On 26.08.2014 15:09, Eric Blake wrote:
On 08/22/2014 10:31 AM, Max Reitz wrote:
039, 060 and 061 all create images with referenced clusters having a
refcount of 0. Because previous commits changed handling of such errors,
these tests now have a different output. Fix it.
Furthermore, 060 created a refblock with a refcount greater than one
which now results in having to rebuild the refcount structure as well.
For bisection purposes, should these test tweaks be made in the same
patch that causes the improved behavior?
We could do that, but patch 6 changed the behavior to "Report an error
instead of trying to fix it", 7 then changed it to "Fix it, but leave
leaks behind" and finally 8 changed it to "Fix it and fix the leaks". It
would be possible to add these changes to each of these patches, but it
would look pretty strange in my opinion.
Benoît commented on this in patch 5 in v1 (which is patch 6 here). I
replied to this in the cover letter of v2: I don't think it really
breaks bisects because the tests were kind of faulty before this series
anyway. They assumed qcow2's repair function would repair blocks with a
refcount of 0 by simply increasing the refcount through the standard
qcow2 functions, which, as this patch 10 shows, will not always work as
intended.
Max
