Re: should we ship linkable gmp/mpfr/mpc or hide them?

[Alistair Crooks <agc%pkgsrc.org@localhost>]

On Sat, Jul 16, 2011 at 11:08:06PM +0000, David Holland wrote:
> On Sun, Jul 17, 2011 at 06:44:44AM +1000, matthew green wrote:
>  > i'm trying to decide if we should ship gmp/mpfr/mpc as something
>  > private to gcc4.5 or first-class part of the base system.
> 
> My inclination is to expose gmp but not the others. This is supported
> by the following:
> 
>    - gmp is used by 49 packages in pkgsrc
>    - mpfr is used by 7, of which 5/7 are gcc packages
>    - mpc (mpcomplex) is used by 3, of which 3/3 are gcc packages
> 
> gmp is therefore the only one with any substantial payoff;
> furthermore, it's used by kdeutils[34] so will be comparatively widely
> installed.
> 
>  >    - if pkgsrc has never versions, we have potential weird
>  >      issues at link or run time for binaries
> 
> This should not be any more of a problem than it is for other builtin
> packages with libraries.
> 
> also, gmp doesn't rev very often.

I'm against installing gmp in base (and all the others as well).  The
fact that other packages use them is neither here nor there - it's a
matter for pkgsrc, not base - and the builtin recognition should just
dtrt with versions.  Where I have problems is that nothing else needs
them in base, and I'm still unsure why a compiler needs
multi-precision arithmetic.  To me, this is just bloat, is not
required, wanted, desired or even "might be nice if we had it".

To add to all of this, the license is GPLv3.

I'd be much more interested in making a layer on top of
(public-domain) libtommath which could do both gmp and openssl-style
bignums.  I have started on this, but could do with some help.  Anyone
interested?

Thanks,
Alistair