I was reading through the mount options and the man paragraph on inode-cache seemed pretty ominous. My question is: what determines whether a btrfs-convert, created volume has 32 bit inode numbers? In order of assumed likelihood: Is it the compiled architecture of the btrfs-progs package? Is it the architecture of the kernel? Is it the inode width of the ancestor filesystem? I'm likely to see a fair amount of churn (for a desktop) on the contents of volumes once converted with a pretty long term support period. I'd like to head off this concern before implementation.
