Properly render gradients when strokeScaling = false

[adroitwhiz]

Description

This was a fun bug 🙂 I like paper.js 🙂

See paperjs#1822 for more info. This is necessary to make bitmap shape tools work properly with gradients, otherwise this will happen:

Related issues

• Relates to Gradients don't render properly on items where strokeScaling = false paperjs/paper.js#1821

Checklist

• New tests added or existing tests modified to cover all changes
• Code conforms with the JSHint rules (npm run jshint passes)

[fsih]

Is this fixable by correctly triggering the _changed event when strokeMatrix changes?
It seems brittle to keep track of exceptions around caching

[adroitwhiz]

I could manually trigger _changed in this block but that seems more fragile to me-- the color code would no longer be responsible for ensuring its own caching behavior.

[adroitwhiz]

And come to think of it that would suffer from the same problem as commented below-- if strokeMatrix stopped being passed in, _changed would stop being called and the gradient would retain its previous incorrect transform

[fsih]

I'm working on understanding the paper.js code, and it looks like the Color class already doesn't keep track of its own matrix; it's relying on the parent item to call _transform on the color any time the item's matrix changes, otherwise if you call toCanvasStyle twice in a row with different matrices you'll get the wrong result.

What if the color stored what was in the matrix and strokeMatrix, and used that to determine whether its cached style was still valid?

Are we assuming that any time a strokeMatrix is passed in, it implies that we're coloring a stroke?

[adroitwhiz]

strokeMatrix, confusingly, actually refers to (IIRC) the matrix by which path points are transformed to obtain the view transformation. It's applied to the fill and stroke when strokeScaling is false.

The problem is that it would also need to be invalidated whenever the view transform changes.

[adroitwhiz]

@fsih Did we determine that this is OK to merge?

[fsih]

Oops, yes I think this looks good to merge!

[jkozniewski]

So there is a good working pr from 2020 and still not fixed in 2022 ?
Any chance for it to be actually merged into main paperjs branch anytime soon ?