Re: [wmii] wmii-3 configuration - in Ruby

On Thu, Jun 08, 2006 at 09:13:00AM +0200, Anselm R. Garbe wrote:
>On Wed, Jun 07, 2006 at 05:15:20PM -0400, Kris Maglione wrote:
>> On Wed, Jun 07, 2006 at 11:42:02AM +0200, Anselm R. Garbe wrote:
>> >libixp is much smaller than the complete lib9p* dependencies of
>> >Plan9 (which are about 50kSLOC including lib9, if not more). And
>> >libixp/client.c keeps the connection between dial and hangup, I
>> >don't think the client part is a bad API of libixp (however the
>> >server side needs polishing, I always agreed to this).
>> >
>> >A sloccount of libixp tells me it consists of 1244 lines, for
>> >everything (socket handling, server/client side, protocol layer,
>> >- only dependency are 173 lines of libcext).
>> >This is 1.5kSLOC vs. 60kSLOC...
>> Yes, I considered mentioning this after I posted. Like I said, though, I
>> already gave up writing libixp bindings to ruby. The interface for
>> ixp_client is sane, I'll admit, but I'd rather rewrite it in ruby whan
>> write bindings. It would actually be easier. I think, though, that I'd
>> rather have lib9pclient bindings, since it fully supports authentication. I
>> think that it's somewhat slower, though, than libixp. I haven't actually
>> benchmarked the libs, just the executables (wmiir and 9p), and wmiir is ~3x
>> the speed.
>
>This might be related to lib9, because nearly every libc call of
>Plan 9 apps ends is wrapped in the Unix environment by lib9.
>Anyway, in 9base I linked everything statically and the
>performance gain was fantastic (however 9base didn't contained
That may be true, but I doubt the impact is that terrible. Unix system calls
are expensive as it is and I can't imagine the impact of a function call
adding that much. The linixulator, for instance, wrapps all system calls with
no discernable performance impact. Perhaps later I'll try statically linking
it and see what the results are.

-- 
Kris Maglione
Things greater than the same thing are greater than each other.  -L. Carroll (Sylvie and Bruno)