Convert Ruby 1.8 to 1.9.2 | Hacker News Reader

Convert Ruby 1.8 to 1.9.2

github.com

35 pointsduck8 comments

For a more thorough explanation of functionality, motivation and how it works: http://tagaholic.me/2011/03/05/one9-upgrade-to-ruby-19-now.h...

Interesting tool. I recently got my Rails 2.3 app running in 1.9.2. This utility wouldn't quite cut it though because most of my effort was devoted to getting remarkable_rails working just so I could get my specs to run.

After all that effort, I found out I'm better off sticking with REE. While my specs all did pass, they took 3 - 4 times longer to run in 1.9.2. I ran experiments on that for about a day, not believing the results because I've been told repeatedly that 1.9.2 is unequivocally faster. Alas, that is not the case. So, that's my "excuse" for not running 1.9.2.

I kind of believe slowness in test runs vs 1.8MRI or REE depends on bytecode compiling (though i have not proof of this). rails and rspec include a massive amount of code that gets loaded/compiled, and that is done every time you run the tests without giving time to YARV to express itself in the faster paths, so this difference is more noticeable.

Would spork or a similar testing tool that loads the framework+gems once and then reuses them for each test run help with this issue do you think?

Hopefully the bytecode would be kept by the VM between each test run. I don't know enough about the ruby interpreter to say, especially not YARV.

it makes sense to me, I shall give it a try and report

If your changes were method changes, one9 should catch them. As explained in the Rails section of this post, http://tagaholic.me/2011/03/05/one9-upgrade-to-ruby-19-now.h..., you can always jump into a troublesome gem's source, install its dependencies and run one9 on its tests.

It's a logic error in remarkable_rails that results in an infinite loop on Ruby 1.9.

Andrew Selder tracked the issue down and provided the fix:

https://rspec.lighthouseapp.com/projects/5645/tickets/1027-s...

In production use, we found we could run double the load on our servers with 1.9 than we could with 1.8. Switching to 1.9 kept us running our app on a cluster of 8 servers as our user base grew, saving us the monthly rental of 8 servers and the overhead of running twice as many web servers.

We never measured how much longer our tests took to run though, since our customers were only ever running in Production mode

Reply on news.ycombinator.com