Fix translucent image colors by correcting alpha premultiplication in image decoding and GPU texture transfers

[Keavon]

Superseded by #4518.

[cubic-dev-ai]

2 issues found across 6 files

Confidence score: 4/5

• node-graph/libraries/no-std-types/src/color/color_types.rs duplicates SRGBA8-to-premultiplied-Color conversion logic, so future color or alpha fixes could diverge across image decoding and GPU readback; extract and reuse a shared helper.
• node-graph/libraries/raster-types/src/image.rs changes stored alpha semantics and could introduce a subtle color regression despite the fix; add a round-trip test using from_image_data with known alpha values.

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="node-graph/libraries/no-std-types/src/color/color_types.rs">

<violation number="1" location="node-graph/libraries/no-std-types/src/color/color_types.rs:363">
P3: This change duplicates the SRGBA8-to-premultiplied-`Color` conversion already used by image-data decoding and GPU readback. Extract the conversion into a shared helper and call it from all paths so future color or alpha fixes cannot leave these implementations inconsistent.</violation>
</file>

<file name="node-graph/libraries/raster-types/src/image.rs">

<violation number="1" location="node-graph/libraries/raster-types/src/image.rs:152">
P3: This premultiplication fix changes stored alpha semantics (no longer alpha-squared) and is the kind of subtle color-space regression that can silently reappear. Add a round-trip test that runs from_image_data on a known semi-transparent RGBA pixel and asserts the stored Color has alpha == v[3]/255 with premultiplied linear RGB (e.g. white at alpha 128 → rgb ≈ 0.502, alpha ≈ 0.502), and that to_png()/to_flat_u8 recovers the original u8 values.</violation>
</file>

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[cubic-dev-ai]

P3: This change duplicates the SRGBA8-to-premultiplied-Color conversion already used by image-data decoding and GPU readback. Extract the conversion into a shared helper and call it from all paths so future color or alpha fixes cannot leave these implementations inconsistent.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At node-graph/libraries/no-std-types/src/color/color_types.rs, line 363:

<comment>This change duplicates the SRGBA8-to-premultiplied-`Color` conversion already used by image-data decoding and GPU readback. Extract the conversion into a shared helper and call it from all paths so future color or alpha fixes cannot leave these implementations inconsistent.</comment>

<file context>
@@ -360,8 +360,8 @@ impl Pixel for Color {
 		// `Image<Color>` pixel convention is linear-light with associated (premultiplied) alpha.
-		let srgba = SRGBA8::new(bytes[0], bytes[1], bytes[2], bytes[3]);
-		Color::from(srgba).apply_opacity(bytes[3] as f32 / 255.)
+		let color = Color::from(SRGBA8::new(bytes[0], bytes[1], bytes[2], bytes[3]));
+		color.map_rgb(|channel| channel * color.a())
 	}
</file context>

[cubic-dev-ai]

P3: This premultiplication fix changes stored alpha semantics (no longer alpha-squared) and is the kind of subtle color-space regression that can silently reappear. Add a round-trip test that runs from_image_data on a known semi-transparent RGBA pixel and asserts the stored Color has alpha == v[3]/255 with premultiplied linear RGB (e.g. white at alpha 128 → rgb ≈ 0.502, alpha ≈ 0.502), and that to_png()/to_flat_u8 recovers the original u8 values.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At node-graph/libraries/raster-types/src/image.rs, line 152:

<comment>This premultiplication fix changes stored alpha semantics (no longer alpha-squared) and is the kind of subtle color-space regression that can silently reappear. Add a round-trip test that runs from_image_data on a known semi-transparent RGBA pixel and asserts the stored Color has alpha == v[3]/255 with premultiplied linear RGB (e.g. white at alpha 128 → rgb ≈ 0.502, alpha ≈ 0.502), and that to_png()/to_flat_u8 recovers the original u8 values.</comment>

<file context>
@@ -148,8 +148,8 @@ impl Image<Color> {
-				let srgba = SRGBA8::new(v[0], v[1], v[2], v[3]);
-				Color::from(srgba).apply_opacity(v[3] as f32 / 255.)
+				let color = Color::from(SRGBA8::new(v[0], v[1], v[2], v[3]));
+				color.map_rgb(|channel| channel * color.a())
 			})
 			.collect();
</file context>

[Keavon]

Superseded by #4518.