definev/gputext

gputext is a library for the Flutter app building toolkit that renders text using the device's graphics processor, known as the GPU, instead of the usual software drawing method. It gives developers a widget called GPURichText that works as a near drop in replacement for Flutter's built in RichText, but draws each letter shape using a coverage shader, a technique for calculating how much of a pixel a shape covers so edges look smooth. The project is built as a Melos workspace, a way of organizing multiple related Dart packages inside one repository. The actual library code lives in packages/gputext, and there is a demo app in the example folder that shows several test pages, including an interactive ASCII dragon animation where every character, including Latin letters, Chinese characters, and color emoji, is drawn by the GPU shader. Because GPU based text rendering cannot cover every situation on its own, the library falls back to Flutter's normal text drawing for characters it cannot render itself, such as emoji or scripts the loaded fonts do not include. This mixed approach means some parts of a screen are drawn by the fast GPU path and other parts are drawn the ordinary way, and the project's benchmark suite measures both separately so it is clear which numbers apply to which drawing method. The repo includes an extensive benchmarking system that compares layout speed, frame timing, memory use, and visual accuracy against the standard Flutter text widget, plus a baseline file for tracking whether changes make things faster or slower over time. That level of tooling suggests this is aimed at teams optimizing performance in Flutter apps with heavy text content, not casual developers. There are important limits stated directly in the README: the library only works with Flutter's newer Impeller graphics engine, it does not support right to left or bidirectional text, and it does not fully use locale settings. It also appears to be a renamed continuation of an earlier project called windfoil_flutter, with a table mapping old names to new ones.