Re: RFC: SDL and X static extension libraries re-revisited
>> Branden Robinson <branden@debian.org> writes:
> The difference is, with a static libXxf86dga (et al.), no
> *applications* will break when this happens. All applications,
> plugins, and libraries based on libSDL will have a shlibdep on it
> anyway. If/when libSDL changes the way it deals with the static X
> extension libraries, it probably *should* bump its soname.
With a static library, if the API changes (read, functions are removed
or signatures are changed) there's a changce applications won't compile
anymore, which is just as bad from our POV. With a shared library, we
at least have the opportunity to notice that earlier. We can have a
test to check the existance (and signatures) of all the entrypoints in
a given release. In the next release one of the entrypoints is
changed, the test fails. A library with a new soname is produced.
Following policy, the package name changes. If the old package
dissapears but there's stuff in testing which still depends on it, we
eventually notice that ("why the $#%^ isn't XFree86 6 making it to
testing?" "oh, it breaks Bar") and recompile/fix/whatever. With the
static library approach we are just releasing (more) stuff that won't
compile out of source. Adding entrypoints isn't as bad, it involves
changing one line in a shlibs file (hint, hint). The only discussion
which I've followed more or less closely is the one relating Xv, and
it's about adding entrypoints, not about changing existing ones,
AFAICR.
A candybar goes to the one that lists all the assumptions I'm making
here...
--
Marcelo | "When it's time to stop living, I will certainly make
mmagallo@debian.org | Death my number one choice!"
| -- (Terry Pratchett, The Last Continent)
Reply to: