He just crawled through hell to fix the browser…

A former React core team engineer at Midjourney, Changlu, is betting that browser text measurement no longer has to be expensive. His library, pretext, aims to deliver fast, secure, accurate, and comprehensive text metrics in pure TypeScript—without forcing the browser to run layout reflows just to figure out line breaks, paragraph height, or where text should wrap. That matters because text-heavy UI patterns like virtualized lists and masonry layouts often stall on the cost of repeatedly asking the browser for dimensions.

In most browsers, measuring text height triggers layout reflow: the engine recalculates geometry for elements across the page to answer questions like “how tall is this paragraph?” Reflow is among the most expensive operations a browser can perform, and it becomes a bottleneck when an app needs thousands of measurements during scrolling or layout. pretext’s core move is to bypass the browser’s standard text measurement pipeline. For widths, it leans on the canvas API, which can measure string pixel widths outside the DOM without layout work. The harder part is height, because height depends on line breaking rules—rules that vary across browsers and languages.

To solve line breaks, Changlu built a custom algorithm that predicts how text wraps, then used automated “clankers” (AI workers) to generate and validate the logic. The workflow described is iterative: produce line-break logic, test it against real text in real browsers, compare results, and repeat until the algorithm is stable. The end product is a simple API: first “prepare” text by segmenting it and caching each segment’s width, then call “layout” to compute total height and line count without touching the DOM or triggering reflow.

The transcript also shows why that capability unlocks new UI techniques. One demo builds a character grid from video frames: the app uses pretext to determine exactly which characters belong in which columns and rows, then renders the video into a tiny offscreen canvas sized to the grid. It reads pixel brightness from the canvas and uses that brightness to decide whether each character cell is drawn or skipped—effectively sculpting an image out of letters. In that pipeline, pretext handles the precision of character placement, while canvas handles the pixel-level rendering.

Even if pretext doesn’t become the default foundation for all UI measurement, it’s presented as proof that browsers don’t have to “own” text measurement. The broader implication is practical: if apps can compute text geometry cheaply and deterministically, developers can build smoother, more scalable text-heavy interfaces without paying the reflow tax.