Recent content by Jazzzny

I worked on this some more and I believe I've come up with a much nicer solution. After comparing several different problematic fonts, I've determined that ATS will reject a font if: 1. The font name in the CFF table's name index and SFNT postscript name in the name table under Name ID 6 are...

Another nice thing for the next release: I've been able to implement a fallback for when ATS/CoreGraphics fails to load a font by using FreeType, meaning that all web fonts will now render correctly:

I had to do quite of reversing in CoreGraphics, but after shimming a few functions and writing what I think is a pretty clever replacement for CGContextClipToMask using CGContextSetCompositeOperation(cg, kPrivateCGCompositeSourceIn), I'm happy to say that the 10.3 build is now rendering...

Looking a lot better!

I don't think I really ran into anything too CoreText-specific, most of the 10.4 code was already implemented using CGFont and I was able to replace the remaining CTFont references with ATSFont.

This is going to need a lot more work, but I'm happy to say that PowerFox will support 10.3!

On GitHub: https://github.com/Jazzzny/powerfox-browser/issues

Hello everyone, I'm happy to announce the release of PowerFox 26.3.0! This release includes the IonPower JIT for substantial JavaScript improvements, and many new features, improvements, bug fixes, and security fixes: Implemented IonPower JavaScript JIT compiler for significant performance...

I'm happy to say that, despite 10.4's incredibly limited and rather broken CoreText, I was able to get Skia up here too!

This still has a few issues, but I've made very good progress towards getting Skia working on PPC, which results in much faster browser compositing compared to the CoreGraphics backend. For comparison, the cross dissolve on the main image of https://powerfox.jazzzny.me/ finishes around twice as...

I have worked through all of the assertions, crashes, and conformance issues I could find, and I can now confirm that the browser is working correctly: Still some room for optimization, but I really don't know how I would approach that. This is already around 21 times faster than JITless. On...

Luckily I was able to get the full gigabyte out, vmmap showed that the original address would indeed have caused a collision, but I was able to shift it up a bit and still stay in mappable memory. Funnily enough the ld64 I am using is seemingly bugged and strips __UNIXSTACK when -dead_strip is...

Yes, this is with the browser, I'll see how much I can get out, thanks.

Did you have to do anything special to get the larger stack to function? When I pull in the linker argument, it creates a binary that instantly crashes with r1 zeroed out.

Thanks, I've located your changes in the appropriate moz.build files. Do you happen to remember the reason for the stack size increase? I didn't have this in for quite a while and it seemed to work okay, though I haven't gotten any benchmarks to run yet.