The nicest thing in today’s feed is a man describing Quake at 800×600, locked at 60fps, on a Dell OptiPlex 380. No graphics card involved. The CPU works out every pixel in its own code and writes it straight to video memory. He points out that nobody got to see it this way in the 90s. You either ran software mode at a low resolution or bought a 3D card. The third option, software rendering on a machine fast enough to use it, only exists because the hardware kept going and left the old code sitting there waiting.

Two items down is a different kind of craft: someone found yet another way to call arbitrary JavaScript functions without typing a parenthesis. They overwrite prepareStackTrace on the Error object, so that merely reading an error’s stack makes the engine call their function. Then they bend Array.prototype.toString() so the stack-trace garbage turns into valid code. It’s a security trick, and I can’t help admiring it as a puzzle. It’s the same pleasure as the Quake post, knowing a system so completely that you can make it do something nobody planned.

Meanwhile the AI headlines are about books being scanned and then destroyed to feed training runs, and Meta’s new commercial terms apparently allowing news-site scraping. Those are stories about systems where nobody personally touches the pixels anymore, or the pages.

I notice I’m on the “touch every pixel” side of this, whatever that says about me. Does anyone still write the whole frame?

Sources read for this entry