Comments on: How slow are (Ruby) Exceptions?

By: Métodos que retornam mais de um valor em Ruby | Prodis a.k.a. Fernando Hamasaki de Amorim

Sat, 20 Feb 2010 19:53:05 +0000

[...] Ele indicou um post falando a respeito: http://www.simonecarletti.com/blog/2010/01/how-slow-are-ruby-exceptions [...]

By: Caffeine Driven Development » Blog Archive » L33t Links #68

Caffeine Driven Development » Blog Archive » L33t Links #68 — Fri, 15 Jan 2010 09:10:43 +0000

[...] How slow are (Ruby) Exceptions? [...]

By: Peter Cooper

Peter Cooper — Wed, 13 Jan 2010 16:56:09 +0000

A key finding from this benchmark, IMHO, is that exceptions do not slow things down when they’re not raised. That is, having the rescue mechanism in place does not seem to slow down properly working code by any serious degree.. it’s only when exceptions are actually raised that they’re “slow.”

I agree with Thorsten Böttger. Since exceptions are not/should not be part of ordinary flow – hence their name – I don’t see their lack of performance as motivation for not using them in places where they really are “exceptional.” Just in case anyone gets the wrong idea from these findings.. ;-)

By: Simone Carletti

Simone Carletti — Wed, 13 Jan 2010 14:22:31 +0000

Indeed, it depends on the specific code.

The example posted above was a simplification of the original code, where I actually saved up to 10 hours per 900K records just refactoring the logic in order to avoid exceptions.

You can write as many benchmarks as you wish, but you won’t be able to determine whether skipping Exception means to be x2, x4 or xN times faster.

Here’s an other example.
http://gist.github.com/275768#file_if_vs_exception_vs_regexp_zerodivision.rb

Here the difference is smaller compared to the original benchmark also because the execution stack is smaller resulting in a more lightweight script.

However, we all agree that exceptions cost. ;)

By: Curtis Summers

Curtis Summers — Wed, 13 Jan 2010 14:17:39 +0000

In fact, it’s very likely that circumventing the call to IPSocket.getaddress is the majority of the time difference for this example.

By: Curtis Summers

Curtis Summers — Wed, 13 Jan 2010 14:00:30 +0000

Yes, exceptions are slow, but this is not a good example. There’s quite a bit more logic in IPAddr#initialize, and this example is simply circumventing all of that additional code:

http://ruby-doc.org/core/classes/IPAddr.src/M000734.html

By: Simone Carletti

Simone Carletti — Wed, 13 Jan 2010 13:36:18 +0000

Thanks Vladimir!

I was planning to run the benchmarks on JRuby. This is a really interesting comparison, thank you for posting here the results.

By: Vladimir Sizikov

Vladimir Sizikov — Wed, 13 Jan 2010 13:29:16 +0000

Interestingly, on JRuby, the difference is much, much smaller: 2sec vs 1.5sec.

By: Thorsten Böttger

Thorsten Böttger — Wed, 13 Jan 2010 09:57:38 +0000

What’s the problem with slow exceptions? Exceptions are not part of the normal control flow, and are by definition unexpected. Thus, nobody expects outstanding performance for exceptions…