"use client";import {  ArticlePage,  ArticlePageAuthor,  ArticlePageCategory,  ArticlePageContent,  ArticlePageDescription,  ArticlePageDetails,  ArticlePageHeader,  ArticlePageMeta,  ArticlePageRelated,  ArticlePageRelatedItem,  ArticlePageRelatedList,  ArticlePageRelatedMeta,  ArticlePageRelatedTitle,  ArticlePageShare,  ArticlePageTextSizeControl,  ArticlePageTextSizeOption,  ArticlePageTitle,  ArticlePageToolbar,  type ArticlePageVariant,} from "@/components/uai/article-page";import { AuthorCardAvatar, AuthorCardName } from "@/components/ui/uai/author-card";import {  ShareMenuChannel,  ShareMenuContent,  ShareMenuCopy,  ShareMenuNative,  ShareMenuSeparator,  ShareMenuTrigger,} from "@/components/ui/uai/share-menu";export function ArticlePagePreview({ variant = "centered" }: { variant?: ArticlePageVariant }) {  return (    <ArticlePage variant={variant}>      <ArticlePageHeader>        <ArticlePageCategory>Engineering · Build performance</ArticlePageCategory>        <ArticlePageTitle>How we cut cold builds from 41 seconds to 9</ArticlePageTitle>        <ArticlePageDescription>          Kiln 4.2 caches parsed templates on disk and skips image work that has not changed. Here          is what we measured, what we threw away, and what is still slow.        </ArticlePageDescription>        <ArticlePageMeta>          <ArticlePageAuthor>            <AuthorCardAvatar name="Ines Moreau" />            <AuthorCardName>Ines Moreau</AuthorCardName>            <ArticlePageDetails>              <time dateTime="2026-09-24">September 24, 2026</time>              <span aria-hidden="true">·</span>              <span>8 min read</span>            </ArticlePageDetails>          </ArticlePageAuthor>          <ArticlePageToolbar>            <ArticlePageTextSizeControl>              <ArticlePageTextSizeOption value="small" aria-label="Small text">                A              </ArticlePageTextSizeOption>              <ArticlePageTextSizeOption value="default" aria-label="Default text">                A              </ArticlePageTextSizeOption>              <ArticlePageTextSizeOption value="large" aria-label="Large text">                A              </ArticlePageTextSizeOption>            </ArticlePageTextSizeControl>            <ArticlePageShare              url="https://kiln.example/blog/cold-builds"              shareTitle="How we cut cold builds from 41 seconds to 9"            >              <ShareMenuTrigger />              <ShareMenuContent align="end">                <ShareMenuCopy />                <ShareMenuNative />                <ShareMenuSeparator />                <ShareMenuChannel href="https://mastodon.example/share?url=https%3A%2F%2Fkiln.example%2Fblog%2Fcold-builds">                  Mastodon                </ShareMenuChannel>                <ShareMenuChannel href="mailto:?subject=Kiln%20cold%20builds&body=https%3A%2F%2Fkiln.example%2Fblog%2Fcold-builds">                  Email                </ShareMenuChannel>              </ShareMenuContent>            </ArticlePageShare>          </ArticlePageToolbar>        </ArticlePageMeta>      </ArticlePageHeader>      <ArticlePageContent>        <p>          A cold build is the first build after cloning a site or clearing the cache. For a          2,400-page documentation site, Kiln 4.1 spent 41 seconds there, and most of it was work we          had already done on the previous machine.        </p>        <h2>Where the time went</h2>        <p>          Profiling twelve large sites showed the same shape: 58% of the time parsing templates, 27%          resizing images, and the rest writing files. Template parsing was the surprise. The same          140 partials were parsed again for every page that used them.        </p>        <blockquote>          We assumed images were the problem because they were the loudest part of the log.        </blockquote>        <h2>What changed in 4.2</h2>        <ul>          <li>Parsed templates are stored in the cache, keyed by their content hash.</li>          <li>Images are only resized when the source or the requested size changes.</li>          <li>Pages write in parallel, up to the number of available cores.</li>        </ul>        <p>          On the same site, a cold build now takes 9 seconds and a warm build takes under 2. Read          the <a href="#cache-guide">cache guide</a> to share the cache between CI runs.        </p>      </ArticlePageContent>      <ArticlePageRelated>        <ArticlePageRelatedTitle>Related reading</ArticlePageRelatedTitle>        <ArticlePageRelatedList>          <ArticlePageRelatedItem href="#incremental">            Incremental builds, explained            <ArticlePageRelatedMeta>6 min read</ArticlePageRelatedMeta>          </ArticlePageRelatedItem>          <ArticlePageRelatedItem href="#ci-cache">            Sharing the build cache in CI            <ArticlePageRelatedMeta>4 min read</ArticlePageRelatedMeta>          </ArticlePageRelatedItem>          <ArticlePageRelatedItem href="#release-4-2">            Kiln 4.2 release notes            <ArticlePageRelatedMeta>Changelog</ArticlePageRelatedMeta>          </ArticlePageRelatedItem>        </ArticlePageRelatedList>      </ArticlePageRelated>    </ArticlePage>  );}