GSoC 2025: Marquee Selection Masking

[mTvare6]

About:

I'm Austin, a statistics student in IITK and I will be working on my project titled marquee selection masking to implement various features related to masking and clipping into Graphite.

Synopsis

The goal of this project is to supplant the Mask node with a mask input to every graphical node, which takes in a general graphic input. This lays the foundation to build further nodes and operations such as one sided stroke alignment, gradient across strokes. Further a new “mask mode” will be introduced, which works at tooling level to enable a way to implement marquee selection masking, which unlike traditional mask tools(clipping) also supports gradient based operations(masking) and supporting much more than simple lasso, freeform or rectangle type selection present in other tools. Further, a system to show the types of nodes merged through the merge node will be implemented, along with their mask information.

Deliverables

• Add a mask input to most graphical node which uses the mask in some form and remove the old mask node, which is then equivalent to merge with mask.
• Implement the infrastructure to render the output of the mask node with SVGs and vello.
• Add the MaskMode to the DocumentMode, with the live masking ants shader running then, which as expected for a DocumentMode affects tooling behaviour.
• Add the system required to propagate nodes merged through the Merge node in the layer panel, along with masks of nodes if present.
• Flood fill node which takes in a point, tolerance and color.

[mTvare6]

Week 1

This week I've mostly been writing the infrastructure neccessary for clipping, as it would be simpler and act precursorial to masking. This involved using a SVG mask tag if any alpha change is involved and otherwise uses clipPath to simplify operations involved.

• Implement clipping masks, stroke align, and stroke paint order #2644

no clip	clipped	fill set to 0

The same works for groups, raster target and vector data.

fill set to 100	fill set to 0

[mTvare6]

Week 2

I started off this week by fixing new bugs and my old PR which didn't work over time.

• Transfer pasted layers to the viewport center when outside the current view area #2306
• Fix regression causing last selected layer to not be deselected #2656

Then into writing a system so that clipPath can be alternatively used instead of mask when possible, as it is faster and has lesser artifacts.

clipPath	mask

Then, along with the help of @Keavon who wrote the frontend part, and assets, I implemented a indicator for clipped layers in Layer panel.

And finally I ended the week by fixing the logic used in clip and mask usage, to ignored layers with stroke.

before	after

[mTvare6]

Week 3

I started off this week by working on part 6 of instance table refactor.

• Instance tables refactor part 6: remove usage of one_instance_* functions #2672

It mostly involved rewriting method to work with many rows across the table instead of the first, and a followup to an old PR

• Fix #[range] to act as a soft min/max #2678

It allows #[range(x, y)] to act like soft_(min|max).

And I ended up implementing masking in vello, which looks exactly the same and also makes inverted masking a breeze compared to doing so with SVGs:

8xU5.mp4

normal	inverted

Then, I worked on stroke alignment, by using the same technique used for masking, as neither vello nor svg have built-ins. I had also implemented paint-order for strokes and fills, which was now straight forward in svg and implemented in vello by changing the order of calling scene.fill and scene.stroke.
Like with other software, stroke align isn't implemented in open vector data for simplicity.

center	inside	outside

Paint Order:

stroke last	fill last

[mTvare6]

Week 4

I started off this week by working on the following bug fixes, which were mostly minor UX fixes
https://discord.com/channels/731730685944922173/881073965047636018/1358697354936713297
https://discord.com/channels/731730685944922173/881073965047636018/1362322748524986519
https://discord.com/channels/731730685944922173/881073965047636018/1360813352376598598

• Improve the click targets for the skew triangles of the transform cage #2703
• Fix grid overlay values getting wiped when switching between isometric/rectangular and back #2702
• Fix inability to deselect nodes, and nodes not being selected when pasted #2691

Then I finished off the week by

• Fixing https://discord.com/channels/731730685944922173/881073965047636018/1362323140784689254, mostly trying to rework the logic in delete_nodes but it keeps breaking in certain edge cases from expected behaviour of previous code
• Changed the bounding box of skew triangles to be rects instead of triangles in Improve the click targets for the skew triangles of the transform cage #2703

[mTvare6]

Week 5

• Fixed bug where Outside stroke alignment would ignore fill due to masking by instead using stroke order to get similiar behaviour if there is fill (otherwise continue with masking)
• Fixed bug with edge case allowing gradients work messing with path's stroke alignment transform
• Rewrote the mask node so that masks no longer rely on raster content but instead work with any content, and render the same with svg or vello, which will eventually allow some nodes to consume the mask and modify data in place if required, or is otherwise sent straight to renderer