Re: NetBSD master CVS tree commits

[Jonathan Stone <jonathan%DSG.Stanford.EDU@localhost>]

In message <199708270019.RAA09717%lestat.nas.nasa.gov@localhost>,
   Jason Thorpe writes:

>On Wed, 27 Aug 1997 09:44:59 +1000 
> Simon Burge <simonb%telstra.com.au@localhost> wrote:

> > I think this is semi bad for netbsd/pmax, since we're still using
> > a really old ldconfig whose source isn't in the tree.

The ldconfig is one I got from Manuel Bouyer.. I understand it's a
stock GNU libc ldconfig, and as such it's *newer* than the a.out
ldconfig in the NetBSD tree. The source should be in the same place as
the binutils source for the snaphsot.

The latest snapshot contains a newer ldconfig, btw.

>NetBSD/pmax won't be any better or worse for this change.  When NetBSD's
>MIPS port start using the already-in-tree ld.elf_so, ldconfig won't be
>needed,

Exactly so. Though it might be a noticeable performance improvement
for exec startup times on old, slow, machines like 12MHz decstations.
I agree the difference is not perceptible on an Alpha; exec costs on
pmaxes are noticeable even to mere humans. Try
        foreach i (/usr/bin/*)
                file $i
        end


>since Elf isn't even supposed to have such a notion as a hints
>file, like a.out libs do.

Jason, could you clarify? Is this part of the ELF spec, or a NetBSD
design decision?  The GNU ELF shared-lib tools *do* use ldconfig and a
hints file...