Multiple bugs / inconsistencies related to nested filter layers on alpha channel

[NotMyFault]

Hi,

 

there are numerous bugs / inconsistencies related to nested filter layers on alpha channel.

My expectations:

• E1 nested filter layers in clipping position get restricted to the parent shape for vector shapes.
• E2 nested filter layers in clipping position get restricted to the content of a group layer, but the results can reach beyond.
• E3 nested filter layers in masking position get restricted to original content of the parent shape, which might get relocated outside that area.
• E4 Nested filters will get restricted to their inherent mask. If the mask is fully black, the filter should not impact any pixel.

The test file shows a RGB color gradient on x-axis, a alpha gradient on Y-axis, a fully opaque rectangle, and a ripple live filter nested to vector shapes or groups, in clipping or masking position, with inherent mask filled or cleared.

The image below has 8 test tiles (A1-A4, B1-B4), rows by letter A, B, columns by number 1 to 4.

• A1 looks OK
• A2 looks OK
• A3 look wrong. violates E4. 
• A4 looks wrong. Blending of group with partial alpha leads to wrong result in the area outside the parent shape. It should deliver the same rendering as the layers without any grouping.
• B1 looks OK, but there are (probably unavoidable) artefacts at the edge on alpha channel coming from green stroke
• B2 looks wrong. violates E4 bad anti-aliasing on bottom edge
• B3 looks OK.
• B4 looks wrong. violates E4 (probably only on alpha channel)

Edit: groups use default blend mode "passthrough". If changing to "normal", results B2, A4, B4 look ok.

B3 is the main problem.

Same but Alpha channel only selected

nested alpha live filer again V2.afphoto

[BeauRX]

@NotMyFaultFantastic job enumerating the conditions! It's hard to pick out exactly what is wrong with an image, just that something does not look right/the way you want. Problems with a single feature are annoying, but this type of issue is much more concerning as it is the core function of compositing the layers. They really should be doing nightly build tests with a suite of these types of very specific edge/corner cases.

To add to your list, adding a Merge Visible layer on top does not duplicate the gradient close to the sum of the combined layers. I believe I have seen this elsewhere but your file provides a great test case for Affinity.

nested alpha live filer again V2_MV.afphoto

[NotMyFault]

20 minutes ago, BeauRX said:

To add to your list, adding a Merge Visible layer on top does not duplicate the gradient close to the sum of the combined layers. I believe I have seen this elsewhere but your file provides a great test case for Affinity.

Thank you for your reply.

The merge visible is actually ok in this case, no bug. The irritation is caused by partial transparency. You need to put a fill layer in white below the merged layer to see correct result, or deactivate all layers below.

[EmT]

Hi @NotMyFault

This is something we are aware of and it has been passed over to our developers for further investigation.