On Sun, Feb 02, 2014 at 10:08:06PM +0000, Filipe David Manana wrote:
> On Sun, Feb 2, 2014 at 9:57 PM, Dave Chinner <david@xxxxxxxxxxxxx> wrote:
> > On Sat, Feb 01, 2014 at 02:05:32AM +0000, Filipe David Borba Manana wrote:
> >> This change adds some new tests for btrfs' incremental send feature.
> >> These are all related with inverting the parent-child relationship
> >> of directories, and cover the cases:
> >> * when the new parent didn't get renamed (just moved)
> >> * when a child file of the former parent gets renamed too
> >> These new cases are fixed by the following btrfs linux kernel patches:
> >> * "Btrfs: more send support for parent/child dir relationship inversion"
> >> * "Btrfs: fix send dealing with file renames and directory moves"
> >> Signed-off-by: Filipe David Borba Manana <fdmanana@xxxxxxxxx>
> > Rather than modifying 030 which will cause it to fail on kernels
> > where it previously passed, can you factor out the common code and
> > create a new test with the additional coverage?
> > i.e. the rule of thumb is that once a test is "done" we don't go
> > back and modify it in significant ways - we write a new unit test
> > that covers the new/extended functionality. Redundancy in unit tests
> > is not a bad thing...
> Right. The only reason I did this, instead of a new test file, is that
> because the former fix which btrfs/030 relates to is not yet in any
> kernel release. Given this fact, what do you think?
Ok, so if it already fails for everyone, then I think we'll be fine
to modify it like this. "done" is a flexible concept when it comes
to unit tests ;)