I do see your points regarding the existence of use cases for this
feature, and I agree that at worst, the implementation of this feature
would provide a way to greatly simplify query design and at best provide
a whole new method of obtaining decision supporting data from a
However I am strongly in disagreement with your fourth point, I.e., that
users will only become aware of it once it has been implemented. This
sort of mentality is what gave us the sad case of late 90s HTML in which
browser vendors assumed that they could use the "if you build it they
will come" argument for feature extension of the HTML spec. That is a
debacle we are still suffering the effects of. Let us not do the same to
SQL and implement SKYLINE on our own, only to have other DBMS vendors
implement it in different ways and then finally when the SQL standard
includes it they try to make some kind of average approximation of the
implementations resulting in *none* of the DBs being compliant. Then
we'll be between the rock of breaking backwards compatibility and the
hard place of unwarranted standards non-compliance.
While Josh did point out that being in the leading group as far as
implementing new functionality goes, I feel that it has to be weighed
against the need to not strike out too aggressively, potentially
isolating ourselves with excessive non-standard syntax or behavior.
While I am convinced there is a strong use case for this functionality
and we should definitely start looking at it, I don't see why we should
be in a rush to get it into core. People have survived without it up to
now, I don't think our userbase will suffer if it is implemented 6
months after <foo commercial DB> implements it, at least, not as much as
it will suffer if we start drifting away from standards compliance.
Just my 2 rupees. :)
Few things from our side:
1. 'Skyline Of' is a new operator proposed in ICDE 2003, one of the
topmost conferences of Data Engineering. Skyline operation is a hot
area of research in query processing. Many of the database community
people do know about this operator, and it is fast catching the
2. The skyline operation is very useful in data analysis. Suppose, if
we have a cricket database, and we want to find the bowlers who have
taken maximum wickets in minimum overs, we can issue an easy-to-write
query using 'Skyline of' syntax as follows:
Select * from Player_Match Skyline Of overs_bowled min, wickets_taken max;
This query gives 25 interesting tuples (result set) out of 24750
tuples in 0.0509 seconds. The same result is obtained in 0.8228
seconds if the following equivalent nested-query is issued:
select * from Player_Match p1 where not exists ( select * from
Player_Match p2 where p2.overs_bowled <= p1.overs_bowled and
p2.wickets_taken >= p1.wickets_taken and (p2.overs_bowled <
p1.overs_bowled or p2.wickets_taken > p1.wickets_taken))
Note that the above time is the time elapsed between issuing a query
and obtaining the result set.
As can be seen, the above query looks pretty cumbersome to write and
is inefficient too. So, which query will the user prefer? As the
number of dimensions increases, writing a nested-query will become a
Btw, how can such a query be written using aggregate function syntax??
3. As far as optimizing the Skyline is concerned, it is still a
research problem since it requires estimating the cardinality of the
skyline result set.
4. Until and unless this operator is implemented in a popular database
system, how can a user ever get to know about it and hence appreciate
Btw, it was our B.Tech final year project, and not a term project :-)
On 3/8/07, *Tom Lane* <[EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>>
Shane Ambler <[EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>> writes:
> Tom Lane wrote:
>> Well, whether it's horrible or not is in the eye of the
>> this is certainly a non-standard syntax extension.
> Being non-standard should not be the only reason to reject a
No, but being non-standard is certainly an indicator that the feature
may not be of widespread interest --- if it were, the SQL committee
would've gotten around to including it; seems they've managed to
everything but the kitchen sink already. Add to that the complete
of any previous demand for the feature, and you have to wonder
> The fact that several
> different groups have been mentioned to be working on this
> indicate that it is worth considering.
It looks to me more like someone published a paper that caught the
attention of a few profs looking for term projects for their students.
Now maybe it really is the best idea since sliced bread and will
in the next SQL spec edition, but color me skeptical. It seems to me
to be a very narrow-usage extension, as opposed to (eg) multi-input
aggregates or WITH/RECURSIVE, which provide general mechanisms
to a multitude of problems. Now even so it would be fine if the
implementation were similarly narrow in scope, but the published
description of the patch mentions a large chunk of additional executor
mechanisms. If we're going to be adding as much code as that, I'd
to see a wider scope of usage for it.
Basically, this patch isn't sounding like it has a reasonable
bang-to-the-buck ratio ...
regards, tom lane
TIP 6: explain analyze is your friend
Pride sullies the noblest character
---------------------------(end of broadcast)---------------------------
TIP 4: Have you searched our list archives?