It's more of a mismatch between the virtual tree and what's in the DOM tree because the comments were injected outside of preact-render-to-string somehow. Not sure if that makes this a bug or more of a feature request to allow skipping comment nodes that weren't created by Preact. Conceptually, the SSR'ed tree must match the virtual tree 1:1 and not be messed with.

Made a PR that would address this over here: #3771

That said I'm not sure if we should merge it as the use case seems to be a bit niche. I'd love to know more about where the comments are coming from in your scenario. It seems to be related to how WordPress' editor's HTML is generated. Can you share more information on that?

Thanks for the explanation, and thanks for looking into this, Marvin. If this is actually the expected behavior in Preact, then we may need to look for an alternative solution.

For context, this is not specifically related to WordPress, but to the partial hydration technique we are trying to implement, which:

• It traverses the DOM to generate a static virtual DOM.
• While doing so, it replaces the interactive parts of that virtual DOM with client components.

The main benefit of that technique (versus the Astro-like island approach) is that interisland communication works out of the box (context, suspense, error boundaries) and it can do client-side transitions. In a way, it's similar to RSC, but lightweight and built on top of regular HTML. The initial HTML can be generated by any technology, not preact-render-to-string. In our case, the HTML is generated by WordPress, but it could be any other platform/language. And that HTML can contain anything, including HTML comments.

Actually, WordPress discourages using HTML comments as they are usually removed by optimization plugins or CDNs (Cloudflare and the like), but WordPress developers still add them sometimes.

The ideal solution would be to add the HTML comments to the static virtual DOM and have a 1:1 match. But Preact is not prepared to deal with comment vnodes (the diffing fails).

Now that I know that this is the expected behavior and not a bug, we'll try to deal with it ourselves. Maybe removing the comments during the DOM -> vDOM phase.

@luisherranz Don't get me wrong, I'm happy to add support for this if there is a strong reason to do so. And it might look like I need that feature for streaming hydration anyway as the pending areas need to be marked with HTML comments.

the pending areas need to be marked with HTML comments

Oh, interesting (and exciting) 🙂

What's your idea about that part? (for example, is the idea to keep vnode representations of those HTML comments in the vdom while the rest of the HTML is streamed?)

I'm going to close this issue as this works as expected, and we will manually remove the HTML comments during our DOM -> vDOM step. I'll keep an eye on what you release for streaming in case we can adopt it in the future!