bcache block interface ongoing support

That might be quite difficult because you can layer bcache at almost any point in the block hierarchy, and because its header is on the front, either rewriting of all data or careful movement of the layer above (LVM, partitions or whatever, which requires knowledge of that layer) is required. And then there are clever bastards like me with an LVM PV on bcache on md, and a VG layered across both bcached and non-bcached PVs, on two md arrays on the same physical disks, and lots of LVs underneath that... changing *that* to the one-lvm-cache-device-per-LV model used by dm-cache without screwing alignment completely or risking data loss seems likely to be terribly difficult.

No, I fear the same rule applies to this as to any fs: people rely on it, so you can't rip it out or randomly change the data format -- though for non-writeback bcache volumes it is probably more practical to change the cache device format, you don't want to require changes to the backing device: even writeback bcaches probably just need to write all the data back and then freely change the cache device: it would be nice if this didn't require downtime for those of us with rootfses on bcache, too.