Bug#646095: Usage of "private" symbol

[Julien Cristau <jcristau@debian.org>]

On Fri, Oct 21, 2011 at 14:22:00 -0200, Gustavo Noronha Silva wrote:

> On Fri, 2011-10-21 at 18:09 +0200, Philipp Kern wrote:
> > On Fri, Oct 21, 2011 at 01:57:42PM -0200, Gustavo Noronha Silva wrote:
> > > This is caused by midori using a symbol that was only exported for being
> > > used in the test runner. I can't see how a rebuild would fix it, are you
> > > sure it did?
> > 
> > This means that you just introduced an ABI break and are obliged to bump SONAME
> > and this would've been a transition that should've been asked for and handled
> > properly.
> 
> You are absolutely right, I am sorry about the transition, I wrongly
> assumed it was a good time to upload, I forgot about the addition of new
> packages and thought it would not block anyone, I'll be sure to ask in
> the future, and I'll do what I can to try to avoid too much disruption.
> 
> Regarding the ABI breakage, it's unfortunate that webkitgtk had to use
> this strategy of exporting some symbols that were not meant to be used
> by applications to be able to run the tests, but they were always
> clearly marked, left out of the public headers and you can see that
> midori even carries its prototype mentioning the fact that it's
> "unstable API". I don't think a soname bump is a good idea for changes
> in such symbols.
> 
At the very least the new webkit will need to declare Breaks on midori
(<< 0.4.1-2) or something like that.  And midori needs to get fixed up,
by the sound of it.

Cheers,
Julien