Further investigation shows this issue affects both qt and kdelibs now, so would request this be addressed in EL-5 as well. +++ This bug was initially created as a clone of Bug #429846 +++ Due to licensing issues, kdelibs dlopens openssl instead of linking. In order to make this work, kdelibs needs to know the soname of both libcrypto and libssl. Upstream kdelibs *tries* to use libssl.so.SHLIB_VERSION_NUMBER (and libcrypto.so.SHLIB_VERSION_NUMBER) as defined by openssl/opensslv.h header. Unfortunately, this doesn't work, since SHLIB_VERSION_NUMBER=0.9.8 and the soname in rawhide is named libssl.so.0.9.8g In order to workaround this, we're currently having to patch kdelibs to hardcode looking for libssl.so.0.9.8g/libssl.so.7 Is the value of SHLIB_VERSION_NUMBER incorrect? Is this usage/expection of SHLIB_VERSION_NUMBER to name sonames invalid? Any suggestions on how to make this work better? -- Additional comment from tmraz on 2008-01-23 09:07 EST -- The file name in rawhide has to be ....0.9.8g otherwise it would conflict with potential compat libraries with different SONAME. We should probably patch the SHLIB_VERSION_NUMBER to reflect that but it could theoretically break other things as I don't know how 3rd party apps use this macro. But don't ask me what I think about dlopening openssl libraries to overcome the GPL incompatibility. So this will not be a priority for me to solve. kdelibs should instead use GPL or GPL compatible licensed code for SSL. Such as nss_compat_ossl library. -- Additional comment from rdieter.edu on 2008-01-23 09:11 EST -- Interested in exposing SHLIB_SONAMEVER (in opensslv.h or somewhere) as referenced in openssl-0.9.8g-soversion.patch ? That could serve nicely. -- Additional comment from tmraz on 2008-01-23 10:13 EST -- I could add that but will it really help you? It will not be in openssl upstream so KDE cannot depend on it. -- Additional comment from rdieter.edu on 2008-01-23 10:24 EST -- fair nuf, I'll try pinging both openssl, kde upstreams and see how best to address this. -- Additional comment from rdieter.edu on 2008-01-23 11:21 EST -- fyi, http://marc.info/?l=openssl-users&m=120110500517612&w=2 -- Additional comment from tmraz on 2008-01-23 17:51 EST -- The soname is actually libssl.so.7 - we patch it to be so otherwise it would be libssl.so.0 - we do this because the abi is not stable so we almost always have to bump sonames on version upgrades The file name is libssl.so.0.9.8g - again we patch it this way otherwise it would be libssl.so.0.9.8 - again we do this so multiple openssl versions (with potentially different ABI) could be installed together. This makes us (not upstream) to be not in sync with SHLIB_SONAMEVER. So this question doesn't make much sense to be reported upstream. As I wrote we could/should probably patch the SHLIB_SONAMEVER to be 0.9.8g but there is a small possibility it might break things. -- Additional comment from rdieter.edu on 2008-01-23 22:04 EST -- Thanks for the clarification. I don't care so much about SHLIB_SONAMEVER, it's the value of SHLIB_VERSION_NUMBER that I'm more interested in, since that is what (upstream) kdelibs uses to determine the name of the shlib(s) to dlopen. -- Additional comment from rdieter.edu on 2008-01-23 22:08 EST -- My take/wish: fix SHLIB_VERSION_NUMBER to match reality. 3rd-parties that use this are broken in the status-quo (e.g. kdelibs). -- Additional comment from tmraz on 2008-01-24 03:02 EST -- (In reply to comment #7) > I don't care so much about SHLIB_SONAMEVER, it's the value of > SHLIB_VERSION_NUMBER that I'm more interested in, since that is what (upstream) > kdelibs uses to determine the name of the shlib(s) to dlopen. Oops, I meant SHLIB_VERSION_NUMBER instead of SHLIB_SONAMEVER in the comment 6. -- Additional comment from tmraz on 2008-01-26 15:40 EST -- Fixed, let's see what breaks - if anything.
I am not too fond of changing this for RHEL-5. There is still theoretical chance that the change could break third party apps which use SHLIB_VERSION_NUMBER in a different way than kdelibs/qt.
arg, I thought we had concluded what the *correct* use/value of SHLIB_VERSION_NUMBER is (should be!). Anything that depends on it being wrong, is just plain broken, imo.
Yes, but the changes in released Red Hat Enterprise Linux distributions should be as conservative as possible. This will be fixed in Red Hat Enterprise Linux 6.