On Sun, Dec 16, 2012 at 02:01:00PM +0100, Jeremie Le Hen wrote: > Hi bapt, kib, > > On Sat, Dec 15, 2012 at 11:56:43AM +0100, Baptiste Daroussin wrote: > > On Sat, Dec 15, 2012 at 03:22:34AM +0200, Konstantin Belousov wrote: > > > On Sat, Dec 15, 2012 at 12:54:19AM +0100, Baptiste Daroussin wrote: > > > > Hi, > > > > > > > > Some of our binary are overlinked, the way we handle the linking > > > > doesn't help > > > > for that. > > > What do you mean there ? Do you mean that some libraries specified for the > > > linking stage of the final binary are not needed for the execution ? > > > > I mean some library are registrer in the binary with DT_NEEDED tag where > > they > > don't need to. > > > > > > > > > > > > > On proposition could be to use pkgconf > > > > https://github.com/pkgconf/pkgconf which > > > > is BSD license pkg-config implementation 100% compatible with > > > > pkg-config. > > > > > > > > What I propose is to create a new PCADD variable for the Makefiles. > > > > > > > > PCADD will invoke pkgconf to gather the libraries and the cflags for a > > > > given > > > > project. > > > > > > > > The second thing would be to create .pc files for all of our libraries. > > > > > > > > for example: > > > > usr.bin/fstat dynamic build is overlinked > > > And how this is better than just removing the unneeded library from > > > the Makefile ? > > > > In that case to statically build you need -lkvm -lutil -lprocstat but if you > > build dynamically you will only need -lproctast meaning you don't need to be > > directly linked to libutil and libkvm. This means a breakage of libutil will > > only have inpact on procstat and no more on fstat for example. > > I think this could be solved by an implicit linker script contained in > .so and .a files, pointing to the real libraries. > > We have already the SHLIB_LDSCRIPT variable to accomplish this for .so > files. It may be possible to do the same for .a files, though this > would require renaming the real archives to something like .a.0 (here > I'm just taking a similar scheme to the one used for DSO). > > As a matter of fact, libprocstat.a would contain: > > GROUP ( /usr/lib/libprocstat.a.0 /usr/lib/libkvm.a /usr/lib/libutil.a ) > > Note that libkvm.a and libutil.a could be linker scripts as well. > > Kib, do you see any problem to this proposition?
Wouldn't you need to completely rewrite the handling of the .a files in the share/mk ? I somewhat dislike the mere thought that .a is not an archive any longer. Does it make sense from the overhead and complexity POV, for such small goal ?
pgpxuDm949pH3.pgp
Description: PGP signature
