[Xorg] The big multiconsole nasty

[Eric Anholt]

On Tue, 2004-07-06 at 09:17, Adam Jackson wrote:
> > <snip>
> > 1) The XFree86 loader: One should assume the OS can dynamically load and
> > link modules and that the OS knows more about this than the X server. In
> > the past when very low level handling of object modules changed the X
> > Server blew up spectacularly and required surgery to very arcane pieces
> > of code none of which would have been necessary if the underlying system
> > services had been used. I might add being able run a debugger on a
> > running process would appear to be a novel concept ;-)
> 
> I have to agree.  Right now we've got both a libc thunk layer and a 
> reimplementation of libdl in the server, all for a marginal (and I hear, 
> rarely realized) gain in module portability.  I have trouble justifying that.  
> I'm currently working at fixing the drivers so the libdl loader is usable 
> again.

I was talking with a NetBSD guy about this at USENIX.  He said he used
the module portability all the time (many vendors provide binaries
compiled for x86 Linux, and x86 NetBSD just is not going to get them),
and I realized that I often also copy drivers between my OSes.  But we
don't need the XFree86 loader to do that -- the elf modules and the libc
layer should be sufficient.  I think we should think seriously about how
much overhead it is to keep the existing libc wrapper while using an elf
loader so we can have working debugging.  It probably isn't that bad.

-- 
Eric Anholt                                eta at lclark.edu          
http://people.freebsd.org/~anholt/         anholt at FreeBSD.org