Fix memory leak in Oj.dump if raise an exception

[Watson1978]

If an exception occurs inside Oj.dump, the heap area has not been released.
The heap area allocated by ALLOC_N must be released.
This patch will fix this problem.

Test code

require 'oj'

data = {
  "foo" => "a" * 1024 * 1024 * 10,
  1 => "Invalid hash key with strict mode"
}

1000.times do |i|
  begin
    Oj.dump(data, mode: :strict)
  rescue
  end

  if i % 100 == 0
    rss = Integer(`ps -o rss= -p #{Process.pid}`) / 1024.0
    puts "#{i}, #{rss}"
  end
end

Before

$ ruby json.rb
0, 49.4375
100, 1051.0
200, 1674.03125
300, 1773.625
400, 1822.171875
500, 1842.078125
600, 1841.015625
700, 1815.65625
800, 1781.53125
900, 1792.28125

1.8 GB of memory was used in my enviroment.

After

$ ruby json.rb
0, 49.40625
100, 49.421875
200, 49.53125
300, 49.59375
400, 49.65625
500, 49.6875
600, 49.71875
700, 49.734375
800, 49.765625
900, 49.796875

50 MB of memory was used in my enviroment.

[ohler55]

Nicely done. Just one tiny request for a change.

[Watson1978]

Just one tiny request for a change.

OK, sure. What kind of request is that?

[ohler55]

Minor nit. Move both dump_body and dump_ensure functions above the dump function and then remove these 2 lines.

[Watson1978]

OK. Fixed at 0ab66a6

Thanks

[ohler55]

Sorry, I've gotten used to using gitlab and forgot to click the submit review.

[ohler55]

Thank you for the fix.

[ohler55]

I don't have your email address but Oj is getting a parser rework and I would be honoured if you would like to act as a reviewer of the code when it is ready for a merge. Early result show the new parser could be as much as twice as fast as the current Oj parser.

[Watson1978]

Sounds great. OK, sure :)