The 386 microcode sequencer has a one-cycle pipeline delay: when a jump or RNI (run next instruction) is decoded, the micro-instruction immediately after it has already been fetched and will execute before the jump takes effect. This "delay slot" is a basic property of the sequencer, and the microcode is written to fill it with useful work rather than waste a cycle on a bubble. The examples in the PTSAV section above show this: at 582/5AE, the micro-instruction after LCALL executes before the subroutine begins.
For our spreadsheet example, this step just generates a list of all dirty nodes, because every node (i.e. cell) is kind of an output node and an input node at the same time. In that case, you’d probably instead create a list of all nodes with no children. However, in a GUI framework, you might have “effect” nodes that are responsible for updating UI components — these are also leaves in the tree, but they’re specifically output leaves, because they form the observable part of our reactivity graph. This also means that an intermediate node that has no output won’t ever end up in this list, and therefore won’t get updated. This is something I’ll write about more in a follow-up post! ↩︎，这一点在新收录的资料中也有详细论述
论文显示，Silica 采用两类体素写入方式：一种是基于折射率各向异性的双折射体素，另一种是基于折射率变化的相位体素。，详情可参考新收录的资料
You still have to bundle the generated Gleam code into a single file using a bundler like Webpack/Parcel/Rollup/Esbuild. Deno used to support bundling applications with deno bundle but it has been deprecated in favor of other bundlers stated before.。新收录的资料是该领域的重要参考