Finalizers defined via define_finalizer are a huge performance hit

[Asmod4n]

Am writing a wrapper for https://github.com/zeromq/czmq and was wondering why it was 20 times slower than https://github.com/methodmissing/rbczmq.

The Culprit is ObjectSpace.define_finalizer, when i free the memory manually its only 2 times slower.

Should this be "fixed" in ruby-ffi or in ruby itself?

[Matijs van Zuijlen]

Hi Asmd4n,

Can you give a small code sample that demonstrates the difference? I really need
to get started on memory management for my own FFI-based gem, so I'm interested
to know what works or doesn't work.

Regards,
Matijs

--
Matijs

[Asmod4n]

Have made a small gist for it.

https://gist.github.com/Asmod4n/ccb86462a727061801de

[Matijs van Zuijlen]

Hi Asmod4n,

That is a big difference. My first guess would be there is significant overhead
in the define_finalizer mechanism. I'll be investigating this further.

On 30/09/14 17:58, Asmod4n wrote:
> Have made a small gist for it.
>
> https://gist.github.com/Asmod4n/ccb86462a727061801de
>
> Am Dienstag, 30. September 2014 09:40:41 UTC+2 schrieb Matijs van Zuijlen:

> [...]

>
> Can you give a small code sample that demonstrates the difference?
>

> [...]


--
Matijs