As a sidequest of a sidequest of a sidequest [1], I'm adding a "Convert to Bilevel…" feature to SerenityOS's PixelPaint.

I've been wanting to implement Floyd-Steinberg error diffusion dithering (something much simpler than this long name might suggest) for literally decades, cool that I finally got around to it :)

1: PDFs can contain an obscure file format called JBIG2, so I needed to write a JBIG2 decoder. That needs tests, there are very few JBIG2 encoders out there, and the ones that exist use way fewer features than what's found in PDFs, so I'm writing a JBIG2 encoder too. JBIG2 can only store black and white pixels, so I needed to write a "convert to bilevel" feature. And since I have that anyway, might as well hook it up!

PR: https://github.com/SerenityOS/serenity/pull/26182

Over the summer break, I added support for clipping to arbitrary paths to Serenity's LibPDF. Here's a before/after of one of my test files, together with a visualization of the clipping paths.

https://github.com/SerenityOS/serenity/pull/26035, the PR that added the feature, has many more before/after screenshots.

The PR looks fairly obvious and straightforward in hindsight, but it took me a while to get it to that shape. My first local proof of concept was extremely slow: For my initial test page (page 1140 of the PDF 1.7 spec), it originally was a 250x (25000%) slowdown. The final version is "just" 5% slower for that page.

There's like 4 larger features missing and LibPDF will be able to render most complicated files :^)

Took a break from jpeg2000 and started on the PDF `sh` operator over the weekend. Many before/after screenshots at https://github.com/SerenityOS/serenity/pull/25717

The most dramatic one:

Over the last few months, I wrote a JPEG2000 decoder for SerenityOS. JPEG2000 is one of the image formats that can appear in PDF files. I finally sent out the last pull request in that JPEG2000 support patch series that connects the image decoder to the PDF library: https://github.com/SerenityOS/serenity/pull/25693

I found writing a JPEG2000 decoder pretty difficult!

I thought it's because it has so much more stuff, but looking at line counts it's in the same ballpark as other image decoders in the system:

JBIG2: ~2.9 kLoC
JPEG: ~2 kLoC
JPEG2000: ~3 kLoC
PNG: ~1.5 kLoC (not counting the deflate code though)
WebP(lossy+lossless): ~3 kLoC

Maybe it's because the other formats are a bit more horizontal, in that you don't need to do all the features before you can decode a simple image. JPEG2000 feels like it needs a lot of machinery working before anything works. Also, the example codestream in the spec is too simple to be useful.

Maybe I'll write a bit more about the format later. In the meantime, the PR description of https://github.com/SerenityOS/serenity/pull/25670 has an overview of the rough steps. (I think it's pretty readable, but it's well possible that it's very cryptic for someone new to the format.)