The urlpreview regexp is grabbing the closing parenthesis in markdown syntax and thus failing when attempting to preview. If the user posts a URL in the form of text

[text](https://link.com)

The closing parenthesis after .com gets pulled in and the preview fails.

Synapse's parser is a lot more feature complete, including oEmbed support, which can enhance url previews:

• URL previewer: https://github.com/matrix-org/synapse/blob/92828a7f958b2cb1925e2a64ed08c2efb6293787/synapse/media/url_previewer.py#L101
• oEmbed parser: https://github.com/matrix-org/synapse/blob/92828a7f958b2cb1925e2a64ed08c2efb6293787/synapse/media/oembed.py#L46

Caching could be a nice bonus, but that can be separated into a separate issue/enhancement.

Currently, all images are set to MAX_IMAGE_EMBED px wide, even if the image is much smaller.

We can detect the image's width/height and adjust it to show min(width or height, MAX_IMAGE_EMBED) for a better user experience.

This can be an optional external dependency: https://stackoverflow.com/questions/6444548/how-do-i-get-the-picture-size-with-pil

Preview json in urlpreview

Other possible ideas:

• RSS feeds
• XML

In my opinion the HTML that is returned by the bot renders in a way that takes up too much space in the chat. In a previous version of the plugin I hacked the following:

If we defer all hardcoded HTML to a simple templating engine users can decide how they want to render previews.

I might give this a try myself sometime.

(The example above was implemented using HTML tables. Hello 1997.)

Currently, urlpreview supports whitelisting certain urls. This way, users can be selective about outgoing requests for security reasons.

If interested I can help with a PR since this is mostly config side.

Originally posted by @ajkessel in #9 (comment)

Thanks for fixing this so quickly. Is it necessary to limit URLs to https? In my copy, I added a ? after s in the regexp.

7009c54#commitcomment-125372320

@peterrus

Hey @coffeebank can you give some context around the choice to not fetch data through the Matrix API anymore? For me that was exactly what made the plugin interesting so I might want to start maintaining a fork but maybe there are some caveats I am not yet aware off.

Thanks!

Currently, if a preview has been generated, and the user edits the message that triggered the preview, urlpreview sends the preview again. Ideally, urlpreview would never send the same preview multiple times (this would be useful even when a message is not edited -- for example, if someone responds to a message with the same link for some reason). An alternative would be to not generate previews on edited messages, or not generate previews unless the hyperlink in the message has been edited.

When someone embeds an URL in a code block (backtick or triple backtick) we probably shouldn't try to fetch a preview because that URL is there for different purposes than a regular URL that is posted to a channel.

I tried to do this with a regex, but my regex wizardry doesn't cut it unfortunately.