Re: Less allocated objects on each request

[Aaron Patterson <tenderlove@ruby-lang.org>]

[-- Attachment #1: Type: text/plain, Size: 2851 bytes --]

On Thu, May 12, 2016 at 03:04:54AM +0000, Eric Wong wrote:
> richard schneeman <richard.schneeman@gmail.com> wrote:
> > That first SVN commit is
> >  
> > *https://github.com/ruby/ruby/commit/779ae78995977305aa5aec9cb5b562dcf54c22e7*
> > 
> > The second commit is:
> > *https://github.com/ruby/ruby/commit/58f800a278b8b70463f4afdbb23a918d8ab441ff*
> > 
> > Eric's proposal: *https://bugs.ruby-lang.org/issues/9425*
> > 
> > 
> > I agree that the patch should one day be unnecessary,
> 
> <snip>  Hi Richard, ressurrecting since I believe the day has come :)
> 
> Since rack.git depends on Ruby 2.2.2+ which has the
> aforementioned string deduplication, I think we can undo
> most (if not all) of these ugly constants and reduce constant
> cache overheads while we're at it.
> 
> I can followup with patches to the list hopefully in the next
> few weeks if nobody beats me to it (occupied with other stuff,
> at the moment)
> 
> To summarize, the following are optimized in 2.2 (or earlier):
> 
>   { "foo" => "bar" } # "foo" is deduped
>                      # "bar" is duplicated unless 'frozen_string_literal: true'
> 
>   base_hash["foo"] = "bar" # same as above
>   base_hash["foo"]         # "foo" is deduped ditto
> 
> The Hash#[] and Hash#[]= optimizations unfortunately do not
> transfer to subclasses like HeaderHash.
> 
>   # string literals in the "when" statement
>   # this has been optimized for many years, now.
>   case var
>   when "foo"
>     do_foo
>   when "bar"
>     do_bar
>   end
> 
> "string".freeze on any string literal is deduped since 2.1,
> which might be useful for performance-critical subclasses of Hash.
> 
> "frozen_string_literal: true" magic comment exists since 2.3;
> but I'm not sure how much it helps and it's easy to introduce
> new bugs, so I'm not enthusiastic about it for old code.
> 
> Regardless, constants are expensive in both space and time.  The
> inline constant cache takes space, and the cache needs to be
> checked every time a constant is accessed.  String literals do
> not have that cost.
> 
> Constants still have one minor advantage: runtime NameError in
> case of typos.  But I don't think that benefit is worth it
> as they make code harder-to-read.
> 
> I mainly bring this up because I had trouble following some
> of the hijack code for webrick:
> http://mid.gmane.org/20160511050451.GA23544@dcvr.yhbt.net
> 
> (Only speaking for MRI optimizations, I don't know other VMs)

Yes please!  If you have time.  Otherwise I can take care of it.

-- 
Aaron Patterson
http://tenderlovemaking.com/

-- 

--- 
You received this message because you are subscribed to the Google Groups "Rack Development" group.
To unsubscribe from this group and stop receiving emails from it, send an email to rack-devel+unsubscribe@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[-- Attachment #2: Type: application/pgp-signature, Size: 455 bytes --]