rpath and LD_LIBRARY_PATH are only relevant for the runtime linking/resolution, right? My concern is that at *compilation time* the wrong version of OpenSSL is being picked up. Is there a way to see the full library/header path the compiler is using so that I can verify the path I specified with --with-headers and --with-libraries is taking precedence over the system paths?
On Tue, Sep 29, 2020 at 1:25 PM Tom Lane <t...@sss.pgh.pa.us> wrote: > Joshua Poehls <jos...@poehls.me> writes: > > I'm compiling PG 12.2 (on CentOS 7) and need to ensure that it uses a > > specific version of OpenSSL (1.1.1h). I've added what I think are the > > appropriate `./configure` flags (see below) and compilation is successful > > but when I run `ldd -d initdb` or `readelf` I can see that it linked to a > > different version of libcrypto/libssl. > > > --with-openssl --with-includes=/path/to/ssl/include > > --with-libraries=/path/to/ssl/lib > > > My theory is that even though I'm adding > --with-libraries/--with-includes, > > the build script is finding and using the version of OpenSSL in /lib64/ > > because it has higher precedence somewhere. > > Linux is generally pretty unfriendly to putting shlibs in nonstandard > places. If you want to do that, you need to add an "rpath" spec to > the calling programs, or set LD_LIBRARY_PATH, or fool around with the > dynamic linker's configuration files. "man ld.so" will get you started. > > regards, tom lane > -- -- Joshua Poehls jos...@poehls.me
