Except that not all posts follow the new ID format. Not all new posts even follow it. It's now a mix of the two.
At one point, they've also chosen a web forum, a wiki lots of devs bought into (but which they later entirely blew away and replaced with Bing searches of their site), and probably a half-dozen attempts at official on-site docs.
When I saw they'd offloaded to StackOverflow, I thought that was an idea that had some promise, but the big question in my mind was how long until it reached the critical mass where it would consist of more out-of-date answers than current ones. And what they'd do when it does.
The answer, of course, is likely move fast and break something, and heaven help the people who had the idea of building something on top of their quicksand.
They seemed to have fixed it now, but for a while the search box would return results from the wiki, which had been removed, leading you on an endless tempting "the answer you want is just around the corner" unhelpful drugery.
I've updated the Graph API getting started guide to mention the fact that IDs are opaque and also mentioned it in our comments guide since comments are one of the few places where we have IDs that are made up of other keys.
Thanks for the feedback - it's good to hear about things we can fix.
It's not documented as a format (very little is) but it's been this way since at least 2011-ish.
If there is a better way to get this information it is also undocumented.
If the API has a usable purpose without providing that info, then I think I'd disagree with you - since you're reverse-engineering the implementation to get more data out.
Reverse engineering is OK, but implicitly carries the risk that stuff will change and break you without warning, in which case you can't really complain, in my opinion.
If the API's purpose can't be fulfilled without that info, then the API was probably broken as designed (you needed to parse the id to get the info, they just didn't document it).
Liking works, but unliking does not in cases where the post_id returned has the profile_id prefixed. So you have to remove the profile_id from the post_id to unlike.
The reason why developers have to reverse engineer is that the Facebook platform doesn't and did never work the way it is documented.
Here's a bugreport for that problem which hasn't been solved since last February! (The top comment here with a workaround for that issue was created by myself: https://developers.facebook.com/bugs/539328566085833?browse=...)
I've suffered from having designed bad/limited APIs before and watching as people dig into bits I didn't want them to dig into, limiting what I can do in the future.
Not sure on the best way to avoid it, but I guess it's a 'nice problem to have' since at least people are using your API.