Re: LuaSocket 3.0 ?

[Matthew Wild <mwild1@...>]

On Thu, Jan 22, 2009 at 11:55 PM, Diego Nehab <diego@tecgraf.puc-rio.br> wrote:
> Hi,
>
>> The code appears straighforward, and I'm happy to submit a patch if
>> that is what is required to get this into the next LuaSocket release.
>
> Does this show up as a performance bottleneck in your
> software? It would be nice to see the numbers.

It would indeed. In fact this is a hard thing to benchmark, since the
effects would hit during allocation and garbage collection, both of
which are difficult to track from Lua, and it would be some work to
test from C I believe.

Basically we are trying to keep all allocations and table-lookups to a
minimum, and this is just one of the things that got highlighted,
happening in the tightest loop in our code. It would be good to prove
the issue conclusively either way.

> If there is a
> significant improvement, I don't see a reason not to add the
> feature. The arguments would have to come in as two
> additional at the end of the list, so as not to kill
> performance.
>

Agreed.

Matthew.