问题
We are trying to create a SQL query in Sybase where we can concatenate multiple rows into one selected row.
Layout:
| Type | Skill |
----------------
| A | 1
| A | 2
| B | 1
ETC
I want the output to be like: A (1,2)
回答1:
Which specific type of Sybase database is this? if ASE, then you either have to use a loop approach (e.g. Cursor), or use a somewhat exotic trick with the UPDATE statement using local variables (too much to describe here quickly, but details are in my book 'tips, tricks and recipes for sybase ase'
回答2:
If using ASE 16 you may be able to roll your own function to simulate the list() function; see my responses in this thread: Implementing group_concat() in ASE
Pay attention to Edison's response about KBA 2260479 ... a bug when using table variables inside a transaction.
----------- per the suggestion to copy pertinent info from sap.com link ...
Referring to an example @ Percona example (sorry, not going to replicate that link, too; the examples attached to this post show the results of matching the percona examples anyway)
To mimic MySQL's GROUP_CONCAT() function in ASE we need to understand some issues/limitations/observations:
ASE does not allow aggregate/UDFs (eg, passing a result set as an argument to a function)
ASE does not allow defining a UDF with a table variable as an input parameter
ASE does not allow the creation of a #temp table inside a UDF
ASE 16 SP02+ DOES allow the creation/use of table variables inside a UDF
ASE DOES allow the exec() construct inside a UDF
the GROUP_CONCAT() argument consists of columns/strings to be appended, an 'order by' clause, and an optional SEPARATOR (see examples at the percona.com link - above); it's not too hard to see that the columns/strings + 'order by' clause are components of a SELECT query
we can simulate passing of a data set to the UDF by instead passing a SQL/SELECT query that represents the desired dataset; this SQL/SELECT query can then be executed via the exec() construct to provide the UDF with the desired data set
The gist of our UDF design:
NOTE: Due to our use of a table variable the following requires ASE 16.0 SP02+
1 - the UDF's input parameter - @sql varchar(XXX) - represents a SQL/SELECT statement supplied by the calling process
1a- @sql must be a complete stand-alone query (ie, you should be able to run the query on its own in a separate ASE session)
1b - the select/projection list of @sql matches the column(s) of the table variable (see next bullet)
1c - @sql includes any necessary 'group/order by' clauses (ie, the UDF will not perform any sort operations)
2 - the UDF creates a table variable with a column defined to hold the results of the @sql query
2a - the column(s) of the table variable must match, datatype wise, with the select/projection list of the @sql query
3 - the UDF populates the table variable via the exec() construct:
exec("insert @table_var " + @sql)
4 - the UDF uses a cursor to loop through the records in the table variable
4a - cursor does not have a 'order by' clause => assume rows are ordered, upon insert to the table variable, based on @sql
Some specifics about this particular UDF:
1 - I've created the UDF in the sybsystemprocs database and named it:
sp_f_group_concat
1a - the 'sp_' prefix means the UDF can be executed from within any database
1b - the 'f_' string allows me to quickly/visually see that this is a function and not a system stored proc
2 - the UDF has been created with the assumption that the incoming SQL/SELECT query will have a select/projection list consisting of a single varchar(100) column
2a - the calling process will need to perform any necessary datatype casting (to char) and any concatenation of columns/strings
2b - the @sql input parameter has been defined as varchar(1000), and the @separator has been defined as varchar(10) with a default of a single comma (',')
2c - the owner of the UDF will need to revise the varchar() lengths based on what they expect to handle in their environment
3 - since the UDF cannot be sure TF:7703 (allow accumulating of data, by row, into a single @variable) is enabled, and the UDF does not perform any sorting, we'll use a cursor to step through the records in our table variable
4 - it's not apparent from the percona.com examples how MySQL's GROUP_CONCAT() function handles appending NULL's with non-NULL values (eg, is the NULL ignored? is the NULL converted to the empty string ''? is the NULL converted to the string 'NULL'?); net result is that the UDF owner/user may need to revisit the UDF and/or @sql design if they find NULL's are not being handled as desired
Hmmmm, can't attach files to stackoverflow posts? ok, so cut-n-paste it is ... yuck, not quite what the source looks like ...
++++++++++++++ sp_f_group_concat.sql
- UDF DDL
use sybsystemprocs
go
if object_id('sp_f_group_concat') is not null
drop function sp_f_group_concat
go
create function sp_f_group_concat
(@sql varchar(1000)
,@separator varchar(10) = NULL
)
returns varchar(1000)
as
/*
sp_f_group_concat
ASE implementation of MySQL's GROUP_CONCAT() function.
See https://ideas.sap.com/D36082 for a discussion of this topic, along
with some examples (as attachments to one of Mark's comments)
Requirements/Assumptions
========================
- ASE version must support a) user defined functions and b) table variables
- @sql is a standalone query that generates a result set consisting of a single varchar column
- @sql includes an 'order by' clause if needed (ie, this function does not attempt to order the results generated by @sql)
History
=======
10/10/2016 Mark A. Parsons Initial coding
*/
set nocount on
declare @string varchar(100),
@string_list varchar(1000)
-- default delimiter to ',' if not supplied
select @separator = isnull(@separator,',')
-- create/populate @strings table
declare @strings table (string varchar(100))
exec("insert @strings " + @sql)
-- assume TF:7703 is not enabled, so can't use a single SELECT to append to a @variable
-- assume @sql has a 'order by' clause and that our cursor will pull from @strings in the same order
declare string_cur cursor
for
select string
from @strings
for read only
-- loop through @strings rows, appending individual strings to @string_list
open string_cur
fetch string_cur into @string
while @@sqlstatus = 0
begin
select @string_list = @string_list + case when @string_list is not NULL then @separator end + @string
fetch string_cur into @string
end
close string_cur
deallocate cursor string_cur
-- send concatenated list of strings back to calling process
return @string_list
go
grant execute on sp_f_group_concat to public
go
++++++++++++++ sp_f_group_concat.test1.sql
- percona.com examples for the group_c table
- first query shows using the default separator (a single comma)
- second query shows using a 3-char separator
use tempdb
go
set nocount on
go
/*
reproduction of the MySQL/GROUP_CONCAT() examples from:
https://www.percona.com/blog/2013/10/22/the-power-of-mysql-group_concat/
*/
if object_id('group_c') is not NULL
drop table group_c
go
create table group_c
(parent_id int NULL
,child_id int NULL
)
go
insert group_c values (1,1)
insert group_c values (1,1)
insert group_c values (1,2)
insert group_c values (1,3)
insert group_c values (1,4)
insert group_c values (2,1)
insert group_c values (2,4)
insert group_c values (2,6)
insert group_c values (3,1)
insert group_c values (3,2)
insert group_c values (4,1)
insert group_c values (4,1)
insert group_c values (5,0)
go
-----
print "
******************************
QUERY # 1 : List of parents and associated children (default separator = ',')
******************************
"
select parent_id,
-- we know child_id_list is relatively narrow in this case
-- so reduce the width of the output via left(--,20)
left( dbo.sp_f_group_concat("select distinct
convert(varchar(100), child_id)
from group_c
where parent_id = " + convert(varchar, parent_id) + "
order by child_id"
, default
)
,20) as child_id_list
from group_c
group by parent_id
order by parent_id
/*
results should look like:
parent_id child_id_list
----------- --------------------
1 1,2,3,4
2 1,4,6
3 1,2
4 1
5 0
*/
go
-----
print "
******************************
QUERY # 1 : List of parents and associated children (separator = ' - ')
******************************
"
select parent_id,
-- we know child_id_list is relatively narrow in this case
-- so reduce the width of the output via left(--,20)
left( dbo.sp_f_group_concat("select distinct
convert(varchar(100), child_id)
from group_c
where parent_id = " + convert(varchar, parent_id) + "
order by child_id"
, " - "
)
,20) as child_id_list
from group_c
group by parent_id
order by parent_id
/*
results should look like:
parent_id child_id_list
----------- --------------------
1 1 - 2 - 3 - 4
2 1 - 4 - 6
3 1 - 2
4 1
5 0
*/
go
++++++++++++++ sp_f_group_concat.test2.sql
- percona.com examples using the engineers/customers/issues/workflow tables
- second query (nested GROUP_CONCAT() calls) uses a temp table to store intermediate results since a single-query solution was just too convoluted/complex
use tempdb
go
set nocount on
go
/*
reproduction of the MySQL/GROUP_CONCAT() examples from:
https://www.percona.com/blog/2013/10/22/the-power-of-mysql-group_concat/
Assumptions
===========
- ASE's identity column attribute generates the same results as
MySQL's AUTO_INCREMENT column attribute
- otherwise the auto-generated customer.id values won't
match the manually entered issues.company_id values
- otherwise the auto-generated engineers.id and issues.id
values won't match the manually entered values for
workflow's engineer_id/issue_id pairs
*/
if object_id('engineers') is not NULL
drop table engineers
go
create table engineers
(id smallint identity
,e_name varchar(30) not NULL
,e_surname varchar(30) not NULL
,url varchar(255) not NULL
)
go
alter table engineers
add primary key (id)
go
insert engineers (e_name, e_surname, url) values ('Miguel', 'Nieto', 'https://www.percona.com/about-us/our-team/miguel-angel-nieto')
insert engineers (e_name, e_surname, url) values ('Marcos', 'Albe', 'https://www.percona.com/about-us/our-team/marcos-albe')
insert engineers (e_name, e_surname, url) values ('Valerii', 'Kravchuk', 'https://www.percona.com/about-us/our-team/valerii-kravchuk')
insert engineers (e_name, e_surname, url) values ('Michael', 'Rikmas', 'https://www.percona.com/about-us/our-team/michael-rikmas')
go
if object_id('customers') is not NULL
drop table customers
go
create table customers
(id smallint identity
,company_name varchar(30) not NULL
,url varchar(255) not NULL
)
go
alter table customers
add primary key (id)
go
insert customers (company_name, url) values ('OT','http://www.ovaistariq.net/')
insert customers (company_name, url) values ('PZ','http://www.peterzaitsev.com/')
insert customers (company_name, url) values ('VK','http://mysqlentomologist.blogspot.com/')
insert customers (company_name, url) values ('FD','http://www.lefred.be/')
insert customers (company_name, url) values ('AS','http://mysqlunlimited.blogspot.com/')
insert customers (company_name, url) values ('SS','https://www.flamingspork.com/blog/')
go
if object_id('issues') is not NULL
drop table issues
go
create table issues
(id smallint identity
,customer_id smallint not NULL
,description text
)
go
alter table issues
add primary key (id)
go
insert issues (customer_id, description) values (1,'Fix replication')
insert issues (customer_id, description) values (2,'Help with installation of Percona Cluster')
insert issues (customer_id, description) values (3,'Hardware suggestions')
insert issues (customer_id, description) values (4,'Error: no space left')
insert issues (customer_id, description) values (5,'Help with setup daily backup by Xtrabackup')
insert issues (customer_id, description) values (6,'Poke sales about Support agreement renewal')
insert issues (customer_id, description) values (4,'Add more accounts for customer')
insert issues (customer_id, description) values (2,'Create Hot Fix of Bug 1040735')
insert issues (customer_id, description) values (1,'Query optimisation')
insert issues (customer_id, description) values (1,'Prepare custom build for Solaris')
insert issues (customer_id, description) values (2,'explain about Percona Monitoring plugins')
insert issues (customer_id, description) values (6,'Prepare access for customer servers for future work')
insert issues (customer_id, description) values (5,'Decribe load balancing for pt-online-schema-change')
insert issues (customer_id, description) values (4,'Managing deadlocks')
insert issues (customer_id, description) values (1,'Suggestions about buffer pool size')
go
if object_id('workflow') is not NULL
drop table workflow
go
create table workflow
(action_id int identity
,engineer_id smallint not NULL
,issue_id smallint not NULL
)
go
alter table workflow
add primary key (action_id)
go
insert workflow (engineer_id, issue_id) values (1,1)
insert workflow (engineer_id, issue_id) values (4,2)
insert workflow (engineer_id, issue_id) values (2,3)
insert workflow (engineer_id, issue_id) values (1,4)
insert workflow (engineer_id, issue_id) values (3,5)
insert workflow (engineer_id, issue_id) values (2,6)
insert workflow (engineer_id, issue_id) values (3,7)
insert workflow (engineer_id, issue_id) values (2,8)
insert workflow (engineer_id, issue_id) values (2,9)
insert workflow (engineer_id, issue_id) values (1,10)
insert workflow (engineer_id, issue_id) values (3,11)
insert workflow (engineer_id, issue_id) values (2,12)
insert workflow (engineer_id, issue_id) values (2,13)
insert workflow (engineer_id, issue_id) values (3,14)
insert workflow (engineer_id, issue_id) values (1,15)
insert workflow (engineer_id, issue_id) values (1,9)
insert workflow (engineer_id, issue_id) values (4,14)
insert workflow (engineer_id, issue_id) values (2,9)
insert workflow (engineer_id, issue_id) values (1,15)
insert workflow (engineer_id, issue_id) values (3,10)
insert workflow (engineer_id, issue_id) values (4,2)
insert workflow (engineer_id, issue_id) values (2,15)
insert workflow (engineer_id, issue_id) values (4,8)
insert workflow (engineer_id, issue_id) values (4,4)
insert workflow (engineer_id, issue_id) values (3,11)
insert workflow (engineer_id, issue_id) values (1,7)
insert workflow (engineer_id, issue_id) values (3,7)
insert workflow (engineer_id, issue_id) values (1,1)
insert workflow (engineer_id, issue_id) values (1,9)
insert workflow (engineer_id, issue_id) values (3,4)
insert workflow (engineer_id, issue_id) values (4,3)
insert workflow (engineer_id, issue_id) values (1,5)
insert workflow (engineer_id, issue_id) values (1,7)
insert workflow (engineer_id, issue_id) values (1,4)
insert workflow (engineer_id, issue_id) values (2,4)
insert workflow (engineer_id, issue_id) values (2,5)
go
print "
******************************
QUERY # 1 : List of issues for each engineer
******************************
"
/*
for display purposes we'll use left() to reduce column widths based on known max widths for the test data
*/
select left(e.e_name + ' ' + e.e_surname, 20) as engineer,
left(dbo.sp_f_group_concat("select distinct
convert(varchar,w.issue_id) + ' (' + c.company_name + ')'
from workflow w,
engineers e,
customers c,
issues i
where w.engineer_id = e.id
and w.issue_id = i.id
and i.customer_id = c.id
and e.id = " + convert(varchar,e.id) + "
order by w.issue_id"
, ', ')
, 80) as 'issue (customer)'
from workflow w,
engineers e,
customers c,
issues i
where w.engineer_id = e.id
and w.issue_id = i.id
and i.customer_id = c.id
group by e.id
order by e_name, e_surname
/*
results should look like:
engineer issue (customer)
------------------------------ --------------------------------------------------------------------------------
Marcos Albe 3 (VK), 4 (FD), 5 (AS), 6 (SS), 8 (PZ), 9 (OT), 12 (SS), 13 (AS), 15 (OT)
Michael Rikmas 2 (PZ), 3 (VK), 4 (FD), 8 (PZ), 14 (FD)
Miguel Nieto 1 (OT), 4 (FD), 5 (AS), 7 (FD), 9 (OT), 10 (OT), 15 (OT)
Valerii Kravchuk 4 (FD), 5 (AS), 7 (FD), 10 (OT), 11 (PZ), 14 (FD)
*/
go
print "
******************************
QUERY # 2 : List of engineers for each customer (nested group_concat() calls)
******************************
"
/*
while technically possible to nest our sp_f_group_concat() calls, the outer
call becomes unwieldly since it will have to duplicate a copy of the inner
call (and the full text for the e_list derived table) for each company;
reason being that the e_list derived table has to be re-created for each
outer call (per company)
to make the code easier to read we're going to materialize the e_list derived table
as a #temp table; for large data sets we'd want to look at the feasibilty of
adding an index for performance reasons
for display purposes we'll use left() to reduce column widths based on known
max widths for the test data
*/
-- build/populate the #e_list table with a set of issue id's and associated engineer lists
if object_id('#e_list') is not NULL
drop table #e_list
go
create table #e_list
(i_id int
,engineer_list varchar(1000)
)
go
insert #e_list
select i.id as i_id,
dbo.sp_f_group_concat("select distinct
e.e_name + ' ' + e.e_surname
from workflow w,
engineers e,
issues i
where w.engineer_id = e.id
and w.issue_id = i.id
and i.id = " + convert(varchar, i.id) + "
order by e.e_name, e.e_surname"
, ', ') as engineer_list
from workflow w,
engineers e,
issues i
where w.engineer_id = e.id
and w.issue_id = i.id
group by i.id
go
-- now run the main query to display isuses/engineer-lists by company
select left(c.company_name, 10) as company,
left(dbo.sp_f_group_concat("select distinct
convert(varchar,e_list.i_id) + ' (' + e_list.engineer_list + ')'
from workflow w,
engineers e,
customers c,
issues i,
#e_list e_list
where w.engineer_id = e.id
and w.issue_id = i.id
and i.customer_id = c.id
and w.issue_id = e_list.i_id
and c.id = " + convert(varchar, c.id) + "
order by w.issue_id"
, ', ' )
, 140) as issue
from workflow w,
engineers e,
customers c,
issues i,
#e_list e_list
where w.engineer_id = e.id
and w.issue_id = i.id
and i.customer_id = c.id
and w.issue_id = e_list.i_id
group by c.id
order by c.company_name
/*
results should look like:
company issue
---------- --------------------------------------------------------------------------------------------------------------------------------------------
AS 5 (Marcos Albe, Miguel Nieto, Valerii Kravchuk), 13 (Marcos Albe)
FD 4 (Marcos Albe, Michael Rikmas, Miguel Nieto, Valerii Kravchuk), 7 (Miguel Nieto, Valerii Kravchuk), 14 (Michael Rikmas, Valerii Kravchuk)
OT 1 (Miguel Nieto), 9 (Marcos Albe, Miguel Nieto), 10 (Miguel Nieto, Valerii Kravchuk), 15 (Marcos Albe, Miguel Nieto)
PZ 2 (Michael Rikmas), 8 (Marcos Albe, Michael Rikmas), 11 (Valerii Kravchuk)
SS 6 (Marcos Albe), 12 (Marcos Albe)
VK 3 (Marcos Albe, Michael Rikmas)
*/
go
++++++++++++++
NOTE: For the example queries you'll notice that the @sql string passed to the sp_f_group_concat() function is basically a copy of the parent query plus an additional 'where' clause to allow limiting the query to just the rows that match the parent query's 'group by' clause (ie, the additional 'where' clause matches the column(s) in the 'group by' clause)
来源：https://stackoverflow.com/questions/43758776/create-concatenate-sql-query-for-multiple-rows