Too bad we cannot preserve original 1.0 semantics and have -O turn off
Objectspace and +O enable it.  Then we can default to -O.  -O really
just keeps getting worse and worse (comparatively to not having it
enabled) and people generally do not realize they can disable it.  So
I have come around on this and also think it should be off by default.

-Tom

On 10/28/07, Charles Oliver Nutter <[EMAIL PROTECTED]> wrote:
> So, it's time to bring up this question. ObjectSpace makes JRuby so much
> slower, I think it's time to consider disabling it by default.
>
> For example, here's the fractal benchmark, which generates a text-based
> mandelbrot:
>
> With ObjectSpace: Ruby Elapsed 45.967000
> Without ObjectSpace: Ruby Elapsed 4.280000
>
> We have also made modifications to allow ObjectSpace.each_object(Class)
> to work even when ObjectSpace is disabled, by tracking an in-memory
> graph of classes and subclasses. This allows the most visible use case
> for ObjectSpace--test/unit--to continue working until we can patch it to
> not use ObjectSpace.
>
> So I have a few questions here:
>
> - Could we now use -O to *enabled* objectspace?
> - Would a requirable library be better? -robjectspace to enable the
> "extended" features for full object graph walking?
> - Do we know of anything else that needs ObjectSpace? Can we get those
> things patched?
> - Do we need to get buy-in from ruby-core for something like this? Can
> we convince them it's in everyone's best interest if ObjectSpace becomes
> an extension?
>
> Over time, our perf numbers have shown ObjectSpace as more and more of a
> penalty. I'm tired of typing -O, but I'm not running a lot of real-world
> stuff day-to-day. What do you all think?
>
> - Charlie
>
> ---------------------------------------------------------------------
> To unsubscribe from this list please visit:
>
>     http://xircles.codehaus.org/manage_email
>
>


-- 
Blog: http://www.bloglines.com/blog/ThomasEEnebo
Email: [EMAIL PROTECTED] , [EMAIL PROTECTED]

---------------------------------------------------------------------
To unsubscribe from this list please visit:

    http://xircles.codehaus.org/manage_email

Reply via email to