On 02/06/14 17:37, Vojtech Szocs wrote:
----- Original Message -----
> From: "Alexander Wels" <awels(a)redhat.com>
> To: "Lior Vernia" <lvernia(a)redhat.com>
> Cc: devel(a)ovirt.org
> Sent: Monday, June 2, 2014 3:35:51 PM
> Subject: Re: [ovirt-devel] Sortable Ui Columns
> On Monday, June 02, 2014 11:27:26 AM Lior Vernia wrote:
>> After discussing this with Alex on another thread, I just pushed this:
>> If it's approved, then it'll take care of the secondary criterion for
>> you, so you only need to pass the criterion that actually interests you.
>> Should simplify our work in the next couple of weeks.
> Okay, I think we need to get everyone on the same page with this, right now
> have several solutions to the problem, and I think we should pick one and go
> with that. Right now we have the following solutions available:
> 1. Attempt to keep the current order in case of 'duplication' in the
> comparator. So the fallback is to keep the current order. This is implemented
> in Liors patch .
> 2. Use the hash code as a fallback in case of 'duplication' in the
> Due to hashcode rules this should result in the 'natural' order in case of
> duplicates. This is implemented in my patch .
> 3. Do ad-hoc fallback mechanism which can specialize to do the proper thing
> based on domain knowledge. It looks like this is what Anmol is doing in his
> gluster sorting patch 
> 4. Replace the SortedSet with a List but lose the ability to automatically
> sort when calling getItems().add(X).
> To me it seems that 3 and 4 are off the table as we want to keep the ability
> automatically sort. And doing ad hoc solutions for every single sorting
> is going to take a lot of time and is going to lead to maintainability issues
> down the road. That leaves 1 and 2 on which I would like to have a
> so we can pick the appropriate method and go forward with that.
I just reviewed Lior's patch and posted some comments:
Initially, I thought that computing item positions based on the order
in which they are returned by Collection's iterator might be unreliable,
now that I think more about it, I think this approach is something we
In other words, in edge case when comparing two items with same logical
property (i.e. two Data Centers with same name), we should honor the
ordering of items within the original Collection provided by the server.
Otherwise, comparing hashCode might change their order in such edge
case. So actually it seems better to preserve Collection's iteration
order rather than preserving natural order in such edge case.
What do you think guys?
Preserving the previous ordering will enable users to sort tabs (that
are sorted client-side...) according to secondary criteria.
Say, for example, I want to have networks grouped by DC, and within each
group ordered by name. No problem, I'll just click on the name column
header first, then click on the DC column header. Since the DC sorting
will preserve the previous ordering for identical DC names, the networks
will be sub-ordered by name.
I think this is a very cool feature for virtually zero cost.
Sub-ordering by hashcode sounds a little random to me. It isn't
consistent with any human-comprehensible logic, it is only consistent
with Java's default ordering. So if I'm not mistaken, it will only
produce predictable results in case a collection isn't explicitly sorted
and its items don't implement Comparable. In which case, it will
converge to the same result as the "stable" ordering (because ordering
will fall back to the original ordering, which was based on hashcode).
So it seems to me like the "stable" sub-ordering produces results that
are always at least as predictable as hashcode sub-ordering, and in most
cases more predictable.
>  http://gerrit.ovirt.org/#/c/28268/
>  http://gerrit.ovirt.org/#/c/28225/
>  http://gerrit.ovirt.org/#/c/28233/
>> On 31/05/14 15:35, Anmol Babu wrote:
>>> Thanks a lot Alexander.Your idea of having a second criteria based
>>> just in case of comparing same values(compare returning 0) looks good to
>>> me and now, I have also done the same in my patch set
have added you as a reviewer as
>>> well. Thanks,
>>> ----- Original Message -----
>>> From: "Alexander Wels" <awels(a)redhat.com>
>>> To: devel(a)ovirt.org
>>> Cc: "Anmol Babu" <anbabu(a)redhat.com>
>>> Sent: Friday, May 30, 2014 5:55:20 PM
>>> Subject: Re: [ovirt-devel] Sortable Ui Columns
>>> This is due to the fact that the sorting is done by a SortedSet instead
>>> a list in the SortedListModel. To fix this we have to do one of two
>>> 1. Change the sorting to use a list of some sort, but apparently that is
>>> not much of an option as people want automatic sorting when they add new
>>> items to the collection.
>>> 2. Make sure that the comparator never returns 0 for two entities that
>>> really not the same. The reason you are seeing the issue is because you
>>> using a different comparator that does return 0 on whatever field you are
>>> comparing against. I had the exact same issue and I solved it by using a
>>> compound comparator. Basically what I did was create a comparator that
>>> contains the field I am trying to compare on and if that returns 0 then I
>>> use a different comparator that is guaranteed to not return 0 for the
>>> I have a patch that implements sorting for all the data center main tab
>>> sub tabs here  that demonstrates how I solved the issue. This patch
>>> hasn't been reviewed yet, and my solution might get rejected, but it
>>> work and doesn't make your entries disappear when you sort.
>>>  http://gerrit.ovirt.org/#/c/28225/
>>> On Friday, May 30, 2014 02:32:17 AM Anmol Babu wrote:
>>>> While applying client-side sort using the sorting infra, to the
>>>> column of the "Volumes" sub tab "Bricks", I had 2
Bricks with same
>>>> name.So,when I sorted it, it removed one of the bricks that had the same
>>>> server name. I found that this issue occurs when the sort values
>>>> are same(i.e, comparator's compare returns 0). Regards,
>>>> Devel mailing list
>>> Devel mailing list
> Devel mailing list