< Previous by Date Date Index Next by Date >
< Previous in Thread Thread Index Next in Thread >

Re: [reSIProcate] gperf updates? Remove some hackery.


The performance gain would be very slight. I feel that there is enough value in being able to regenerate the table when we add a header/param/method to make me want to stick with a gperf-generated hash function. I don't think that doing it every build is worthwhile though.

Best regards,
Byron Campen


I always been a fan of removing the gperf dependency all together. If you are willing to use just a little bit more memory for the tables, it is pretty trivial to get more or less any reasonable hash function to not collide for some size table. I suspect that making the table that much larger would be a drop in the bucket of overall memory use in resip.

On Oct 12, 2007, at 7:56 PM, Byron Campen wrote:

        I think Jason is saying that most embedded systems won't have gperf
3. I'm saying that the answer is to not build new hash functions
using the regular build. Put it aside as a separate script that can
be run independent of the regular build, and just have all the
generated files live in the repository. This would remove a build
dependency that is continually giving newcomers grief (always a good
thing), cut down on the complexity of the build system a little (also
good), and probably help performance a tiny bit (I imagine gperf will
generate slightly better code than our hacked-up scripts will).

Best regards,
Byron Campen

I must be missing something. What does changing gperf have to do
with embedded builds?

Thanks
A

-----Original Message-----
From: Byron Campen <bcampen@xxxxxxxxxxxx>
Date: Fri, 12 Oct 2007 21:21:30
To:Jason Fischl <jason@xxxxxxxxxxxxxxx>
Cc:"Alan Hawrylyshen" <alan@xxxxxxxxxxxx>,       "resiprocate-devel
resiprocate-devel" <resiprocate-devel@xxxxxxxxxxxxxxxxxxxx>
Subject: Re: [reSIProcate] gperf updates? Remove some hackery.

        I would like to point out that generating fresh hash functions on
every build is of limited value, especially considering the hoops you
have to jump through to add a method/header/parameter anyway.

Best regards,
Byron Campen

The only thing i'll point out is that this will break the build on
lots of embedded systems. What does it really buy us?

On 10/9/07, Byron Campen <bcampen@xxxxxxxxxxxx> wrote:
        I'm all for it.

Best regards,
Byron Campen

I've noticed that the newer versions of gperf permit case-
insensitive
hashes to be generated. I find this oddly satisfying given that the
GNU/FSF position on this previously was 'over my dead body'.

What are peoples' feelings around updating our build process to
leverage gperf 3.X and remove the hackery that edits the output of
the hash function generators?

Thoughts?

Alan

Alan Hawrylyshen
a l a n a t p o l y p h a s e d o t c a


_______________________________________________
resiprocate-devel mailing list
resiprocate-devel@xxxxxxxxxxxxxxx
https://list.resiprocate.org/mailman/listinfo/resiprocate-devel




_______________________________________________
resiprocate-devel mailing list
resiprocate-devel@xxxxxxxxxxxxxxx
https://list.resiprocate.org/mailman/listinfo/resiprocate-devel

_______________________________________________
resiprocate-devel mailing list
resiprocate-devel@xxxxxxxxxxxxxxx
https://list.resiprocate.org/mailman/listinfo/resiprocate-devel