(Yes, that means applications need to start being concious of what
fs
they are being run on, or at least the fedora configuration needs to do
that check for them)
Right, and it's concerning to me that Fedora is committing to btrfs by default before
important applications have become more enlightened about running on btrfs. If upstream
changes don't land in time for Fedora 33, we will be implicitly expecting users to be
aware of these pitfalls and leave them to implement manual workarounds. I'd imagine a
good bit of thought and work will have to go into creating, testing, and upstreaming those
patches, so I think it's very possible that an appreciable number of changes will not
land in time for Fedora 33.
For example with virtualization I'd think that the changes would need to happen around
the level of libvirt, and not to specific a front-end like GNOME boxes or virt-manager.
It's also probably not sufficient to just set nodatacow on the default VM image
directory as users may use a non-default directory for qcow2 images. Hence I don't
think these issues will always have trivial solutions.