fix(render): unwrap a bare shiki class on pre back to a plain fence

[hendrikheil]

Keeps shiki-highlighted code blocks round-tripping to a plain fence instead of leaking a ::pre{.shiki} block-attribute wrapper.

The leak was introduced in #214, which added the ::pre{attr} wrapper form together with the userBlockAttrs strip guard — that guard never matched the bare shiki class.

The fix

The ::pre{attr} wrapper form treats any class that survives userBlockAttrs as a user attribute. userBlockAttrs strips shiki's injected class, but the guard only matched class="shiki " with a trailing space (the multi-token dual-theme form). Single-theme shiki emits a bare class="shiki", which slipped through and forced every highlighted code block to serialize as a ::pre{.shiki} wrapper — especially visible for code blocks nested in component slots (which also produced a duplicate closing ::).

Widened the guard from value.startsWith('shiki ') to value.startsWith('shiki').

Tests

packages/comark/test/plugins/highlight.test.ts:

• bare class="shiki" → plain fence
• multi-token class="shiki shiki-themes … dark:…" → plain fence (unchanged)
• highlighted code block nested in a component slot → plain fence, no ::pre

Full suite green (1079 passed, 2 todo); existing shiki SPEC fixtures unaffected.

🤖 Generated with Claude Code

[vercel]

@hendrikheil is attempting to deploy a commit to the NuxtLabs Team on Vercel.

A member of the Team first needs to authorize it.

[pkg-pr-new]

Open in StackBlitz

comark

npm i https://pkg.pr.new/comarkdown/comark@259

@comark/angular

npm i https://pkg.pr.new/comarkdown/comark/@comark/angular@259

@comark/ansi

npm i https://pkg.pr.new/comarkdown/comark/@comark/ansi@259

@comark/html

npm i https://pkg.pr.new/comarkdown/comark/@comark/html@259

@comark/nuxt

npm i https://pkg.pr.new/comarkdown/comark/@comark/nuxt@259

@comark/react

npm i https://pkg.pr.new/comarkdown/comark/@comark/react@259

@comark/svelte

npm i https://pkg.pr.new/comarkdown/comark/@comark/svelte@259

@comark/vue

npm i https://pkg.pr.new/comarkdown/comark/@comark/vue@259

commit: c5cfd14

[farnabaz]

Thanks for the PR @hendrikheil
I think the changes are somewhat too much and changing line 131 in attributes.ts would be enough:

-    if (key === 'class' && tag === 'pre' && typeof value === 'string' && value.startsWith('shiki ')) {
+    if (key === 'class' && tag === 'pre' && typeof value === 'string' && value.startsWith('shiki')) {

But as for code prop, does it come from Nuxt Content / MDC packages? Comark parser does not add code prop anymore.

[farnabaz]

We can keep code attribute in drop list for now to ease the migration and remove it later.

[hendrikheil]

The PR really contains two fixes:

1. strip guard: I can collapse it to startsWith('shiki'), I only used /^shiki(?:\s|$)/ so a genuine user class that happens to start with shiki (e.g. ::pre{.shiki-toggle}) isn't stripped as if it were highlighter output. Cheap insurance, but I'm fine going with the simpler form if you'd rather have that 👍
2. highlighting idempotency: highlightCodeBlocks merges the entire existing pre.class as a user class, so running it twice on the same tree produces shiki . shiki, which then leaks back out as ::pre{.shiki} regardless of the strip guard

[hendrikheil]

Actually, I'll just fix the code issue in nuxt-studio directly, no need to do it in comark imo. I've also dropped the rehighlighting path. Let's keep this PR focused on the primary issue we're encountering in nuxt-studio right now...

[farnabaz]

Thanks