets the traction it deserves from that developer community. The cpp-tests are great at checking out how all the features work, and they are indispensable.

[TedGriggs]

One more thing that's not related to the above issue, but is a part of what I'm working on... On Linux when I run cpp-tests, and I choose the fake shadow test, I don't see a shadow. I do see a shadow on your online version: https://axmol.dev/wasm/cpp-tests/cpp-tests

[TedGriggs]

Not sure this is useful, however, as I am running both tests on the same system.

[halx99]

Not sure this is useful, however, as I am running both tests on the same system.

please update to latest

[halx99]

8c451e4 this commit fix the issue

[rh101]

@TedGriggs It would be best if the Axmol v3 you are using is kept up-to-date, given that it is still in active development, with no actual production release as of yet. If you encounter any issues, update to the latest version and re-test, since the issue may have already been fixed.

[TedGriggs]

Will do!