<?xml version="1.0" encoding="utf-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
    <channel>
        <title>Comparing performance of Lucre2D (alpha experiment) to Processing on a Pi 4</title>
        <link>https://diode.zone/videos/watch/6e97d9df-9637-44cc-a533-af4560e4f752</link>
        <description>I quickly hacked together a 2D graphics layer for my functional-reactive / dataflow model 'Lucre', with a Processing inspired front API, which allowed me to translate the 'Tree' L-System sketch by Daniel Shiffman. While running smoothly on a laptop, the test on a Pi 4B immediately reveals that my library has hardly usable performance at this stage. First part of the video shows the few-frames-per-second of Lucre2D, followed by the totally smooth animation coming from Processing 3.</description>
        <lastBuildDate>Tue, 20 Oct 2026 15:49:05 GMT</lastBuildDate>
        <docs>https://validator.w3.org/feed/docs/rss2.html</docs>
        <generator>PeerTube - https://diode.zone</generator>
        <image>
            <title>Comparing performance of Lucre2D (alpha experiment) to Processing on a Pi 4</title>
            <url>https://diode.zone/lazy-static/avatars/61b689b7-84d5-43c7-9fa8-e63a7fd1d952.png</url>
            <link>https://diode.zone/videos/watch/6e97d9df-9637-44cc-a533-af4560e4f752</link>
        </image>
        <copyright>All rights reserved, unless otherwise specified in the terms specified at https://diode.zone/about and potential licenses granted by each content's rightholder.</copyright>
        <atom:link href="https://diode.zone/feeds/video-comments.xml?videoId=6e97d9df-9637-44cc-a533-af4560e4f752" rel="self" type="application/rss+xml"/>
    </channel>
</rss>