PDF does not have native support for box shadows, so we have to render it into an image and then include it in the PDF.

Suggested implementation: We add the corresponding API on the Typst side, and based on the parameters we generate an SVG and use SVG filters to get the desired effect (e.g. Gaussian blur). Then, we simply convert this SVG using usvg and include the svg in the desired position. Implementing this should be pretty straight-forward, the only tricky thing might be positioning the SVG and getting the dimensions right, but that shouldn't be impossible to achieve.

PDF does not have native support for box shadows, so we have to render it into an image and then include it in the PDF.

Suggested implementation: We add the corresponding API on the Typst side, and based on the parameters we generate an SVG and use SVG filters to get the desired effect (e.g. Gaussian blur). Then, we simply convert this SVG using usvg and include the svg in the desired position. Implementing this should be pretty straight-forward, the only tricky thing might be positioning the SVG and getting the dimensions right, but that shouldn't be impossible to achieve.

It doesn't have to be an image, it could be based on gradients no?

I don't think you can easily simulate a Gaussian blur with a gradient, although I haven't tried it. But still, rasterizing would be a good enough first step, I think.

I don't think you can easily simulate a Gaussian blur with a gradient, although I haven't tried it. But still, rasterizing would be a good enough first step, I think.

You can do almost anything with mesh gradients, but yeah

i think its important to mention https://github.com/typst-community/harbinger , a package that aims to implement shadows (using svg filters and such, as well as with gradients), which we haven't published yet (mainly because of problems with the rendering of svg filters, i should make an issue about it actually)

i have a couple ideas that could remove the need for svg filters (mainly doing with mesh gradients aproximating gaussian blur, like enivex suggested), but i haven't have the time to implement them yet.