Make average compatible across ruby versions

[albertoalmagro]

Summary

Since Ruby 2.6.0 NilClass#to_d is returning BigDecimal 0.0. This breaks average compatibility with prior Ruby versions. This patch makes average return nil in all Ruby versions when there are no rows.

Other Information

Reverts #34601
Fixes #34850

r? @rafaelfranca

[kamipo]

Just enough value&.respond_to?(:to_d) ? value.to_d : value or value && value.respond_to?(:to_d) ? value.to_d : value.

Suggested change

	when "average" then cast_average(value)
	when "average" then value && value.respond_to?(:to_d) ? value.to_d : value

[albertoalmagro]

Thanks for the suggestion @kamipo change applied. 🙏

[albertoalmagro]

r? @kamipo

[kamipo]

Returning nil for average no records is preferable to me.

The reason that I pasted the all operation cases in #34601 (comment) was for when "sum".

rails/activerecord/lib/active_record/relation/calculations.rb

Lines 400 to 405 in e7717ec

	def type_cast_calculated_value(value, type, operation = nil)
	case operation
	when "count" then value.to_i
	when "sum" then type.deserialize(value || 0)
	when "average" then value.respond_to?(:to_d) ? value.to_d : value
	else type.deserialize(value)

Databases also returns nil for sum no records, but now Active Record regards no records sum as 0 instead of nil.

I believe that is historical reason, but it is a good timing to discuss the odd inconsistency for no records sum and average.

[rafaelfranca]

I think it makes sense to sum return 0 and average return nil. For sum it doesn't make difference if there is no records or if there is many that the sum returns 0, but for average it does make a difference since you don't know if 0 is because you have no record or because you have two records with -10 and 10 as value.

[kamipo]

I don't know I get your point, you means that the difference (1) and (3) is less important than the difference (2) and (4), right?

root@localhost [test] > create table t (c int);
Query OK, 0 rows affected (0.03 sec)

root@localhost [test] > select sum(c) from t; -- (1)
+--------+
| sum(c) |
+--------+
|   NULL |
+--------+
1 row in set (0.00 sec)

root@localhost [test] > select avg(c) from t; -- (2)
+--------+
| avg(c) |
+--------+
|   NULL |
+--------+
1 row in set (0.00 sec)

root@localhost [test] > insert into t values (10), (-10);
Query OK, 2 rows affected (0.00 sec)
Records: 2  Duplicates: 0  Warnings: 0

root@localhost [test] > select sum(c) from t; -- (3)
+--------+
| sum(c) |
+--------+
|      0 |
+--------+
1 row in set (0.00 sec)

root@localhost [test] > select avg(c) from t; -- (4)
+--------+
| avg(c) |
+--------+
| 0.0000 |
+--------+
1 row in set (0.00 sec)

[rafaelfranca]

Yes. I can't think in a case where I need to know the sum of all records is 0 because there are no records or because all records are 0.

[kamipo]

I see, I agree that the difference is less important.