Re: GMP transition: 4.3.2 to 5.0.1?
On Fri, Mar 18, 2011 at 10:24:15 +0100, Julien Cristau wrote:
> Well I tried to mention some ways this was suboptimal in my previous
> mail. Probably too late to do anything about it right now though, other
> than decide if we want to revert.
OK, I might see another way out. As far as I can tell, the
incompatibilities introduced in gmp 5 are the removal of mpn_bdivmod and
mpn_neg_n, and the rest of the functions should stay compatible between
gmp 4 and gmp 5. If that's not true, please do tell me, as the below is
based on that assumption.
So it *looks* like having libgmp.so.3 and libgmp.so.10 coexist in an
address space should be safe (although it'd probably make sense to have
both libraries linked with -Bsymbolic to be extra careful). We can then
re-introduce gmp 4.3.2 as a separate source package (e.g. gmp4),
building *only* the libgmp3c2 binary package. Then get both libgmp3c2
and libgmp10 in testing for a while, to avoid a massive transition of
all reverse deps at once. Then at a later point once everything's
rebuilt and transitioned to testing, we can get rid of 4.3.2.