[prev in list] [next in list] [prev in thread] [next in thread] 

List:       kde-devel
Subject:    Re: Message System
From:       Ashley Winters <jql () accessone ! com>
Date:       1998-07-20 8:44:25
[Download RAW message or body]

On Mon, 20 Jul 1998, Steffen Hansen wrote:

> On Sun, 19 Jul 1998, Martin Konold wrote:
> 
> > On Sun, 19 Jul 1998, Sven Radej wrote:
> > 
> > > I just wanted to point out that I don't like the idea to link small
> > > tools like kbiff to mega-lib like CORBA. Not that I have a simple
> > > solution...
> > 
> > What is the problem about linking a tiny app to a MEGA lib dynamically?
> > We could even link kdelib to corba by default.
> 
> Couldn't it cause the program to start up more slowly on systems with
> a slow runtime linker? I saw somewhere that Linux was critizised for
> having a not so efficient linker.

All I know is that Linux is slow. I have no idea if anything else is
faster...

> Is the startup time a function of the size of the dynamic libs or only the
> number of them?

As I understand it, the size of the dynamic lib only determines the
disk-access time. Each dynamic lib has an very large but fixed
minimum overhead (I'd guess around a millisecond) for the mmap().

The real penalty comes from the symbols. Qt-1.3 has 111k worth of them,
kdecore+kdeui have another 93k, and most of that is slurped in at startup.
Compare that to only 18k for Linux libc. C++ symbols take up ~3x the space
of C symbols because of the detailed name-mangling g++ uses. But it's the
price you must pay for C++.

A 10M dynamic lib with only 5 symbols should load in about the same time
as a 50K dynamic lib with the identical 5 symbols. I'd prefer the 50K lib
nonetheless.

Ashley Winters

> greetings,
> --------------
> Steffen Hansen                           | Dave, put down those Windows 
> email: stefh@mip.ou.dk, stefh@imada.ou.dk| disks, Dave, Dave! 
> URL:   http://www.mip.ou.dk/~stefh       | I'm sorry HAL.

[prev in list] [next in list] [prev in thread] [next in thread]