resident buildworld

David Rhodus sdrhodus at gmail.com
Wed Mar 30 09:22:57 PST 2005


On Tue, 29 Mar 2005 15:42:26 -0800 (PST), Matthew Dillon
<dillon at xxxxxxxxxxxxxxxxxxxx> wrote:
> 
> :
> :On Tue, Mar 29, 2005 at 04:31:19AM +0000, David Rhodus wrote:
> :> On Tue, 29 Mar 2005 06:16:37 +0200, Joerg Sonnenberger
> :> <joerg at xxxxxxxxxxxxxxxxx> wrote:
> :> > On Tue, Mar 29, 2005 at 04:12:49AM +0000, David Rhodus wrote:
> :> > > I was thinking today about how we patch the vendor files during the
> :> > > buildworld process and the overhead added by running patch(1) all the
> :> > > time. Perhaps we should use resident(8) more during buildworld on
> :> > > program such as make, gcc, patch and other friends that are run
> :> > > several times.
> :> >
> :> > I think those tools are statically linked. If not, they should.
> :> >
> :> > Joerg
> :>
> :> Why do you think they should be static binaries ?
> :
> :Makes them less depending on the environment. We add NOSHARED for
> :the build and cross tool stage.
> :
> :Joerg
> 
>     Plus the startup time is a lot quicker... something that normally
>     does not matter unless you are running the same program over and
>     over again (which is what happens in a buildworld).
> 
>     resident does not make static binaries run any faster, so it would
>     not improve performance.
> 

The make, gcc, path that are run from the base userland before we have
created the bootstrap obj directory are dynamic.

-DR





More information about the Kernel mailing list