ruby 2.0.0 has been released, I wanted to use it. I have built
it the day after on my notebook, but it took more than two times as much time
to run all the specs for my recent rails project, than with
falcon-gc. I’m certainly getting the wrong results, don’t I? Sure I
do. Yesterday I grabbed my pickaxe, to dig out the root of the problem.
My first thought was it’s a bug in my build. I have tried a lot of extra
cflags, but no avail. I also installed the
homebrew version. Same
Or, with tables (sorry RedCloth markdown tables are just plain ugly):
This doesn’t look too good.
OK, what if it’s my computer? I’m using a late-2010 Macbook Air with 4GB RAM soldered on it. A better computer, a closer target (I’m deploying to 64bit ubuntu server), and more RAM would help. Thus, I have set up a development environment on a Ubuntu 12.04.02 box, with a 8-core i7 3660, with 32GB RAM, with no load whatsoever. I have ran the same tests (but with a bit different versions though) there too:
Hey, where is my big difference? It certainly shows what Funny Falcon has done with 1.9.3, but 2.0.0 results are just A-OK. I can live with this performance difference.
Ah well. What takes much more on my computer than it should? It’s time to put the chainsaw away, and get a plasma blaster.
My weapon of choice was
ruby-prof. I wanted to profile my specs
though, and it is not trivial to attach
Fortunately, Colin MacKenzie IV wrote
rspec-prof gem to grease the
wheels. It’s just adding the gem, putting specs into a
... end block, and looking at the generated files afterwards from
To make the long story short it turned out I have created a new user
account for every single controller tests, which encrypted its password
bcrypt. For those who don’t know this, bcrypt
is well known being resistant against brute-force, because its iteration
count can be tuned to slow processing down.
Therefore the answer for my question why ruby 2 is slower is because I asked it to be slower.
Ask yourself: why your class spend precious time doing unnecessary stuff like encryption during tests? Of course you have to, if this is your core business. It’s not the case though in most situations though: patch through the slow method then:
has_secure_password requires are not hard to find
either (I haven’t tested it, sorry):
The new results are stunning:
I don’t know what happened with rbenv’s 2.0.0-p0, but I don’t even care.
All the other results are expected, it’s safe now to switch to