On Mon, Jul 06, 2009 at 11:01:16PM -0500, Eric Sandeen wrote:
> Christoph Hellwig wrote:
> > Check that we correctly update the timestamps when writing to a file
> > through an mmap mapping. Currently fails for XFS due a VFS bug but
> > succeeds for many other filesystems.
> >
> >
> > Signed-off-by: Christoph Hellwig <hch@xxxxxx>
>
> Test seems mostly ok, but aren't you missing 215.out in the patch?
Yeah, looks like it. It's in my local tree, though :)
> > +sleep 1
>
> Is there any slight chance a sleep 1 won't catch it? Would sleep 2 do
> better to be sure?
A one second sleep should always increase the wall time by one second,
shouldn't it? We can do the sleep 2 to be double safe in case of
rounging issues, though.
> > + echo "FAIL: mtime not update after mapped write"
> > + status=1
> > +fi
> > +
> > +if [ "$ctime_diff" -eq "0" ]; then
> > + echo "FAIL: ctime not update after mapped write"
> > + status=1
> > +fi
>
> Should these just be _fail calls? Otherwise...
>
> > +# success, all done
> > +echo "*** done"
> > +rm -f $seq.full
> > +status=0
>
> don't you always exit w/ status 0 and remove the full output?
>
> I guess i'm not sure when to use _fail and when to let the golden output
> difference kill the test, I suppose it doesn't much matter, it should
> pass & fail correctly either way.
You're right, the status=1 is overrident currently and it only fails
due to the golden output difference. I'll respin it to just kill the
superflous status assignment.
|