Alas, granularity very quickly turns into users clicking through piles of crap without thinking about it. With great power comes great user error.
Given the amazing strides in usability we've seen in nearly every other field it baffles me why everyone isn't onboard with the fact that we can take the learnings from elsewhere and bring them to the domain of permissions.
Permissions are almost always hierarchical and grouped into classifications that make it easier to present the user with fewer more meaningful choices than asking the user to approve whether an app can see each contact on their phone one-by-one.
I'm honestly a bit cynical (puts on tinfoil hat) that marketers have held us back here since a lack of granular permissions aligns quite well with their effort to grab as much personal data as possible.
Consider the permissions for the lowly keyboard app...
A proper understanding of fine-grained permissions basically requires a working knowledge of how that permission might be or has in the past been abused.
And ultimately, fine-grained permissions are probably answering the wrong questions. The user expresses some basic trust via the initial app installation; what permissions ultimately help with is deciding whether or not to keep trusting the developer. If the app ask for lots of unexpected stuff, it's probably malware and should be uninstalled. If the permissions seem reasonable, the app is probably fine, and the user just wants to delegate responsibility to the app to do what it needs to do to get shit done.
It's really /all/ about trust. If you can't trust a random app, installation is a high-friction event. Check the stars, number of users, read a bunch of recent reviews, carefully go through permissions providing access for exactly what's needed. If you /can/ trust a random app, you can just install it, use it to read the fscking QR code and go on with your day. The need for trust is why we've ended up with centralized app stores with stringent content policies, and all the false positives that come along with it.