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

Re: ABI breakage in libmysqlclient15off_5.0.24-1 - binNMU possible?




On 2006-08-25 Steve Langasek wrote:
> On Fri, Aug 25, 2006 at 10:25:25PM +0200, Christian Hammers wrote:
> 
> > MySQL introduced a ABI incompatibility in libmysqlclient15off_5.0.24-1
> > which already caused a lot of segfaults and crashes esp. with the Perl
> > bindings. A patch is available at least.
> 
> > Is it possible to schedule a binNMU rebuild for all packages depending on
> > libmysqlclient15off without a package name change and just change
> > the line "dh_makeshlibs -V"?
> > I guess not as it was necessary back in 2006-03-15 at the last ABI
> > breakage?
> 
> What do you mean, 'change the line "dh_makeshlibs -V"'? A binNMU can't
> include any changes to the source, but this line shouldn't appear in the
> packages that *depend* on libmysqlclient15off anyway (or at least it's not
> relevant to this change).
Sorry for being unclear, the dh_makeshlibs command is of course in my source
package which I will have to upload anyway to revert the accidental ABI
change.

> Please fix the library first -- ideally with a package name change or by
> restoring the ABI,
Hm, yes, that "or" was exactly my question, in cases where an unintended ABI
change has been around only for a week in unstable, is it enough to upload
a fixed version and rebuild everything?

> but at least with a bumped shlibs -- and then we can binNMU the packages 
> depending on it as needed.
I upload 5.0.24-2 once it has been built here. It contains *no* package name
nor soname change but just a fixed libmysqlclient.so.15 and an updated shlibs
file. And a NEWS entry for apt-listchanges to warn people to rebuild
binaries that were compiled locally last week. 

bye,

-christian-



Reply to: