I upgraded a test database not too long ago to 18.6 (to do proper testing before we start upgrading productions) and got a feedback from dev that one of the queries got errors.
The query was quite long and strange and it used a few levels of views so I found it really difficult to analyze. The errors were quite strange as well:
select … where rownum<=2000 got ORA-1722 invalid number
select … where rownum<=100 worked
Also, the query returned a large concatenation of strings and columns, while I changed this to “select *” I got ORA-3113 and the alert log showed ORA-7445 [_intel_fast_memcmp]
I thought it’s an optimizer thing so I tried to compare plans and use hints to work around the problem. It fixed some of the cases, but not all of them.
So I opened an SR to see what support has to say. I was quite impressed (compared to other experiences I had in such cases) when they managed to reproduce the problem and found a related bug. The bug talks about kafcpy module and rowsets, so the fix was to set “_rowsets_enabled” to false. Unfortunately, the bug is marked internal and there is no information at all about it (bug number is 26609158)
Then I searched and found Mike Dietrich’s post about rowsets that might be relevant.
In the meantime, we are checking the workaround and waiting for a patch (that was requested for my specific environment).