Remove Advisory lock

[brendon]

This is in favour of scope locks or locking all the records in a table if no scope exists.

[brendon]

@fractaledmind, sorry this is a little out of the blue, but I wondered if you had any thoughts about why my sqlite tests might be deadlocking? I've had a go at integrating your gem but there still appears to be a problem. The problem test in particular is:

positioning/test/test_positioning.rb

Lines 31 to 51 in 25fcdca

	def test_no_duplicate_row_values_when_creating
	ActiveRecord::Base.connection_handler.clear_all_connections!
	list = List.create name: "List"
	students = []
	10.times do
	threads = 20.times.map do
	Thread.new do
	ActiveRecord::Base.connection_pool.with_connection do
	students << list.authors.create(name: "Student", type: "Author::Student")
	end
	end
	end
	threads.each(&:join)
	end
	assert_equal (1..students.length).to_a, list.authors.map(&:position)
	list.destroy
	end

I'd love to keep sqlite as a compatible database for this gem but am getting close to dropping it as the ability to keep the list ordering clean is the primary purpose of this library :)

I frequently encountered your work on getting sqlite working well in production for Rails. Well done! :)

[fractaledmind]

Let me dig in tonight and get back to you.

[brendon]

Thanks @fractaledmind :D

[rossme]

Hi @brendon, I was running the tests earlier and I had to increase the pool size to 20 in database.yml which solved the issue for me. Not sure if that helps or not. I've used acts_as_list and this is a terrific improvement over that gem. Nice job.

[brendon]

Thanks @rossme, that doesn't seem to have helped in the CI environment. I feel like there's something different about it vs the Mac environment.

[brendon]

Thank you for your kind words too! :D

[brendon]

Managed to remove default_transaction_mode: IMMEDIATE also. Maybe the tests aren't robust enough though.

[brendon]

Also managed to remove the activerecord-enhancedsqlite3-adapter gem. Sqlite tests pass with or without the row locking (as expected). They fail for the other databases without the locking, I guess indicating that the sqlite connection pool only gives write access one at a time to each thread which is what the row locking does for other databases. Just conjecture on my part.

Has anyone tried these modifications in production?

[brendon]

I've merged this into main and will do a new minor release since there is a breaking change. Please get in touch if you notice any issues.