T-SQL grouping by either of two columns - sql-server

Question

I have a table where I hold chat history as follows:
Id From To Text Hour
=================================================
1 A B Msg_A_B1 00:01
2 A B Msg_A_B2 00:02
3 B A Msg_B_A1 00:03
4 A B Msg_A_B3 00:05
5 C A Msg_C_A1 00:11
6 A C Msg_A_C1 00:12
7 C A Msg_C_A2 00:14
8 D B Msg_D_B1 00:17
I want to create a chat header list from data for a spesific user. The rules are
I want to get the start (first) "Hour" of each chat
and last message of chat for a spesific user
ordered by "Hour" ascending
For example if the user is "A" I want to get
Correspondant Text Hour
=======================================
B Msg_A_B3 00:01
C Msg_C_A2 00:11
Or for user "B" :
Correspondant Text Hour
=======================================
A Msg_A_B3 00:01
D Msg_D_B1 00:17
I can possibly do it by using Temporary tables, but I am seeking a simpler and faster solution.
This information might lead to the use of Stored Procedures, but a proper use of Views is also accepted.

Answer 1

What you are missing is a grouping column to mark a chat between A and B as "belonging together" without looking if A or B is the From or the To.
It is your table design, which makes things difficult. Below my suggestion I will add some hints how this might be done better:
Your mockup to simulate your issue:
DECLARE #mockupTable TABLE(Id INT,[From] VARCHAR(100),[To] VARCHAR(100),[Text] VARCHAR(100),[Hour] TIME(0))
INSERT INTO #mockupTable VALUES
(1,'A','B','Msg_A_B1','00:01')
,(2,'A','B','Msg_A_B2','00:02')
,(3,'B','A','Msg_B_A1','00:03')
,(4,'A','B','Msg_A_B3','00:05')
,(5,'C','A','Msg_C_A1','00:11')
,(6,'A','C','Msg_A_C1','00:12')
,(7,'C','A','Msg_C_A2','00:14')
,(8,'D','B','Msg_D_B1','00:17');
--The query
WITH cte AS
(
SELECT t.*
,CONCAT(CASE WHEN t.[From]>t.[To] THEN t.[To] ELSE t.[From] END,'-',CASE WHEN t.[From]>t.[To] THEN t.[From] ELSE t.[To] END) AS ChatID
FROM #mockupTable t
)
,FindFirstAndLast AS
(
SELECT cte1.ChatID
,(SELECT TOP 1 Id FROM cte cte2 WHERE cte2.ChatID=cte1.ChatID ORDER BY cte2.[Hour] ASC) AS FirstId
,(SELECT TOP 1 Id FROM cte cte2 WHERE cte2.ChatID=cte1.ChatID ORDER BY cte2.[Hour] DESC) AS LastId
FROM cte cte1
GROUP BY cte1.ChatID
)
SELECT fal.ChatID
,tFirst.[From] AS FirstFrom
,tFirst.[To] AS FirstTo
,tFirst.[Hour] AS FirstHour
,tLast.[From] AS LastFrom
,tLast.[To] AS LastTo
,tLast.[Text] AS LastText
FROM FindFirstAndLast fal
INNER JOIN #mockupTable tFirst ON fal.FirstId=tFirst.Id
INNER JOIN #mockupTable tLast ON fal.LastId=tLast.Id;
The idea in short:
The first CTE will create a ChatID by concatenating the From and the To in a sorted way. Doing so a message from A to B will get the same ChatID as a message from B to A.
The second CTE will use a correlated sub-query to find the first and the last message id, grouped for the previously computed ChatID.
The final SELECT will use these message ids to join the appropriate rows.
The result is coming with everything you need. It's on you, to put it in the format needed:
+--------+-----------+---------+-----------+----------+--------+----------+
| ChatID | FirstFrom | FirstTo | FirstHour | LastFrom | LastTo | LastText |
+--------+-----------+---------+-----------+----------+--------+----------+
| A-B | A | B | 00:01:00 | A | B | Msg_A_B3 |
+--------+-----------+---------+-----------+----------+--------+----------+
| A-C | C | A | 00:11:00 | C | A | Msg_C_A2 |
+--------+-----------+---------+-----------+----------+--------+----------+
| B-D | D | B | 00:17:00 | D | B | Msg_D_B1 |
+--------+-----------+---------+-----------+----------+--------+----------+
Some ideas about the desing
I would use
one table Person for your chatting persons.
a second table Chat for a chat with a ChatID.
one m:n mapping table ChattingPerson with JoinTime, a ChatID and a PersonID, both as FKs. Here you can set timestamps like LastAction or mark the status (active, has left, ...)
one more table Message for the messages with time, text, and ChatPersonID as FK.
Your advantages
The opener can explicitly invite more persons (or limit it to one for a person2person chat), or just wait for participants.
Starting a chat creates the row in the Chat table, the first row in the ChattingPerson table to mark the opener, and eventually a first message row.
Following messages add - if not existing yet - a row to the ChatPerson (with a new participant) and the message row.
The ID to the ChatPerson-table will give you the ChatID and the PersonID.
You can filter per chat and/or by person.
There can be separate chats between A and B over the time
You can control the type of chat with a PersonCount-Constraint
You can enforce, that a new ChatPerson can only be added by the opener
You can create certain chat types (like "person2person") with a template
Happy Coding :-)

Answer 2

Let's do it Creating a view:
First let's load the data in table t1:
create table t1 (Id int,[From] varchar(10),[To] varchar(10),Text varchar(100),Hour time(0))
insert into t1 values (1,'A','B','Msg_A_B1','00:01')
insert into t1 values (2,'A','B','Msg_A_B2','00:02')
insert into t1 values (3,'B','A','Msg_B_A1','00:03')
insert into t1 values (4,'A','B','Msg_A_B3','00:05')
insert into t1 values (5,'C','A','Msg_C_A1','00:11')
insert into t1 values (6,'A','C','Msg_A_C1','00:12')
insert into t1 values (7,'C','A','Msg_C_A2','00:14')
insert into t1 values (8,'D','B','Msg_D_B1','00:17')
Then let's create the view
create view vChats
as
with cte as (
select left([text],len([text])-1) as chat,
t.*
from t1 as t
),
cte2 as (
select chat,
min(hour) as minHour,
max(text) as maxText
from cte
group by chat
),
cte3 as (select distinct [From] as [User]
from t1
UNION
select distinct [To] as [User]
from t1
)
select c3.[User],
t.[To] as Correspondant,
c.maxText as [Text],
c.minHour as [Hour]
from cte2 as c
inner join t1 as t ON c.maxText = t.[Text]
inner join cte3 as c3 ON c3.[User] = t.[From]
UNION
select c3.[User],
t.[From] as Correspondant,
c.maxText as [Text],
c.minHour as [Hour]
from cte2 as c
inner join t1 as t ON c.maxText = t.[Text]
inner join cte3 as c3 ON c3.[User] = t.[To]
After that you can use to get all the communications for each user like this:
select *
from vChats
where [User] = 'A'