ofNoise() -- Memory bad access issue

[nicoleyimessier]

When running my app with address sanitizer, it quickly catches a bad access that happens within the noise library.

[bakercp]

@nicoleyimessier Did this cause any additional issues or did you find a solution?

[ofTheo]

Huh. This could be an issue.

Looking through the code for slang_library_noise3 I think the variable ii could be negative as is shown in the screen grab above.

Here is the relevant part in the code:
https://github.com/openframeworks/openFrameworks/blob/master/libs/openFrameworks/utils/ofNoise.h#L327

Going back to ii being set from i and i from xs I don't see anything that would stop any of those from being negative.

If ii is negative then there will be an out of bounds with perm

Would love a sanity check on this though
:)

[alexp-nl]

Recently, I saw some strange and erratic behaviour in a piece of my openFrameworks code using ofNoise. After a long search, I could trace it back to the problem mentioned above. Try debugging into the call ofNoise(-13.7,14.6) and see what happens. Indices in perm become negative. Replacing all the %256 by &0xff solves this problem. Would be nice to fix this in future releases of of.

[ofTheo]

@alexp-nl - it definitely seems like an issue to me.
for some reason the values I get out of perm are fine on macOS even if ii and jj are negative.

changing them to use &0xFF does fix the negative ii and jj it but the noise curve changes.

Do you get erratic values with this code?

void ofApp::draw(){

	ofSetColor(0);
	
	float t = -1234;

	ofNoFill();
	ofBeginShape();
	for(int i = -1024; i < 1024; i++){
		 int ii = i &0xff;
		 int badi = i % 256;
		
		 float v = ofNoise(-16.7 + ((float)i)*0.005, -14.6 );
		 
		 float x = ofMap(i, -1024, 1024, 0, ofGetWindowWidth());
		 float y = ofGetHeight()/2 + v * 100;
		 
		ofVertex(x, y);
	}
	ofEndShape(false);
}

%256

&0xFF

[ofTheo]

Note for positive values both look the same.

The issue is the fix changes ii and jj like so:

i = 0, i %256 = 0, i &0xFF = 0
i = -1, i %256 = -1, i &0xFF = -255
i = -2, i %256 = -2, i &0xFF = -254

hence the different graphs when negative.

[alexp-nl]

The curve should change, because the %256 version is wrong. It depends on data outside of the perm array. In my case, the erratic behaviour was caused by the changing of this data outside (before) the perm array. So two calls with the same parameters to ofNoise resulted in the different return values.
Try replacing the definition of perm
unsigned char perm[512]
by
std::array<unsigned char,512>perm
and (in MSVS) compiling in debug mode. This catches out of range indices in perm. In the %256 case, this happens quite often, the &0xff version works fine.
Great if the problem it is resolved in of 0.12.1!

[ofTheo]

Thanks @alexp-nl - this should be fixed in master / nightly now.
Appreciate you bringing this up and can't believe it took this long to fix 😬

[alexp-nl]

Thanks @alexp-nl - this should be fixed in master / nightly now. Appreciate you bringing this up and can't believe it took this long to fix 😬

Glad I could help, and thanks to everyone who has made openFrameworks what it is today!