Rule Suggestion: disallow expression on same line as SELECT (if num_expressions>1)

Given this example:

SELECT Md5(-1)            AS gl_account_key,
       0                  AS gl_account_number,
       'Unknown'          AS gl_account_name,
       'Unknown'          AS gl_major_category,
       'Unknown'          AS reporting_major_category

SQLFluff fixes as:

SELECT Md5(-1)            AS gl_account_key,
    0                  AS gl_account_number,
    'Unknown'          AS gl_account_name,
    'Unknown'          AS gl_major_category,
    'Unknown'          AS reporting_major_category

Whereas I think the best practice should be:

SELECT 
    Md5(-1)            AS gl_account_key,
    0                  AS gl_account_number,
    'Unknown'          AS gl_account_name,
    'Unknown'          AS gl_major_category,
    'Unknown'          AS reporting_major_category

In other words: If the number of expressions is greater than one, none of them should share a line with SELECT and all should be indented the standard 4 characters.

Thoughts? Agree/disagree?

I qualify based on the number of expressions because a case can be made that this:

SELECT
    mycol
FROM mytable

…is less desirable/readable than:

SELECT mycol
FROM mytable

[aaronsteers]

@pwildenhain - Indeed. Thank you for linking the two. Where the two differ, as far as I can tell, is just the comment from Alan on #352 related to multiple first line columns being okay, whereas here I’ve suggested only allowing when a single expression is named.

Taking the safe assumption that line length is already enforced, I think I’m okay with all of these:

select col_a, col_b from my_table

select col_a, col_b
from my_table

select col_a
from my_table

Whereas these would be anti-patterns:

select col_a,
    col_b
from my_table

select col_a, col_b
    col_c
from my_table

[alanmcruickshank]

I’m open to this as a rule, with a few potential exceptions.

• If there is only a single column in the select, then the rule does not apply (as per your last example), this would be similar to the logic of L013, which allows SELECT 1 FROM blah without an alias for the column.
• Your first example actually has a hanging indent, and I think these should still be allowed, although perhaps this should be configurable for this rule. I’ll add some examples below.
• The configuration from how much to indent by default should come from the master config and respect the overall settings for tabs/spaces etc…

e.g.

GOOD (Scalar query)

SELECT 1
FROM foo

(Hanging indent, i.e. a & c have the same indent. There’s some logic in L003 to work this out that should probably be more generic.)

SELECT a, b,
       c, d
FROM bar

BAD

SELECT a,
    b
FROM foobar

… should fix to …

SELECT
    a,
    b
FROM foobar