Josh Stone píše v Út 05. 03. 2013 v 09:44 -0800:
On 03/05/2013 07:59 AM, Adam Jackson wrote:
> On Mon, 2013-03-04 at 14:07 -0800, Josh Stone wrote:
>
>> So given that this library's use is pretty well contained, might it be
>> OK to go ahead and update in F18?
>
> Yeah, that's fine.
>
> In the future, consider following the glibc pattern of fixing the soname
> for all but truly-world-breaking changes, and using symbol versions to
> annotate API additions. That way a package that uses an API introduced
> in dyninst 8.2 will get an rpm Requires for foo.so(dyninst-8.2)(64bit),
> which will make yum automatically search for a sufficiently new dyninst
> package without breaking the soname.
Is that feasible for C++ APIs? I mean, it might be possible if you're
*really* careful about hiding class changes, but this project is not
structured that way.
it is, see eg. the wxWidgets library, they are really good in that
> Minor numbers really do not belong in sonames, for this exact
reason.
> Every soname string is essentially a unique major version number.
Well, "minor" is relative, and upstream is consciously choosing not to
preserve ABI in this release, so the soname change is appropriate.
Dan