Cache size accounting on ZFS Datasets with compression not working about ccache

[jrosdahl]

@trofi: This bug report essentially implies that st->st_blocks can't be trusted, at least not on ZFS, so it's in conflict with the proposal in #442. Do you have any insights or opinions?

[trofi]

Yeah, storage becomes more and more complicated.

And to add on top of it even when block size is reported it does not necessarily mean on-disk storage if the deduplication is at play and is able to share data across files within ccache cache directory. Some tools use slower fiemap() syscall to find out the physical layout on disk for a particular file.

Perhaps it's a good time for ccache to give up on blocks accounting and use the logical size with added likely_size_on_disk() like windows case does?

[jrosdahl]

Perhaps it's a good time for ccache to give up on blocks accounting and use the logical size with added likely_size_on_disk() like windows case does?

Yes, that's the only reasonable solution I can think of as well.