This is the mail archive of the
guile@cygnus.com
mailing list for the Guile project.
Re: -fpic (Well, that's interesting.)
Telford Tendys <telford@eng.uts.edu.au> writes:
> > > [not using PIC code]
> > Doesn't that end up causing the library to be loaded for every process?
> At the moment, yes it does but that's a trade-off that some people might
> consider reasonable if it buys them as much speed as was previously noted.
Yeah, it'd be startup time versus runtime performance. Short-running
programs would notice the former; long-running emacsen and web
browsers and data crunchers would notice the latter.
> At some future stage, the ld.so program might be smart enough to recognise
> cases where the non-pic library gets loaded into the same memory zone each
> time it gets used so it can overlap those code pages, maybe this is wishful
> thinking.
The University of Utah did some experimentation with a dynamic-linker
server (actually, an "active object server which manages executable
programs and their components", providing linking services, program
analysis, and "custom application construction") on Mach in their OMOS
project.
ftp://mancos.cs.utah.edu/papers/omos.html (abstract & link to postscript)
http://www.cs.utah.edu/~mecklen/omos/iwooos.html (html version)
The one Utah person I've talked to about it put a positive spin on it,
but the work was still under way at the time. I don't know what's
become of it since, or if they uncovered any serious drawbacks. Maybe
they've folded some of it into their Flux/Fluke work?