On 11/30/12 4:27 PM, Dave Chinner wrote:
> On Fri, Nov 30, 2012 at 10:08:46AM -0600, Eric Sandeen wrote:
>> On 11/30/12 10:06 AM, Christoph Hellwig wrote:
>>> On Thu, Nov 29, 2012 at 12:59:55PM -0600, Eric Sandeen wrote:
>>>> This will cause the $SCRATCH_DEV to be fscked if it was used in
>>>> the prior test. Without this I don't think it gets done unless
>>>> specifically requested by the test.
>>> This one looks good.
>> Hm now that I think of it perhaps I should remove the explicit
>> _check_scratch-es if they happen at the end of the run, just to
>> try to speed things up.
I'll send as another patch; I don't think there are really very
>>>> Also recreate lost+found/ in one test so that e2fsck doesn't
>>> This one I can't make any sense of. Care to send it separately
>>> with a good explanation?
>> Ok, sure.
>> Basically, test does an rm -rf of the scrach mnt, but e2fsck
>> thinks that a missing lost+found/ is cause for complaint and a
>> failure exit code, which then stops the tests :(
> Shouldn't e2fsck be fixed? i.e. if you have a corrupted filesystem
> and it's missing lost+found, how are you expected to create it? by
> mounting your corrupted filesystem and modifying it and potentially
> making the corruption worse?
No, e2fsck fixes it, but reports that as an exit error condition
even if nothing else is found.
>> (hum, now that I think about it, maybe a broken scratch device
>> shouldn't stop the test series, but should just log a test
>> failure? What do you think?)
> Stop it - we should be leaving a corpse that we can dissect to find
> out what went wrong. For a corrupted scratch filesystem, running
> another test will eat the slowly rotting corpse and leave nothing
> useful behind for diagnosing the failure...
True, in most cases you could re-run the test, but maybe not.
Ok, will leave that as-is.