Well that's weird.  Now it's all working in MRI.  I'm not sure which
of my changes helped, or if I sleep-walked last night after one of my
lobes declared a coup.

Here's what I did with CurlFFI:

https://github.com/aitrus/curl_ffi/commit/3e72173c9f60238f855c6e23f54e553ad7da3608

And here's what I did in StreamlyFFI:

https://github.com/aitrus/streamly_ffi/commit/48568e7792cd7bc8a300d0fac99238799364c374

I need to go back and remove my frantic use of ".dup" and see what the
effect is.  I should probably walk back my changes to see where it
went right.

Scott

On Nov 11, 11:09 pm, Scott Gonyea <gon...@gmail.com> wrote:

> Interesting.  I used JRuby (1.5.3) and the ffi gem (0.6.3 /
> rubygems.org).  Works just peachy.  I ran it at 50 and then 100
> iterations.  Everything seems to get along.
>
> sgonyea@skg:~/Sites/gsolr$ ruby bench.rbhttp://toddler.tomkittens.com:8080/solr
> "hi"
> Rehearsal ------------------------------------
>  GSolr
>   4.626000   0.000000   4.626000 (  4.626000)
>  RSolr
>   9.435000   0.000000   9.435000 (  9.436000)
> -------------------------- total: 14.061000sec
>
>        user     system      total        real
>  GSolr
>   3.523000   0.000000   3.523000 (  3.523000)
>  RSolr
>   8.903000   0.000000   8.903000 (  8.903000)
> sgonyea@skg:~/Sites/gsolr$ ruby bench.rbhttp://toddler.tomkittens.com:8080/solr
> "hi"
> Rehearsal ------------------------------------
>  GSolr
>   8.134000   0.000000   8.134000 (  8.134000)
>  RSolr
>  18.710000   0.000000  18.710000 ( 18.710000)
> -------------------------- total: 26.844000sec
>
>        user     system      total        real
>  GSolr
>   6.805000   0.000000   6.805000 (  6.805000)
>  RSolr
>  17.557000   0.000000  17.557000 ( 17.557000)
>
> Scott
>
> On Nov 11, 11:01 pm, Scott Gonyea <gon...@gmail.com> wrote:
>
>
>

> > Strange.  I replied in my e-mail, but it didn't post.  Hoping it
> > wasn't plugged up in Google somewhere.  Here's what I wrote:

>

> > I'm pretty sure it does.  I ran into another issue, while using the
> > FFI from Github:

>

> >https://gist.github.com/673816

>

> > I'm checking the values, all along the way, and I'm not handing it any
> > nils, but it won't believe me. The ffi 0.6.3 gem won't complain about
> > this (but it will blow up at a few spots).

>

> > Sigh :(  Thank you very much, for your help.

>

> > Scott

>

> > On Nov 11, 10:21 pm, Wayne Meissner <wmeiss...@gmail.com> wrote:

>

> > > Thats interesting.

>

> > > Does curl use threads internally?  If so, build the ffi gem from
> > > github and try that.  Earlier versions don't deal well with callbacks
> > > coming from non-ruby threads.

>

> > > Also try running it with JRuby.  It handles threads a lot better.

>

> > > On 12 November 2010 15:25, Scott Gonyea <gon...@gmail.com> wrote:

>

> > > >https://gist.github.com/673753

>

> > > > That will show you the script I'm running, the gdb backtrace, and ruby-
> > > > prof (if that helps at all).  I basically was benchmarking my library
> > > > with the one I started branching from.  It's all good, until I start
> > > > running a whole lot of iterations.  I made some changes, which seems
> > > > to have made it less frequent... until I up the number of iterations.

>

> > > >https://github.com/aitrus/curl_ffi/commit/99a231aea24b2d7d63bccaa5c23...

>

> > > > That was the one fix.  This GSolr junk of mine uses StreamlyFFI and
> > > > CurlFFI:

>

> > > > StreamlyFFI:https://github.com/aitrus/streamly_ffi
> > > > CurlFFI:https://github.com/aitrus/curl_ffi

>

> > > > Scott