In a join, how to prefix all column names with the table it came from

Question

I\'m analysing a rather horrible legacy database/codebase, trying to reduce server load by combining queries into joins (including an email alert cron job that typically inv

Answer 1

I ended up just building the field set for the query, as as of 2020 this still isn't supported.

But, being a lazy programmer, I obviously didn't want to manually type this all out for all of the tables in my query. So I wrote a query to build the select statement:

SELECT
    CONCAT(table_name, ".", column_name, " AS ", CHAR(34), table_name, ".", column_name, CHAR(34)) field_names
FROM
    information_schema.columns
WHERE
    table_schema = "my_database"
    AND table_name IN(
        "table_1",
        "table_2"
    );

which will output something like:

| field_names                        |
|------------------------------------|
| table_1.id AS "table_1.id"         |
| table_1.name AS "table_1.name"     |
| table_2.id AS "table_2.id"         |
| table_2.number AS "table_2.number" |

That can then easily be copied into your SELECT statement.

Answer 2

@alden-w, You may add TABLE_SCHEMA condition to where to do not mix up same table names from different schemas

WHERE c.TABLE_SCHEMA='YOUR_SCHEMA_NAME' AND c.TABLE_NAME IN (....)

Answer 3

You could name the fields in your query and give them aliases:

SELECT     ah.whateverfield1 AS 'ah_field1',
           ah.whateverfield2 AS 'ah_field2',
           l.whateverfield3 AS 'l.field3',
           [....]
FROM       class_alerts_holding ah 
INNER JOIN class_listings l ON l.id = ah.lid 
INNER JOIN class_users u ON u.id = ah.uid
LEFT JOIN  class_prodimages pi ON pi.pid = ah.lid

Its a bit of work to manually set up if you have that many fields, but you can simplify this with this query...

SHOW FULL FIELDS FROM your_table_name;

...and a good text editor and copy & paste.

Answer 4

I am convinced that such feature to prefix and/or postfix fields names with a table name in a join SHOULD BE INCLUDED INTO ANSI SQL STANDARD. Currently, in year 2019, there is still no elegant cross-platform way to do it, and all what's left is ugly-looking and error-prone manual hacking with aliases, or platform-specific solutions involving dynamic sql. Everyone would really benefit from having ability to specify custom prefix or/and postfix to fields denoted by 'dot-star' (.*). Sample select after adding such feature would be:

select a.* use prefix,b.* use postfix '_b' from table_a a inner join table_b b on a.id=b.id

As you can see, by default prefix or postfix would equal table name (or alias name), and can be overridden with any desired string literal.

Also what's aching to be added to standard, is ability to exclude certain fields from 'starred' (*) output, which is a shortcut to select all fields. I would add except keyword to list fieds which I do not want to be included for reasons of reducing network data transfer or/and brevity, e.g. :

select * except large_binary_data_field,another_notneeded_field,etc from my_table

Such feature would allow to avoid necessity of explicitly specifying full (and potentially large) list of fields which are needed as opposed to only specifying star and a few fields which are not needed.

So please, whoever reading this post and being able to reach out to ANSI SQL standard influencers, you know what to do )

P.S. yet another ugly, but at least automated & generic dynamic sql wrapper

For the Python advocates who work with psycopg, here is the convenient sub I use (strictly internally, as it's prone to possible sql injections)

def get_table_fields(table,alias,prefix='',suffix='',excluding=''):
    if type(excluding)==str: excluding=excluding.split(',')
    cur.execute('select * from '+table+' where 0=1');cur.fetchall()
    if not (cur.description is None):        
        return ','.join([alias+'.'+col.name+' '+prefix+col.name+suffix for col in cur.description if not (col.name in excluding)])

And the calling code, where I am joining 3 tables and want to avoid fetching large data field from the datasets table:

sql="""select %s,%s,%s from tasks t,features_sets f,datasets d 
        where 
                t.is_active=true and f.is_active=true 
                and f.task=t.id and t.train_dataset=d.id 
    """ % (
        get_table_fields('tasks','t',prefix='ts_'),
        get_table_fields('features_sets','f',prefix='fs_'),
        get_table_fields('datasets','d',prefix='ds_',excluding='data')
    )

which gets unrolled for me into mighty

select t.id ts_id,t.project ts_project,t.name ts_name,***,
    fs_id,f.task fs_task,f.name fs_name,f.description fs_description,***,
    d.id ds_id,d.project ds_project,d.name ds_name,***
from tasks t,features_sets f,datasets d 
    where 
        t.is_active=true and f.is_active=true 
        and f.task=t.id and t.train_dataset=d.id 

where *** means tons of other useful fields, some of them are common for more than one table (hence the need for prefixing). cur is obviously the psycopg cursor, and 0=1 condition is intended to retrieve only fields names without real data.

Answer 5

I've found something usefull in this question MySQL concat() to create column names to be used in a query? . I think that this can be one of the solutions.

Answer 6

Based on the solution proposed by koljaTM and AndriyM, maybe an even better solution is to write your query like this:

select
  '--TABLE_AAA:--', TABLE_AAA.*,
  '--TABLE_BBB:--', TABLE_BBB.*,
  '--TABLE_CCC:--', TABLE_CCC.*,
  '--TABLE_DDD:--', TABLE_DDD.*
from ...

Unfortunately this is still not good enough in cases when one (or more) of the tables contains more column names than can fit on the screen width. (So you might see on your screen 20 columns but still not be visible on the screen the name of the table from which they come.)

It would still have been better if SQL provided a way to automatically prefix the column names with the table names...