Should not change serializable value for the enum attribute

[kamipo]

EnumType#serializable? which allows values only in the enum mapping
was added in 72fd0ba, and it is called only in HomogeneousIn, it
caused the serialization inconsistency whether the values are in IN
clause or not.

# SELECT "users".* FROM "users" WHERE "users"."role" = '0'
User.where(role: "0")

# SELECT "users".* FROM "users" WHERE "users"."role" = '1'
User.where(role: "1")

# SELECT "users".* FROM "users" WHERE "users"."role" IN (NULL)
User.where(role: ["0", "1"])

Even if we introduce validation for the enum serialization in the future,
we should at least not cause the inconsistency for now.

Fixes #41474.

cc @eileencodes

[eileencodes]

Can you instead remove the super from serializable??

[yevgenko]

@eileencodes I've tried to restore serializable?, but test has failed.
Looks like removing super from serializable? alone can't fix the problem, but removing serializable? can.

test failed

Using sqlite3
Run options: --seed 20368

# Running:

.........F

Failure:
EnumTest#test_find_via_where_with_values.to_s [/vagrant/rails/activerecord/test/cases/enum_test.rb:69]:
--- expected
+++ actual
@@ -1 +1 @@
-#<Book id: 1, author_id: 1, format: "paperback", name: "Agile Web Development with Rails", status: "published", last_read: "read", nullable_status: nil, language: "english", author_visibility: "visible", illustrator_visibility: "visible", font_size: "medium", difficulty: "medium", cover: "hard", isbn: nil, external_id: nil, published_on: nil, boolean_status: "enabled", tags_count: 0, created_at: "2021-02-19 16:30:24.429872000 +0000", updated_at: "2021-02-19 16:30:24.429872000 +0000", updated_on: "2021-02-19">
+nil



bin/test test/cases/enum_test.rb:64

...................................................................

Finished in 2.017204s, 38.1716 runs/s, 126.9083 assertions/s.
77 runs, 256 assertions, 1 failures, 0 errors, 0 skips

[yevgenko]

Alternatively we could cast stringified digits into numeric type, for example:

0ffa6e5

What do you think? :)

[kamipo]

@yevgenko It will break the existing behavior effed76.

Can you instead remove the super from serializable??

I'm not sure I get your point, so let me explain why the value.blank? || mapping.has_key?(value) || mapping.has_value?(value) part should not be left in the serializable?.

EnumType#serialize is mapping.fetch(value, value), it means that all values were serializable before 72fd0ba.

rails/activerecord/lib/active_record/enum.rb

Lines 146 to 148 in effed76

	def serialize(value)
	mapping.fetch(value, value)
	end

After 72fd0ba, serializable? is introduced which checks a value is whether cast-able or not, it is used only in HomogeneousIn.

rails/activerecord/lib/arel/nodes/homogeneous_in.rb

Lines 50 to 52 in effed76

	casted_values = values.map do |raw_value|
	type.serialize(raw_value) if type.serializable?(raw_value)
	end

So now, all values are no longer always serializable in HomogeneousIn, it is the cause for the issue #41474.

I'm not sure about we'd like the enum serialization more stricter in the future, at least for now the serialization inconsistency whether in HomogeneousIn or others should be fixed.