Date: Fri, 18 Mar 2005 14:43:13 -0500 From: David Schultz <das@FreeBSD.ORG> To: Daniel Eischen <deischen@FreeBSD.ORG> Cc: sobomax@portaone.com Subject: Re: cvs commit: src/lib/msun/i387 fenv.c fenv.h Message-ID: <20050318194313.GA50903@VARK.MIT.EDU> In-Reply-To: <Pine.GSO.4.43.0503181353530.661-100000@sea.ntplx.net> References: <20050318.114956.74684448.imp@bsdimp.com> <Pine.GSO.4.43.0503181353530.661-100000@sea.ntplx.net>
next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, Mar 18, 2005, Daniel Eischen wrote: > On Fri, 18 Mar 2005, Warner Losh wrote: > > > > Yeah, well, that was your karmic punishment for touching the internals > > > of FILE. Serves you right! > > > > I didn't touch the internals of FILE. I just helped clean up the > > mess. :-( > > That was me, but as I recall we ended up keeping binary compat > by reusing an existing field and extending it (green's idea I > believe). I think the real pain came when we tried to fix > the predefined stdio, stdin, and stderr descriptors to prevent > ABI breakage for future FILE changes. > > Has anyone experimented with symbol versioning yet? I recall > someone wanting to do that (dfr?). I looked into it, but I don't have time to implement it. Basically, 90% of the support is already there, and it would not be hard to finish for someone who understands the rtld internals already (I don't). I successfully created a working versioned math library by simply making a linker spec and passing the --version-script option to ld. All that's needed, I think, is to teach the runtime linker to check the version fields. Then we just make a bunch of specs for the libraries we want to version.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20050318194313.GA50903>