Add support for passing a block to IO#readlines and IO.readlines.

[ioquatix]

IO#readlines can allocate memory proportional to the file size. This can be a problem. While IO#each_line exists, IO#readlines is convenient. So, let's extend IO#readlines and IO.readlines to accept a block, and yield one line at a time, to avoid memory usage issues when dealing with large files.

cc @youchan

[ioquatix]

Here is a small example program I was using for testing:

#!/usr/bin/env ruby

PATH = ARGV.pop || '/tmp/fifo'

class Producer
  def initialize
    @queue = Thread::Queue.new
  end

  def dequeue
    @queue.pop
  end

  def run
    Thread.new do
      # read data from named pipe
      File.readlines(PATH) do |line|
        @queue.push line
      end
    end
  end
end

# consumer
producer = Producer.new
producer.run
loop do
  p producer.dequeue
end

Most of the credit goes to @youchan.

[ioquatix]

@mame mentioned that this function can already be done by IO.foreach. I agree there is significant overlap.

[nobu]

Can be closed?

[ioquatix]

I am fine to close this, however I think what this shows is that there is a lack of consistency in the IO interface. I've actually made exactly the same mistake before, expecting IO.readlines takes a block.

The other problem is, whether we should encourage this approach at all. IO.readlines could work for trivial cases, but fail with big files. Users may not be aware of the limitations. For Ruby 3.3, I'm fine with not making any changes. However, I wonder if we should consider this more carefully. After all, the public interface on core classes will guide our users/programmers and I think we should guide them towards incremental/lazy each_line style patterns.

[ioquatix]

If you check https://github.com/search?q=%2FIO%5C.readlines%28.*%3F%29do%2F&type=code you can see plenty of examples of poor quality code, e.g. IO.readlines.each do which with this PR becomes IO.readlines do. Without this PR, it should be IO.foreach(...) do. So, please ask yourself the question, why do people ignore IO.foreach? I think it's because IO.readlines name matches closely with how people are thinking about the problem. In other words, IO.foreach is not obvious.