trwnh, 10 months ago

@evan hot take, why not just use the opengraph namespace/iris directly?

if the goal is to reduce duplication of data, i'd probably recommend the following

• og:title = as:name
• og:type = og:type (no way to map this one)
• og:url = as:url
• og:image = as:image

for optional ones:

• og:description = as:summary

all the other ones i'd just use ogp.me iris directly

evan, 10 months ago

@trwnh that's about what I'd expect as a mapping. Have you seen it used to create Page objects, for link previews?

trwnh, 10 months ago

@evan i have only ever seen link previews implemented out-of-band due to trust issues :/

evan, 10 months ago

@trwnh you mean, on the receiving servers?

trwnh, 10 months ago

@evan yes, and it creates a sort of "thundering herd" unintentional ddos for web sites on underpowered providers. there have been ideas on how to solve the trust issue, like randomly verifying some small subset and tracking reputation, or like deferring to a semi-centralized pre-trusted "link preview provider" that is shared between several instances, or even simply ignoring the issue and depending on people to report misleading links. (i prefer the provider approach)

evan, 10 months ago

@trwnh another option is accepting the sender-provided metadata to start, and then verifying on a schedule that recognizes that thundering herd problem -- say, a randomized delay between 1 and 300 seconds, with an exponential backoff. And I definitely think having a trusted shared caching server can help immensely.