Environment Information

$ ruby -v
jruby 9.4.6.0 (3.1.4) 2024-02-20 576fab2c51 OpenJDK 64-Bit Server VM 21.0.1 on 21.0.1 +jit [arm64-darwin]
$ uname -a
Darwin janko.local 23.4.0 Darwin Kernel Version 23.4.0: Fri Mar 15 00:12:41 PDT 2024; root:xnu-10063.101.17~1/RELEASE_ARM64_T8103 arm64

I'm using activerecord-jdbc-adapter version 70.2.

Expected Behavior

I expected same raw connection to be considered identical by JRuby when using as hash key with compare_by_identity turned on.

Actual Behavior

require "bundler/inline"

gemfile do
  gem "activerecord", "7.0.8.1"
  gem "activerecord-jdbc-adapter", "70.2"
  gem "jdbc-sqlite3"
end

require "active_record"

ActiveRecord::Base.establish_connection(adapter: "sqlite3", database: ":memory:")

hash = {}
hash.compare_by_identity

hash[ActiveRecord::Base.connection.raw_connection.connection] = "1"
p hash[ActiveRecord::Base.connection.raw_connection.connection]

The above script prints out nil instead of "1", because JRuby considers the connection objects different, even though they have the same #object_id and #hash, and are equal in terms of #==, #eql? and #equal?.

I discovered this while debugging test failures in sequel-activerecord_connection gem, which enables Sequel to reuse Active Record's DB connection. Sequel internally stores transaction state in a hash with compare_by_identity turned on, where hash keys are raw connection objects. Sequel's JDBC adapter itself doesn't appear to have any issues, this only happens if I swap retrieving the connection to go through Active Record.

I could disable compare_by_identity in the sequel-activerecord_connection gem when using JRuby, but I thought I'd report it since I already did the hard work of debugging this issue, and found it surprising that two objects can communicate they're identical but aren't always 😅