Re: Notice that lists comprehension at least twice as slow

[José Valim]

Hello Alex,

Thanks for the benchmarks.

The reason this happens is because our list comprehensions consider everything in filters except false or nil to be true.

So this additional check costs us a bit. We could certainly optimize this though, since operators (like < and >) and other BIFs are known to return just booleans, allowing us to skip this check, I just haven't got to this optimization stage (it surprises me that this check is causing all this difference though).

On Saturday, July 21, 2012 10:22:05 PM UTC+2, Alex Zatvorskiy wrote:

Hey,

I have been doing some benchmarks. So, I have found that elixir's lc at least as slow then erlang's original feature.

There are some benchmarks:

• https://github.com/AZatvornitskiy/nodexs/blob/rc/test/erl.erl
  
• https://github.com/AZatvornitskiy/nodexs/blob/rc/test/ex.ex
  

Just for notice )) 

P.S: On the other hands, you should spread your purpose, and roadmap in general.

Stay well )) 

[José Valim]

Hello Alex,

I have pushed a commit to Elixir master that skips the conversion check if the expression is known to return a boolean:

https://github.com/elixir-lang/elixir/commit/6a7602d974f13924a6db6a52cd495f26945864f6

This brings Elixir's performance to the same level as Erlang's for such common comprehensions.

Cheers!

José Valim

www.plataformatec.com.br

Founder and Lead Developer

[Alex Zatvorskiy]

Thanks a lot, I'll proceed