[quagga-dev 8677] Re: trivial quagga patches

Stephen Hemminger shemminger at vyatta.com
Thu Apr 7 16:35:13 BST 2011


On Thu, 7 Apr 2011 10:23:15 +0100 (BST)
paul at jakma.org wrote:

> On Wed, 6 Apr 2011, Stephen Hemminger wrote:
> 
> >      bgpd: inline lock/unlock
> 
> > --- a/bgpd/bgpd.h
> > +++ b/bgpd/bgpd.h
> 
> > +static inline void
> > +bgp_flag_set (struct bgp *bgp, int flag)
> > +{
> > +  SET_FLAG (bgp->flags, flag);
> > +}
> 
> I have to say I don't like these inlining patches. Unless they come 
> with benchmarks across a diverse range of setups showing actual 
> performance improvements, there's as much reason to believe they hurt 
> performance as improve it - and I won't take them.

Smaller code is faster code, 

> It'd be *much* better to supply patches to the build system to allow 
> multi-unit compiling. Inlines in headers can /never/ be uninlined, 
> but the compiler can always inline code if you tell it to or, better, 
> give it enough information to make intelligent decisions (e.g. using 
> previous-profile directed compilation).

Gcc multi-unit compiling is broken in many versions.

-- 



More information about the Quagga-dev mailing list