build: bump Astro from 6 to 7

[stasadev]

The Issue

See https://docs.astro.build/en/guides/upgrade-to/v7/

How This PR Solves The Issue

Astro 7 changes that needed code fixes:

• New Rust compiler is stricter about HTML. Every non-void element now needs a closing tag; the old Go compiler used to fix this by itself. Added the missing </Layout> tags in survey.astro and newsletter.astro.

• Markdown now runs through a processor. Moved remarkPlugins and rehypePlugins into unified({...}) from @astrojs/markdown-remark so the deprecation warning is gone.

• Astro 7 also moves to Vite 8 (handled by the dependency bumps).

• src/lib/search-index.js used an undefined fs (the import was named fs2). The bug was there before but stayed hidden; Astro 7 fails the build:done hook on it. Renamed the import to fs.

• package.json override:

  "overrides": {
    "astro-rehype-relative-markdown-links": {
      "astro": "$astro"
    }
  }

  astro-rehype-relative-markdown-links does not yet list Astro 7 as a supported peer, but it works fine. The override makes it use our Astro version and avoids a peer-dependency install error. See Support Astro v7 (update peer dependency range) vernak2539/astro-rehype-relative-markdown-links#84

Manual Testing Instructions

Compare https://20260723-stasadev-astro-7.ddev-com-front-end.pages.dev/ vs https://ddev.com/

• https://20260723-stasadev-astro-7.ddev-com-front-end.pages.dev/blog/markdown-features-demo/ vs https://ddev.com/blog/markdown-features-demo/
• https://20260723-stasadev-astro-7.ddev-com-front-end.pages.dev/usage-stats/ vs https://ddev.com/usage-stats/

Click different links and confirm that everything works as expected in both desktop and mobile views.

Automated Testing Overview

Release/Deployment Notes

[cloudflare-workers-and-pages]

Deploying ddev-com-front-end with    Cloudflare Pages

Latest commit:	924884f
Status:	✅  Deploy successful!
Preview URL:	https://c5b224db.ddev-com-front-end.pages.dev
Branch Preview URL:	https://20260723-stasadev-astro-7.ddev-com-front-end.pages.dev

View logs

[rfay]

Claude Review: differential crawl of both deployed sites

Instead of eyeballing the three listed pages, I compared all 216 sitemap pages on prod (ddev.com, confirmed Astro v6.3.5) vs the preview deploy (confirmed Astro v7.1.3). The script fetches each path from both hosts and diffs three things per page — HTTP status, visible text (tags stripped), and all link/asset targets — after normalizing out build-specific noise (content hashes, generator version, scoped-style CIDs). A second pass scrubs live dynamic data to separate real differences from time-based churn.

Results — no regressions

Check	Result
HTTP status, all 216 pages	200/200 on both — no broken pages
Text identical after scrubbing dynamic data	205 / 216
Remaining 11 diffs	all explained as non-regressions (below)
Generated outputs	search.json 10 entries both · blog/rss.xml 50 items both · /404 renders both

The 11 residual differences:

• Cloudflare email obfuscation (8 pages: contact, privacy, and several blog posts) — the production zone rewrites emails into [email protected] + a cdn-cgi decode script; the pages.dev preview shows the true email. Edge feature of the prod zone, not this PR. Same cause as the /cdn-cgi/l/email-protection link diffs.
• Live dynamic data — homepage GitHub star count (127→128), usage-stats Amplitude numbers. Time-based, expected.
• /get-started/ — a newline vs space inside a download-button label (↓\n(Intel/AMD)); inline whitespace browsers collapse identically. Cosmetic only.

Link/asset diffs reduced to build-hash changes plus giscus emitting giscus.app/client.js as a runtime setAttribute (Astro 7) vs a static src (Astro 6) — giscus is present and configured on both. Expected Vite/Astro-bump artifacts.

Highest-risk change verified clean: the markdown processor migration (remarkPlugins/rehypePlugins → processor: unified({...})). On markdown-features-demo every structural feature matches exactly — callouts (76), figures/figcaptions (4/4), autolinked headings (10), shiki blocks (28), copy buttons (16), table wrappers (16), emoji labels (21).

Code review

All 8 files look correct and minimal:

• astro.config.mjs — plugin lists moved verbatim into unified(); order preserved.
• search-index.js — the fs2→fs fix is a real latent bug fix (the old name was undefined).
• newsletter.astro / survey.astro — added the missing </Layout> the old Go compiler had been auto-closing.
• package.json deps/override/allowScripts, plus the two whitespace-only content/doc edits — all benign.

Recommendation: approve. The preview deploy building and serving all 216 pages is itself proof the Astro 7 build succeeds. The only PR-attributable rendering difference is the cosmetic /get-started/ whitespace, which is visually identical in-browser.

Comparison script used (compare.sh)

#!/usr/bin/env bash
# Compare prod (Astro 6) vs preview (Astro 7) for every sitemap path.
set -u
WD=~/tmp/astro7-diff
cd "$WD"
PROD="https://ddev.com"
PREV="https://20260723-stasadev-astro-7.ddev-com-front-end.pages.dev"
mkdir -p prod prev norm diffs
: > status.tsv
: > report.txt

# Normalizer: remove build-specific noise so only real diffs remain.
normalize() {
  perl -0777 -pe '
    s{/_astro/([A-Za-z0-9_\-]+?)\.[A-Za-z0-9_\-]{6,}(\.[a-z0-9]+)}{/_astro/$1$2}g; # asset content hashes
    s{Astro v[0-9]+\.[0-9]+\.[0-9]+}{Astro vX}g;                                    # generator version
    s{data-astro-cid-[a-z0-9]+}{data-astro-cid-X}g;                                 # scoped style ids
    s{astro-[a-z0-9]{8,}}{astro-X}g;                                                # scoped class names
    s{\?v=[a-z0-9]+}{?v=X}gi;                                                        # cache-busting query
  '
}
# Visible text: strip script/style, then all tags, collapse whitespace.
textof() {
  perl -0777 -pe 's{<script\b.*?</script>}{}gis; s{<style\b.*?</style>}{}gis; s{<!--.*?-->}{}gis; s{<[^>]+>}{ }gs;' \
   | perl -0777 -pe 's{&nbsp;}{ }g; s{&amp;}{&}g; s{&#39;}{'"'"'}g; s{&quot;}{"}g;' \
   | tr -s ' \t\r\n' ' ' | sed 's/^ //; s/ $//'
}
# Link/asset targets, normalized and sorted.
linksof() {
  grep -oiE '(href|src)="[^"]*"' | sed -E 's/^(href|src)=//I' | tr -d '"' | sort -u
}

n=0
while IFS= read -r url; do
  path="${url#https://ddev.com}"
  [ -z "$path" ] && path="/"
  slug=$(echo "$path" | sed 's#/#_#g; s/^_//; s/_$//'); [ -z "$slug" ] && slug="home"
  n=$((n+1))
  pc=$(curl -s -o "prod/$slug.html" -w "%{http_code}" "$PROD$path")
  vc=$(curl -s -o "prev/$slug.html" -w "%{http_code}" "$PREV$path")
  printf '%s\t%s\t%s\n' "$pc" "$vc" "$path" >> status.tsv

  if [ "$pc" = "200" ] && [ "$vc" = "200" ]; then
    normalize < "prod/$slug.html" | textof > "norm/$slug.prod.txt"
    normalize < "prev/$slug.html" | textof > "norm/$slug.prev.txt"
    normalize < "prod/$slug.html" | linksof > "norm/$slug.prod.links"
    normalize < "prev/$slug.html" | linksof > "norm/$slug.prev.links"
    tdiff=$(diff "norm/$slug.prod.txt" "norm/$slug.prev.txt")
    ldiff=$(diff "norm/$slug.prod.links" "norm/$slug.prev.links")
    if [ -n "$tdiff" ] || [ -n "$ldiff" ]; then
      { echo "########## $path";
        [ -n "$tdiff" ] && { echo "--- TEXT DIFF ---"; echo "$tdiff"; }
        [ -n "$ldiff" ] && { echo "--- LINK/ASSET DIFF ---"; echo "$ldiff"; }
        echo; } >> diffs/differences.txt
      echo "DIFF  $path" >> report.txt
    else
      echo "OK    $path" >> report.txt
    fi
  else
    echo "STATUS pc=$pc vc=$vc  $path" >> report.txt
  fi
  [ $((n % 40)) -eq 0 ] && echo "...processed $n"
done < urls_full.txt
echo "DONE: $n pages"

_{🤖 Review assisted by Claude Code}

[rfay]

Manual clicking around also found no issues.

[rfay]

Claude caught a regression in get-started:

[rfay]

I took the liberty of pushing the tiny change to windows.astro.

[rfay]

Claude commented on compressHTML — worth consideration, low risk

The default flipped from true (HTML-aware) to 'jsx' (React-style), a documented breaking change requiring whitespace review. The repo doesn't set it, so it takes the new default. My 216-page crawl found no whitespace regressions attributable to it (the only whitespace diff was the \n bug above, which is unrelated). So it's low-risk as-is; optionally pin compressHTML: true in astro.config.mjs to guarantee Astro-6-identical output, or leave it and rely on the crawl evidence.