[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Would an ABI change of apt for DDTP support still be accepted?

On Wed, Sep 27, 2006 at 11:42:35PM -0700, Steve Langasek wrote:
> On Thu, Sep 28, 2006 at 02:35:19AM -0300, Otavio Salvador wrote:
> > > On Wed, Sep 27, 2006 at 07:13:21PM +0200, Luk Claes wrote:
> > >> Would you still accept an ABI change of apt to support description
> > >> translations into etch?
> > > I gather that "ABI change" means an soname change?  In that case, no, sorry,
> > > I think it's too late in the release cycle to be changing this for apt.
> > I would like to ask you to review again your position. This code is
> > around since 3 years ago and in use on Ubuntu too. Are too few
> > packages that will need recompile.
> And yet the request comes as we should be preparing to feature-freeze apt
> *completely* for etch, not thinking about changes that require a recompile
> of all reverse-deps.

Right. I'm to blame here that I was overly cautious with putting new
code into libapt in unstable. 

There were no translations available on ftp.debian.org until end of
July (when aj did a one-time import) and without those the code was
not really testable for real-world use. When the translations were
importet and I asked for testing on debian-devel I got little feedback
on the actual code in experimental. 

That's why I considered it so late for uploading to unstable. I didn't
wanted to upload it without real-world testing because of the risk of
having to break the ABI yet again to fix mistakes in the code.

> BTW, I count 18 binary packages that would need a rebuild for this.  This is
> a decent-sized library transition in its own right.

We may have to recompile the rdepends of libapt anyway because of
(recent g++ upload 4.1.1ds1-14 has a g++ regression)

Upstream gcc bugreport:

Matthias is still waiting for a comment from upstream on this. It
maybe enough to recompile apt with the current g++, but it maybe that
the only save option is to change the soname and recompile a rdepends.

> > There's no API changes from APT side so just binary NMUs are enough
> > AFAIK.
> So what is this ABI change that doesn't involve API changes?

There is a API change involved. But it is backwards compatible so a
recompile will be good enough. To make use of the translated
descriptions the applications needs to be changed though. Patches are
available for aptitude, python-apt, synaptic, libapt-front (0.3). 

I hope this helps and I'm sorry for the bad timing with this request :/


Linux is not The Answer. Yes is the answer. Linux is The Question. - Neo

Reply to: