Support multiple indexes on the same column when loading the schema

[agrobbin]

We use Postgres for quite a few of our Rails applications, and one of the things we use constantly are partial unique indices, allowing us to have multiple indices on 1 column with different restrictions based on other column data. While everything in our production systems works fine, and adding new partial unique indices via migrations also works fine, loading multiple indices on the same column (String or Array) via our db/schema.rb file turns out to not work quite as expected.

If you have a schema like this:

ActiveRecord::Schema.define do
  create_table :sponsorships do |t|
    t.string "user_id"
    t.index ["user_id"], name: "index_sponsorships_on_user_id_unique", unique: true, where: "(status = 1)", using: :btree
    t.index ["user_id"], name: "index_sponsorships_on_user_id", using: :btree
  end
end

When you run rails db:schema:load without this patch, your local database will only include the index_sponsorships_on_user_id index, skipping the index_sponsorships_on_user_id_unique index.

With this patch, all indices are loaded into the database, regardless of whether the columns in question already have an index.

As far as I can tell, this bug affects at least Rails 4.0+.

[rails-bot]

r? @matthewd

(@rails-bot has picked a reviewer for you, use r? to override)

[kamipo]

How about it?

       def index(column_name, options = {})
-        indexes[column_name] = options
+        indexes << [column_name, options]
       end

https://github.com/rails/rails/pull/26019/files#diff-21d4fbe002689dc4b0ab29f021585457L338

[agrobbin]

Oh, duh, totally forgot that iterating over an array of arrays will yield each item in the subarray to each. Much simpler change, thanks!

[agrobbin]

@matthewd @maclover7 would love to get this reviewed and merged into Rails core. Any chance one of you has some time to take a look?

[rafaelfranca]

Backported in 809757f

[agrobbin]

Thanks @rafaelfranca!