Hi, Jochen,

Thanks for the patience in working with us through this set of 
intermittent errors.  The test cases you've provided were extremely 
helpful.  Since the test cases contains nearly 100 conditions in each 
query, it took me a while to sort through them and ping point the root 
cause of the problem.  It should be fixed now.  Please download the 
nightly snapshot <https://codeforge.lbl.gov/snapshots.php?group_id=44> 
and give it a try when you get the chance.

For everyone else who is working with version 1.1.1 released on July 
6th (which enables group-by with string values as sort keys).  If you 
are experiencing troubles, please try the latest nightly snapshot too. 
If you don't need the functionality of group-by on string values, you 
may go back to version 1.1.0 as well.

Thanks.

John

PS: Just in case you are curious about the cause of the problem, here 
is a bit more information.  Overall your where clauses are in the 
conjunctive form (i.e., joined with AND operators).  Within some 
terms, there are a few disjunctive conditions.  Some of the 
disjunctive conditions produce no hit.  In some cases, such as when 
the indexes are not yet available, the query conditions will be 
evaluated by examining the base data files.  The function that 
performs this task neglected to set the return bit vector object when 
there is no hit.  This ill-formed bit vector will cause other terms of 
the disjunctive conditions to be evaluated incorrectly.  Hence, you 
will some time see incorrect number of hits overall.

PS: In my rush to implement group-by functions on string values, I 
have introduced a couple of bugs that can cause the program to crash. 
  They have been identified while updating the built-in tests to check 
for the no-hit disjunctive conditions.  These errors affect FastBit 
version 1.1.1.

PPS: For those who haven't used the night snapshot before, the gforge 
installation appears to create the night snapshot at about 3AM Pacific 
Time (USA).  You will need to download the snapshot after 3AM July 17 
in order to get the latest code.


On 7/15/2009 10:54 AM, Jochen Schlosser wrote:
> Hi John,
> 
> I finally managed to reproduce my errors in a small test scenario. 
> Please take a look into runTests.sh, query.cpp and the README file. Both 
> errors which I reported yesterday can be reproduced using the runTest.sh 
> script.
> 
> I guess the source is rather easy and I hope you'll be able to locate 
> the error...
> 
> 
> Best regards,
> 
> Jochen
> 
> 
> P.S.
> Here is the output ot the TestSkript... I hope it corresponds to the one 
> on your system!
> ##########
>  Observe different results in subsequent queries
>   Run 1
>   >> Hits retrieved via Fastbit: 0
>   >> Hits extracted via getQualifiedInts: 0
>   Run 2
>   >> Hits retrieved via Fastbit: 3
>   >> Hits extracted via getQualifiedInts: 3
> 
>  Observe missing values from getQualifiedInts()
>   >> Hits retrieved via Fastbit: 20
>   >> Hits extracted via getQualifiedInts: 19
> ###########
> 
> 
> 
> 
_______________________________________________
FastBit-users mailing list
[email protected]
https://hpcrdm.lbl.gov/cgi-bin/mailman/listinfo/fastbit-users

Reply via email to