Re: gnome-socket API proposal
- From: Miguel de Icaza <miguel gnu org>
- To: hp redhat com
- CC: dereks kd-dev com, rmoya mail fujitsu es, zunino cli di unipi it, gnome-devel-list gnome org
- Subject: Re: gnome-socket API proposal
- Date: Tue, 14 Sep 1999 09:36:55 -0500
> Well yeah, but the overhead is only for a small set of applications. If
> you add it to libgnomeui, you make libgnomeui bigger and that impacts
> _all_ applications.
Not really. What does it mean the library is bigger? It means:
1. A few more symbols are included in the library for linking
2. It means that probably one extra page of coded is added.
Under a typical setup, this extra page does not even play a role.
until you touch the code on that page, the page is not even loaded
from the hard drive.
> That may not seem like much if it's just this sockets
> stuff, but we have to draw a line or we end up with all kinds of bloat in
> gnome-libs. Do you want to replace libgnomeui with a giant
> Bonobo-print-XML-glade library?
If I could get a way of linking those shared libraries, i would love
to create a gigantic library that reflects the *best* setup.
In SystemV they have libraries for everything. -lnsl, -lsocket. In
Linux we dump them all in libc. Do they have an impact? No, they do
not have an impact.
> What are the criteria we are going to use
> to ensure libgnomeui doesn't get any more bloated?
Again, what is "bloat"?
> I think the sensible criterion for inclusion in libgnome/libgnomeui is
> "almost all apps would potentially benefit from this code." This network
> code will only be used by maybe 10% of apps. gnome-dns is used less than
> that even. That's why I was proposing that we remove it, along with lots
> of other stuff (see libgnomeui/TODO).
but the networking stuff looks very very good. It sounds like even an
obvious thing to have in glib.
] [Thread Prev