libaio-0.3.106-3.2.i386 seems to contain two copies of the library and from the spec file this seems to be deliberate, the library is built twice, once with the old soname, once with the new. I don't see why it's necessary to build and ship the library twice like this: wouldn't it be sufficient to simply build a stub library with the old SONAME which depends on the library with the new soname? But regardless, the library built for backwards-compatibility should be shipped in a different package so that it only gets installed when necessary/desired.
See this bug: https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=133253 The library needs to be built twice.
I don't see how that addresses either of my points. 1) why it is not simply sufficient to have a stub library with the old SONAME (or even a symlink) given that the libraries are otherwise functionally identical 2) why the backwards-compat library (of whatever nature) is not in a separate package.
1) You cannot simply have a stub library with the old SONAME as the functionality was different. The return value for some of the library calls changed from -errno to errno. 2) from the changelog for libaio: * Thu Jan 27 2005 Jeff Moyer <jmoyer> - 0.3.103-3 - Move away from the separate compat packege, as it won't be installed by default and ISV's don't necessarily ship RPM's. As such, the compat package would have to be explicitly installed. - Build the library twice. Once with the old SONAME and once with the new one. The compat approach taken before did not work since the library uses versioned interfaces.
How exactly are the libraries functionally different other than in SONAME? The spec file just unpacks the tarball twice and builds it twice, once with a different soname.
That's a good point. Let me investigate this some more. Thanks!
OK, I misremembered the details of this particular change. The libraries are meant to be the same, save for the SONAME. What are the options for achieving this goal without building the library twice? Also, what is the motivation for changing it? As it stands, the library is 19K for i386, 20K for x86_64, 25k for ppc64.
I would guess that just creating a stub compat library linked against the new library would be sufficient but if it's not worth, but given the size of the library it's not necessarily worth it; WONTFIX seems fine.
Given comment #7, closing this bug.