Details
-
Bug
-
Status: Closed
-
Major
-
Resolution: Invalid
-
10.1.3.1
-
None
-
None
-
zOS with ibm15
Description
A number of tests that printout runtimestatistics fail on zOS with ibm15, because instead of a 'time' of 0, the time is some large number.
Tests that fail like this:
derbylang/derbylang.fail:lang/aggregateOptimization.sql
derbylang/derbylang.fail:lang/distinctElimination.sql
derbylang/derbylang.fail:lang/predicatePushdown.sql
derbylang/derbylang.fail:lang/predicatesIntoViews.sql
derbylang/derbylang.fail:lang/staleplans.sql
derbylang/derbylang.fail:lang/wisconsin.java
encryptionAll/encryption/storemats.fail:store/access.sql (in various encryption schemes)
This is the kind of diff I see:
---------------------
-
-
- Start: access jdk1.5.0 storemats:storemats 2006-04-13 09:46:01 ***
2912 del
< Execute Time = 0
2912a2912
> Execute Time = -3638779204078642344
2921 del
< next time (milliseconds) = 0
2921a2921
> next time (milliseconds) = 3638779204078642344
2934 del
< next time (milliseconds) = 0
2934a2934
> next time (milliseconds) = 3668750241784351640
2991 del
< Execute Time = 0
2991a2991
> Execute Time = -4015137258338621104
3000 del
< next time (milliseconds) = 0
3000a3000
> next time (milliseconds) = 4015137258338621104
....
------------------------------
- Start: access jdk1.5.0 storemats:storemats 2006-04-13 09:46:01 ***
-
As this problem occurs only with ibm15 and not with ibm14, I think somewhere in the runtime statistics printing code we may be using BigDecimal somehow, and it's not done in an encoding-safe way. But that is just a guess.
I'm leaving this one as Major, because it's easy to miss things when looking through test failures this way, but I can also understand that this would be considered minor, as apparently the paths taken by the optimizer are actually correct and it's only the printing out that has the large numbers. Yet again on the other hand I imagine it makes runtimestatistics less useful, with these disconcertingly large number of milliseconds.