fix: rasterize node document leakage

[Ayush2k02]

Simple enough fix which skips the flawed deduplication check and generates a different id ( UUID ) for each new image

Screen.Recording.2025-12-28.at.7.46.56.PM.mov

Closes #3533

[Ayush2k02]

Don't fully understand this syntax but from what I can see this generates an ID for an image and was used in deduplication for similar images . Since we are no more deduplicating similar images and each new image is getting its own ID , figured it would be best to use a simple generate_uuid instead

[Ayush2k02]

Without doing so , its still causing the same image leakage for me

[Ayush2k02]

This ones ready to be reviewed from my side

[TrueDoctor]

I need a bit more time to look at this, the workaround implemented here does not feel quite right, I'm also wondering if this will effect the performane as it is currently implemented

[0HyperCube]

@TrueDoctor yes it will be bad for performance.

The hashing code for images is included below:

Graphite/node-graph/libraries/raster-types/src/image.rs

Lines 104 to 116 in fa45efa

	// TODO: Evaluate if this will be a problem for our use case.
	/// Warning: This is an approximation of a hash, and is not guaranteed to not collide.
	impl<P: Hash + Pixel> Hash for Image<P> {
	fn hash<H: Hasher>(&self, state: &mut H) {
	const HASH_SAMPLES: u64 = 1000;
	let data_length = self.data.len() as u64;
	self.width.hash(state);
	self.height.hash(state);
	for i in 0..HASH_SAMPLES.min(data_length) {
	self.data[(i * data_length / HASH_SAMPLES) as usize].hash(state);
	}
	}
	}

However this will lead to very many conflicts when you have images such as the cases below:

Since there are 1000x1000 pixels and the vast majority are white, these two both hash to the same thing.

It is necessary for correctness that these two different images aren't treated as the same thing by the SVG rendering code. Otherwise it looks like the rasterize node is broken. Actually the rasterization works fine.

An example graphite file where this collision occurs is include below:
rasterize_node_conflicts.graphite

[TrueDoctor]

Yeah the hash is broken, I guess we could do an eq check after the hash calculation, that should hopefully be fast enough. But I guess we would need to benchmark if disabling the deduplication alltogether would be faster

[Ayush2k02]

An example graphite file where this collision occurs is include below:
rasterize_node_conflicts.graphite , like do we import this JSON somewhere to load this ourselves ?

What is this ?? and how is this used ?

[0HyperCube]

@Ayush2k02 Yes, you can import this graphite file into the Graphite editor. file -> open and select the file.

Note that the extension is .graphite.json since GitHub doesn't directly support uploading files using .graphite extension. You can either change the extension after you download the file or just directly select file (you might have to alter the filter in your file picker).