static libraries' policy

[Patrice Dumas]

On Tue, Nov 08, 2005 at 06:06:53PM +0300, Dmitry Butskoy wrote:
> During the review processes I've noticed that reviewers very differently 
> concern to presence of the static libraries in the devel packages. Some 
> of them force me to get rid of it, but some say nothing about it at all.

I, personally, think that we should be flexible on that matter. I can't see
what harm is done when a static library is shipped together with a dynamic
library. My tests show that when using -l the default for the linker is to 
pick the dynamic library (at least on i386). So a user wanting to statically 
compile has to use the library .a file full path, or add -static (correct
me if I'm wrong, I may have missed something). 

Therefore a user using static libraries in fedora is doing that on purpose,
so forcing him to bypass the rpm system and recompile by hand the library
seems a bit unusefull to me. Statically linked executables are still usefull 
because there may be run over a wide range of linux versions.

For some libraries the static library versions may be unusefull, but when
the library is of use to scientists or sysadmins, I think a static version 
should be provided, in case the user finds it usefull. They may be shipped
outside of -devel packages, but should be available with rpm IMO. I don't
think that shipping lapack or netcdf without a static library would be
wise and help anyone.

Of course libraries included in a package or fedora packages compiled 
against a static library when a dynamic library exist should be forbidden
but that's another issue and there is a consensus on that point.

--
Pat