On Fri, Nov 22, 2002 at 09:27:40AM +0100, Martin Husemann wrote:
> On Fri, Nov 22, 2002 at 05:17:00PM +0900, Bang Jun-Young wrote:
> 
> > BTW, what advantages can I get from separate tool and embedded data into
> > binary, compared to what I'm proposing?
> 
> Only one file,

A win over two files.

> first time startup is as fast as later ones,

You should have a binary prebound before you run it, so this is not
an advantage.

> startup time is (probably marginaly) faster (a bit more work inside the binary, but one
> namei less).

Yes, it is. But I don't think an additional namei call will slow down
loading considerably.

> And it's completely optional, even on a per-binary basis
> (typical scenario could be: all the base system is prelinked, plus binaries
> where it helps a lot [mozilla] are prelinked).

My proposal is also completely optional.

The most serious problem with what you're explaining is that it's even
more difficult to implement while it doesn't give better performance or
convenience. Simply adding prerelocation data to the end of a binary
is not allowed, since it's likely to break ELF semantics. In order to
avoid it, you may have to modify a number of places in the binary. 

Jun-Young

-- 
Bang Jun-Young <junyoung@netbsd.org>