Check Your Subroutines - We are delivering our latest release to internal test today. Had a code review yesterday. Many issues were found. We are fixing the highest priority probl...
The database error was that the code was trying to fit a large number into a NUMBER(9) column. Seems simple enough. I said we were probably selecting a column that was bigger than precision 9. We looked at the code. Nope. Source columns were also NUMBER(9).
Then I said we must be getting our insert and select order mixed up. This was a large insert statement. We pulled the column names and the values we were pulling into an Excel spreadsheet. Yep. Something got out of order in the select. Just a couple out of order can cause some pain.
A manager wanted to know what was going on. Why did this work in one environment and not another? Well it all depends on what type of data (and how big the numbers are) in the columns we got switched.
Further analysis of the rest of the SQL showed some more SQL statements out of whack.
It was easy enough. I inspected the catalog to figure out the sets of data I needed to choose from. There is no NVL function in this DB. So instead I had to use CASE expressions. I coded up the changes, compiled them into the database, and executed my function. The good news is I got all the primary data I needed. Unfortunately the copy of production data I had did not exercise all paths.
Okay. I could just clone some records in my own schema, modify them to simulate the second data set, and rock and roll. Unfortunately I don't have my own schema. I just share a schema with my whole team. We usually just make sure all objects are owned by a role. And we all have that role. That works until somebody creates an object and forgets to reassign ownership.
Wouldn't you know it? Somebody created the table I needed in our common schema. But they did not share ownership. I could not even select rows from the table. I needed to modify data in that table. What was my solution? We were running out of time, so I created a similar table with another name. Then I modified my function to use this other table. Hack? Yes. But it worked to test out my logic.