Hi, Will,

Thanks for the report.  There was a couple of problem with the
ordering function that caused this problem you've observed.

- as you've observed, the values of s are all 1s, which causes the
ordering function attempting to order with other columns.  In this
process, the ordering flag "DESC" is not carried appropriately.  The
updated code attempts to keep the option "DESC" when the user has
specified "DESC" to all order by keys.  In your case, you have one
order by key and one specification of "DESC", so the "DESC" flag is
preserved for all sorting operation.  See if you agree with this
logical and let us know.

- there was another bug in the code that causes the sorting procedure
itself to only work for ASCENDING order.  This problem was hidden
because the test case intended to catch this was not thoroughly
examined until now -- thanks for bring this problem up to our attention.

A fix has been checked into the SVN repository as SVN revision 853.
Please give it a try when you get the chance.

Thanks again.

John

PS: you can access the source code following the instruction at
https://code.lbl.gov/scm/?group_id=44





On 4/17/16 6:25 PM, Will Glozer wrote:
> Hello,
> 
> When FastBit is asked to use descending sort on an aggregation column
> with 1 distinct value it attempts to sort by all columns and in doing
> so breaks the row ordering. Here's an example input CSV:
> 
> 11,43,0
> 11,43,1
> 13,43,0
> 13,43,1
> 17,43,0
> 17,43,1
> 19,43,1
> 19,97,1
> 
> Load: 
> 
> `ardea -d test -m "a:int,b:int,c:int" -t test.csv`
> 
> Query:
> 
> `ibis -d test -q "select a, b, sum(c) as s from test where 1=1 order
> by s desc limit 1000"`
> 
> 19, 43, 1
> 19, 43, 1
> 17, 43, 1
> 13, 43, 1
> 11, 97, 1
> 
> Thanks,
> Will
> 
> 
> 
> _______________________________________________
> FastBit-users mailing list
> [email protected]
> https://hpcrdm.lbl.gov/cgi-bin/mailman/listinfo/fastbit-users
> 
_______________________________________________
FastBit-users mailing list
[email protected]
https://hpcrdm.lbl.gov/cgi-bin/mailman/listinfo/fastbit-users

Reply via email to