<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/">
  <channel>
    <title>RoobLi Posts</title>
    <link>https://www.roobli.org/posts/</link>
    <description>Short dated posts on RoobLi, newest first</description>
    <atom:link href="https://www.roobli.org/posts/index.xml" rel="self" type="application/rss+xml"/>
    <item>
      <title>Ten projects, shown working</title>
      <link>https://www.roobli.org/posts/2026-10-22-projects</link>
      <guid isPermaLink="true">https://www.roobli.org/posts/2026-10-22-projects</guid>
      <description>The project pages were overviews of about 300 words.</description>
      <content:encoded><![CDATA[<p>The project pages were overviews of about 300 words. Each is now a proper introduction: who the tool is for, its features shown running, a real session, how it is built and where it stops. Every number comes from the project’s own tests, benchmarks or releases. Start with <a href="https://www.roobli.org/projects/noto" class="internal internal-link alias" data-slug="projects/noto">Noto</a>, <a href="https://www.roobli.org/projects/tls-companion" class="internal internal-link alias" data-slug="projects/tls-companion">TLS Companion</a> or <a href="https://www.roobli.org/projects/tmux-radar" class="internal internal-link alias" data-slug="projects/tmux-radar">tmux-radar</a>; the rest are on <a href="https://www.roobli.org/projects/" class="internal internal-link alias" data-slug="projects/index">the projects page</a>.</p>
<p><a href="https://www.roobli.org/essays/emudhra-weak-keys" class="internal internal-link alias" data-slug="essays/emudhra-weak-keys">Eight CA keys that fit in one byte</a> is rewritten at three times its length, with certificate anatomy from the real certificates, ECDSA worked by hand on a 19-point curve, two new figures and labs to run yourself. It is also <a href="https://www.roobli.org/essays/emudhra-weak-keys/zh" class="internal internal-link alias" data-slug="essays/emudhra-weak-keys/zh">in Chinese</a>.</p>
<p>Screenshots are stored at twice their display size, so they stay sharp on dense screens, and a click opens one full screen: arrows or a swipe page through the rest, a click or pinch zooms in, Escape or a drag down puts it back.</p>
<p>This timeline now records every change here, not only posts. The filter above shows one kind at a time.</p>]]></content:encoded>
      <pubDate>Thu, 22 Oct 2026 00:00:00 GMT</pubDate>
    </item>
    <item>
      <title>苹果的 container：一个容器一台虚拟机</title>
      <link>https://www.roobli.org/posts/2026-09-30-apple-container</link>
      <guid isPermaLink="true">https://www.roobli.org/posts/2026-09-30-apple-container</guid>
      <description>在 Mac 上跑 Linux 容器，Docker Desktop 的做法是起一台共享的 Linux 虚拟机，把所有容器都塞进去。苹果开源的 container 反过来：每个容器一台轻量 Linux 虚拟机，直接用 Virtualization.framework 起，不另带 VMM 二进制。到 9 月 30 日它在 GitHub 上已有约 5 万 star。 区别一句话就能说清：容器假在视图和配额，microVM 假在硬件和客户机内核。每个工作负载一份内核，隔离边界就不再是「大家共用一个内核」。 代价也跟着来。内核要自己带，仓库里有一份为快速启动裁过的配置，内核从「宿主机碰巧是什么」变成了你要...</description>
      <content:encoded><![CDATA[<p>在 Mac 上跑 Linux 容器，Docker Desktop 的做法是起一台共享的 Linux 虚拟机，把所有容器都塞进去。苹果开源的 <code>container</code> 反过来：每个容器一台轻量 Linux 虚拟机，直接用 Virtualization.framework 起，不另带 VMM 二进制。到 9 月 30 日它在 GitHub 上已有约 5 万 star。</p>
<p>区别一句话就能说清：容器假在视图和配额，microVM 假在硬件和客户机内核。每个工作负载一份内核，隔离边界就不再是「大家共用一个内核」。</p>
<p>代价也跟着来。内核要自己带，仓库里有一份为快速启动裁过的配置，内核从「宿主机碰巧是什么」变成了你要维护的东西。它的定位是开发用的笔记本，不是机房；服务器上要把 OCI 镜像放进 microVM，还是看 Kata 或 libkrun。</p>]]></content:encoded>
      <pubDate>Wed, 30 Sep 2026 00:00:00 GMT</pubDate>
    </item>
    <item>
      <title>A fruit fly’s wiring diagram has been hooked up to Doom</title>
      <link>https://www.roobli.org/posts/2026-09-29-doomfly</link>
      <guid isPermaLink="true">https://www.roobli.org/posts/2026-09-29-doomfly</guid>
      <description>A fruit fly’s wiring diagram has been hooked up to Doom.</description>
      <content:encoded><![CDATA[<p>A fruit fly’s wiring diagram has been hooked up to Doom. I didn’t run DoomFly itself, only the pieces under it, and the most useful thing in the project is the author’s own review. The full male CNS connectome (about 166,000 neurons) really does drive the controls: cut every edge and the control disappears. But the baseline model has fixed weights, its controls stay active when the screen is painted black, and the actions are read from four descending neurons with fixed gains. A newer experimental mode adds a plasticity rule on about 4,200 connections; the author reports that it still fails the visual, conditioning and survival checks.</p>
<p>So the honest sentence is the author’s: a full connectome, simulated with approximate dynamics, drives an engineered Doom interface. A character that moves is not a fly that understands the game.</p>
<p>The part underneath is better news than the headline. The male CNS data sits in a public bucket with no account needed, and a whole-brain run of the female FlyWire brain took 14 seconds on a CPU, compile included.</p>]]></content:encoded>
      <pubDate>Tue, 29 Sep 2026 00:00:00 GMT</pubDate>
    </item>
  </channel>
</rss>
