[Fedora-packaging] Re: satic libs package naming

[Patrice Dumas]

On Fri, Apr 20, 2007 at 01:42:59PM +0200, Ralf Corsepius wrote:
> 
> Of cause not .. I guess, I know what Pat's "special needs" are:
> Cross-distro packaging:
> 
> - Sharing binaries across different distros, e.g. running Fedora built
> apps on SuSE, Debian, Ubuntu and older RH based distros.

Exactly. Statically linked libc requires a recent enough kernel, this
is what is currently limiting the portability of the executables I
generate.

> - Storing binaries for long-term to be able to reproduce results long
> time after the distro they had been built on has been EOL'ed

In fact this doesn't work very well if I remember what was said in the
long thread I referred to.

> - Reproducibility of results by eliminating the impact of multilibs, tls
> etc.

That is not one of my needs, but some scientists need that. I am not 
convinced that statically linking is a definite solution, but it may
help.
 
> He tries to achieve this by statically linking apps. This works for some
> very simple apps (Pure numeric apps often qualify as such - Complex
> mathematical algorithms but very simple wrt. system resources), but
> doesn't work in general.

I agree with you that it doesn't work in general, but it works well for
me and I believe my case is very common within scientists: 
* use only mathematical libs
* have heterogeneous linuxes (both in time and in distros)
* want to run on all the computers in the lab
 
> The g77/gfortran issue he reported in recent days probably are related
> to this ;)

Not exactly, but indeed statically compiling allows to compile with a
compiler and run on a platform where it isn't supported. g77 ICE on
some of my codes, so I cannot recompile on some distros, but still can
run gfortan compiled codes on those distros. Static linking is not 
always needed, but in practice it is, for example when libgfortran isn't
available.

--
Pat