Support viewing static HTML

[johanneskoester]

Similar to static HTML served via huggingface spaces, the dataset viewer could render HTML file structures in iframes. This way, one could upload HTML reports generated by Snakemake , Datavzrd or Multiqc and give them a DOI. This would be extremely valuable for data science transparency and reproducibility.

[ArjunJagdale]

Similar to static HTML served via huggingface spaces, the dataset viewer could render HTML file structures in iframes. This way, one could upload HTML reports generated by Snakemake , Datavzrd or Multiqc and give them a DOI. This would be extremely valuable for data science transparency and reproducibility.

Hi! This would be a great addition for dataset transparency.

Would the intended behavior be something like:

• Rendering .html files in an iframe tab if detected in the repo?
• Restricting access to specific files like index.html or all .html files?
• Preventing external script loads for safety?

I’d love to explore a minimal version of this feature. Let me know if it would help to break it down into smaller pieces!

[julien-c]

could be quite cool!

[ArjunJagdale]

Hi again! 👋

I’d love to help implement this feature and wanted to quickly clarify a few things before starting:

1. Should we start by rendering just a single index.html file from the dataset root, or aim to support full HTML folder structures (e.g., with local CSS/JS/assets)?
2. Any security requirements we should follow when embedding the HTML (e.g., iframe sandboxing, CSP, block external scripts)?
3. Would we prefer it shown as a new tab in the dataset viewer, like “Report” or “HTML Preview”?

[ArjunJagdale]

Hi @julien-c, could I get a confirmation on this from you or someone from the team? It looks like a great addition—definitely worth trying out!

[severo]

I think we want to support text columns that contain HTML, ie: each cell is a self contained HTML file (no CSS/JS/asset files)

As far as I understand, the dataset-viewer API (ie, this repo) would "just" give the hint that a column contains HTML, and how to show it in the dataset viewer would be handle on the Hub (which is in a private repository).
An alternative is to change nothing on the API side, and try to detect HTML content on rendering.

@lhoestq do you have an idea of how to tag a specific text column to tell that it contains HTML (or something else, by the way, ie: JSON, Python code, etc)?

@cfahlgren1 for vis

[lhoestq]

we could do the same as JSON, i.e. just check in the front end if string looks like HTML when row is expanded and if so check if it parses into HTML (if its a safe size).

[ArjunJagdale]

Got it — so the focus is on inline HTML in text columns, like how JSON is handled. Just to confirm, showing full HTML reports isn’t a current goal, right?

[cfahlgren1]

Got it — so the focus is on inline HTML in text columns, like how JSON is handled. Just to confirm, showing full HTML reports isn’t a current goal, right?

I think it would be similar to json viewer. If a row is expanded and we detect it as html we show a button to open rendered html in modal? wdyt @severo

[severo]

yes, I think it's what we want. Potentially more complex (security), because we don't want to show the HTML code, but the rendered page in an iframe.

[ArjunJagdale]

From what I understand (correct me if i am wrong):

• The idea would be: if we detect a string cell that looks like HTML, we show a button to render it in a modal (via iframe), instead of showing the raw code.
• This would need careful handling (iframe sandboxing, blocking external scripts, etc.), but seems doable on the frontend side.

Is this something we’d want to move forward with as a feature in the dataset viewer?
I'd love to explore it (if it's within my scope), or contribute in any way if the work is already in progress — there's a lot to learn at HF either way!

[severo]

It will be done on the Hub's side only. So: nothing to do in this repo. I'm letting it open to keep track of the feature request until it's implemented on the Hub.