How to do this in SQL (PostgreSQL Window Function?) - sql

Question

I have a situation in SQL (PostgreSQL specifically) that I'm struggling with. The schema/model that I'm working with is not under my control and not something I'm able to alter, so I am trying to figure out the best way to deal with the cards I've been dealt.
First, the schema, simplified for this question, but essentially it's invoice (Type = T) and transaction (Type <> T) lines combined into the same table. There can and will be n-number of tranaction lines per invoice and n-number of invoices per client.
Id
Type
InvoiceNo
ClientId
100
I
100
1
99
X
0
1
98
S
0
1
97
T
0
1
96
I
99
1
95
X
0
1
94
S
0
1
What I ultimately would like to end up with is something like the below, with the Invoice (Type = I) records removed and the Transaction (Type <> T) records that fall after each Invoice record populated with it's corresponding InvoiceId value.
Id
Type
InvoiceNo
ClientId
99
X
100
1
98
S
100
1
97
T
100
1
95
X
99
1
94
S
99
1
So far, the closest I've been able to get, which isn't very close, is using the below SQL:
select
t1.Id,
t1.Type,
t2.InvoiceNo,
t1.ClientId
from table AS t1
join (select
Id,
InvoiceNo,
ClientId
from table
where type = 'I') as t2
on t1.ClientId = t2.ClientId
where t1.ClientId = t2.ClientId and t1.Id <= t2.Id and t1.Type <> 'I'
The result of that looks something like the below, which works fine for the first invoice per client and then creates extra transaction records for each invoice
Id
Type
InvoiceNo
ClientId
99
X
100
1
98
S
100
1
97
T
100
1
95
X
100
1
95
X
99
1
94
S
100
1
94
S
99
1
Any help or guidance is much appreciated!
** Updated with more complex example **
Source:
Id
Type
InvoiceNo
ClientId
1
X
0
1
2
I
97
1
3
S
0
2
4
X
0
2
5
S
0
1
6
I
98
2
7
S
0
1
8
X
0
1
9
I
99
1
10
T
0
1
11
S
0
1
12
X
0
1
13
I
100
1
Playing with the answer below, I came up with:
select * from (select t.*,
max(InvoiceNo) filter (where type = 'I') over (partition by clientid order by id DESC) as imputed_invoiceno
from t) as x
where Type <> 'I';
Which gets me close:
Id
Type
InvoiceNo
ClientId
imputed_invoiceno
12
X
0
1
100
11
S
0
1
100
10
T
0
1
100
8
X
0
1
99
7
S
0
1
99
5
S
0
1
99
1
X
0
1
99
4
X
0
2
98
3
S
0
2
98
Best case result:
Id
Type
InvoiceNo
ClientId
12
X
100
1
11
S
100
1
10
T
100
1
8
X
99
1
7
S
99
1
5
S
99
1
1
X
97
1
4
X
98
2
3
S
98
2

Answer 1

Based on your sample data, you can use a cumulative window function:
select t.*,
min(invoiceno) filter (where type = 'I') over (order by id desc) as imputed_invoiceno
from t;