sql pivot function with a defined variable in the IN statement - sql

Question

Can anyone tell me how I can make this work? I either get "ORA-56901: non-constant expression is not allowed for pivot|unpivot values" or I get "ORA-00917: missing comma" depending on whether I put &mb1 in quotes or not in the IN statement. Thank you!
define mb1= ADD_MONTHS(TRUNC(SYSDATE,'MM'),-1);
select *
from
( select COLLECTOR
,Month
,low_activity_days
from dwh_prod.low_activity_days_collect_t) src
pivot
(
sum(low_activity_days)
for month in ('&mb1')
) piv;

Answer 1

You could set your substitution to a fixed value instead, so it isn't trying to to a calculation inside the pivot clause. As you're working with dates you can generate a date literal:
set termout off
column x_mb1 new_value mb1
select 'date ''' || TO_CHAR(ADD_MONTHS(TRUNC(SYSDATE,'MM'),-1),
'YYYY-MM-DD') || '''' as x_mb1
from dual;
set termout on
select * from (
select COLLECTOR ,Month ,low_activity_days
from dwh_prod.low_activity_days_collect_t
) src
pivot ( sum(low_activity_days) for month in (&mb1) ) piv;
The first query - with the output hidden by termout generates a string:
select 'date ''' || TO_CHAR(ADD_MONTHS(TRUNC(SYSDATE,'MM'),-1),
'YYYY-MM-DD') || '''' as x_mb1
from dual;
X_MB1
-----------------
date '2017-03-01'
The column ... new_value command then sets &mb to that. The substitution in your really query then becomes:
select * from (
select COLLECTOR ,Month ,low_activity_days
from dwh_prod.low_activity_days_collect_t
) src
pivot ( sum(low_activity_days) for month in (date '2017-03-01') ) piv
The problem with your original code is that you're ending up with a reference to sysdate inside the substituted pivot clause, which throws the ORA-56901 exception (since sysdate isn't constant). You can use any fixed value in the pivot, so you could do:
define mb1=''01-APR-17''
... for month in (&mb1) ) piv;
with the qoutes around the string as part of the definition, or
define mb1='01-APR-17'
... for month in ('&mb1') ) piv;
without, or other variations. But because those end up as string you're relying on implicit date conversion, and relying on your NLS settings; with the date literal I'm generating that isn't an issue. If you really wanted to use that date format you could still do:
define mb1='01-APR-17'
... for month in (to_date('&mb1', 'DD-MON-RR')) ) piv;
or select the date in that specific format with the new_values method:
column x_mb1 new_value mb1
select to_char(add_months(trunc(sysdate, 'MM'), -1), 'DD-MON-RR') as x_mb1 from dual;
... for month in (to_date('&mb1', 'DD-MON-RR')) ) piv;
Using the date literal format still seems simpler to me though.
I'm not sure why pivoting a single value is particularly useful though. Maybe your real query is doing more, but simple aggregation and a filter look more appropriate here:
select collector, sum(low_activity_days)
from low_activity_days_collect_t
where month = add_months(trunc(sysdate, 'MM'), -1)
group by collector;
If you want a rolling 12-month summary you don't need to build the pivot quite that dynamically; you can generate a month-offset number based on the current date, which will give you a fixed range of numeric values you can pivot on (rather than dates), something like:
select * from (
select collector, low_activity_days,
months_between(trunc(sysdate, 'MM'), month) as month_offset
from low_activity_days_collect_t
where month >= add_months(trunc(sysdate, 'MM'), -13)
and month < trunc(sysdate, 'MM')
) src
pivot (sum(low_activity_days) as months_ago
for month_offset in (1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12));
which will give columns collector, 1_months_ago, 2_months_ago etc.
What you can't easily do is make the column headings represent the actual month (e.g. 2017-04, 2017-03, ...), but you don't seem to have been trying to do that anyway; and unless you alias the pivoted column terms the ones generated in this example have to be treated as quoted identifiers (because they start with numbers).
If you do want month headers you could generate those with some more substitution variable shenanigans, with something like this before the query:
set termout off
column x_m1 new_value m1
column x_m2 new_value m2
...
column x_m12 new_value m12
select to_char(add_months(trunc(sysdate, 'MM'), -1), 'YYYY-MM') as x_m1,
to_char(add_months(trunc(sysdate, 'MM'), -2), 'YYYY-MM') as x_m2,
...
to_char(add_months(trunc(sysdate, 'MM'), -12), 'YYYY-MM') as x_m12
from dual;
column 1_months_ago heading &m1
column 2_months_ago heading &m2
...
column 12_months_ago heading &m12
set termout on
You might find it easier to use a proper reporting tool to query and format/display the results for you though.