From: Xavier Shay <xavier-list@...>
Date: 2011-06-03T04:32:00+09:00
Subject: [ruby-core:36700] Re: [Ruby 1.9 - Bug #3924] Performance bug (in require?)



On 31/05/11 3:55 PM, Xavier Shay wrote:
> Two other options that maybe you will like:
> 1) store some metadata against entries in $LOADED_FEATURES to indicate
> whether they have been cached. rb_ary_push will not set this, so a quick
> scan can be made of $LOADED_FEATURES on require to see if a recache is
> needed
> 2) On require, recache if the count of $LOADED_FEATURES does not match
> the count of the cache. This isn't 100% correct --- you could rb_ary_pop
> then rb_ary push to work around it --- but perhaps if more acceptable?
>
> I think #2 if the limitation is acceptable. Items added by rb_ary_push
> would be located on the filesystem if possible (such as "./t" in your
> example), other wise left as is (such as "enumerator.so").
I have put together a proof of concept of this:
https://gist.github.com/8bb23db32d861cbe952f

It removes the need for both lib/enumerator.rb, and the changes I made 
to loading mkmf in ext/exmke.rb

I still need to test it with amalgamite (was on a plane so didn't have 
internet to get it).

If you think the limitations are ok, I will work it back into my other 
patch set.

Also, I note a change to load.c has just gone into trunk related to 
performance. I will add this to my list of benchmarks in the next few days.

Cheers,
Xavier