https://docs.aws.amazon.com/AmazonS3/latest/userguide/access...
(I don't know anything about this personally, but since a lot of people are indicating an interest in this detail of the story, figured I'd try and surface that link better!)
[1] If they ever actually turn off path-style addressing, come find me and I'll PayPal you a dollar. I don't think it'll ever happen.
Also to try to avoid having to special-case any logic in terraform etc.
Say you're working on a family of sites for tradespeople like plumber.io, electrician.io, carpenter.io, etc. A fair number of people from India have "occupational surnames" like Miller, Contractor, Builder, Sheriff, etc. Suddenly one Mr. Dev Contractor registers a bucket "contractor-dev" and you have to special-case your bucket names in your terraform.
Other things I think we do better on:
* The account is the top-level thing we publish a cert for. Without knowing the bucket name you can't really do anything. With S3's global namespace, each bucket has a cert published which makes all buckets discovered as soon as they're created.
* Not default open to the world
* The R2-managed public bucket cname is shared and the URL for the bucket is random (i.e. just a UUID). Additionally, if you delete and recreate the bucket with the same name IIRC that random UUID is changed.
* We have a lot of sensible extensions like automatically creating a bucket on upload (granted not possible for S3 since buckets are global), setting per-object TTLs, handling unicode more gracefully (I think normalizing the key name is a saner choice with fewer foot guns even if there's some potential compatibility issues when you try to synchronize from a filesystem where you have two files with different forms but same normalized), etc etc etc.