Invalid length parameter passed to the LEFT or SUBSTRING function UNION ALL - tsql

Question

I have two queries that use the SUBSTRING function within a CASE statement like so:
CASE
WHEN Answer.ChoiceTitle = 'Neither Likely or Unlikely'
THEN 'Neither Likely nor Unlikely'
WHEN Answer.ChoiceTitle LIKE '[1-5]%'
THEN SUBSTRING(Answer.ChoiceTitle, 3, LEN(Answer.ChoiceTitle) - 2)
ELSE Answer.ChoiceTitle
END AS Recommendation
Both queries run perfectly fine when run separately but when I try to combine both result sets with a UNION ALL I get the error message:
Invalid length parameter passed to the LEFT or SUBSTRING function
Whilst trying to figure out why this error is occurring I added the below to each statement and the UNION ALL now works perfectly fine.
MIN(LEN(Answer.ChoiceTitle)) OVER() AS MinLength
Why would I be getting this error?
Execution Plans
Planned Execution Plan with UNION ALL - https://www.brentozar.com/pastetheplan/?id=rksFnuLS-
Actual Execution Plan of first statement - https://www.brentozar.com/pastetheplan/?id=r1Z-pO8HW
Actual Execution Plan of second statement - https://www.brentozar.com/pastetheplan/?id=rkCTh_IBb

Answer 1

This is most likely causing your error: LEN(Answer.ChoiceTitle) - 2
When that evaluates to less than 0, it will throw an error.
Try this instead:
CASE
WHEN Answer.ChoiceTitle = 'Neither Likely or Unlikely'
THEN 'Neither Likely nor Unlikely'
WHEN Answer.ChoiceTitle LIKE '[1-5]%' and LEN(Answer.ChoiceTitle) > 2
THEN SUBSTRING(Answer.ChoiceTitle, 3, LEN(Answer.ChoiceTitle) - 2)
ELSE Answer.ChoiceTitle
END AS Recommendation
Since you are just getting rid of the first two characters, you could use stuff() instead like so:
CASE
WHEN Answer.ChoiceTitle = 'Neither Likely or Unlikely'
THEN 'Neither Likely nor Unlikely'
WHEN Answer.ChoiceTitle LIKE '[1-5]%'
THEN stuff(Answer.ChoiceTitle,1,2,'')
ELSE Answer.ChoiceTitle
END AS Recommendation
This will give you an empty string if the length is less than 3, otherwise it will remove the first two characters of Answer.ChoiceTitle.
As to why the combined query with union all throws an error when the others run alone do not:
I'm seeing this difference in the execution plans:
Hash Match > (Question & Survey nested loop) & (Compute Scalar > Answer) {Bottom right of execution plan without error}
vs
Hash Match > (Bitmap > Parallelism > Question) & (Compute Scalar > Answer) {Bottom right of execution plan with error}
the nested loop version may be filtering rows that cause the error prior to the hash match, thus avoiding the error.
It is possible that using option (maxdop 1) to prevent parallelism would avoid the error as well (confirmed) on the query that is currently throwing it
this just comes down to when the scalar function is being evaluated for the rows in the answer table, pre or post filtering out the rows you want to run the expression on.
The cost is higher in the union all version, and it exceeds the cost threshold for parallelism, which is why you wouldn't see the same error when run alone where it does not go parallel (specifically parallel in the same way) with the lower cost when run alone.
So basically the parallel plan is running the substring() sooner than your other plans, before the rows that throw errors are filtered out.