Re: GMP transition: 4.3.2 to 5.0.1?
On Sat, Mar 19, 2011 at 13:23:41 -0500, Steve M. Robbins wrote:
> So, given that the incompatibility between GMP 4 and 5 is removing a
> couple of public functions, you believe it is safe to have the shared
> libs coexist in the archive?
>
It looks that way.
> That's fine. Can you help me work out the specific steps I should
> take now?
>
> In your previous message, you suggest to re-introduce gmp 4.3.2 as a
> separate source package (e.g. gmp4), building *only* the libgmp3c2
> binary package.
>
Right.
> Raphael Geissert has requested also to change gmp to build libgmp-dev
> as a real package. I would also make it provide libgmp10-dev because
> that package name was used in experimental and a few packages are now
> using it.
>
> In addition, I made the new -dev package also provide libgmp3-dev
> because of the strange situation of the Haskell compiler ghc [1]. My
> intention was for that to be temporary and remove it once ghc got
> bootstrapped everywhere. In another of your messages you say
>
> Again, the -dev package needs to keep providing libgmp3-dev
> anyway, so changing the build-deps is unnecessary churn.
>
> What did you mean by this? Do you mean that the new -dev package
> needs to continue providing libgmp3-dev forever? Or did you mean that
> since it is presently providing it, there is no need to change all the
> build-deps at once; just let things alone until such time that we
> remove this "provides"?
>
The latter. There's no hurry to drop the Provides, and keeping it
reduces the disruption in the archive since it means the move from
libgmp3-dev to libgmp{,-10}-dev can be staged over a longer period.
> Assuming you meant the latter, here is my understanding of the
> next steps:
>
> 1. Upload gmp4, as described above.
> 2. Upload gmp introducing libgmp-dev as the real package, providing
> virtual packages libgmp3-dev and libgmp10-dev.
>
> Is this accurate? Anything else?
>
I don't think 2 is necessary, or at least not right now. It's fairly
independent as far as I can tell, and it's a cosmetic issue more than
anything else.
As I mentioned in my previous mail it may be a good idea to investigate
linking the libgmp's with -Bsymbolic to ensure that internal symbols are
looked up in the right library. Not a blocker though, so we can
probably leave that aside until/unless issues pop up if you prefer.
Thanks for your work, and let me know if anything's unclear still.
Cheers,
Julien
Reply to: