Hi all,
I'm running into what I suspect to be a linker related issue.
When I compile Ruby + Openssl extension on a machine running snv_75, the
resulting binary is able to load the compiled extension without any
problems . . .
root at dn17 nightly-nd >/export/prashant/ruby_builds/1.8/bin/ruby -e
"require 'openssl'"
root at dn17 nightly-nd > #extension loaded fine.
Now if I were to hook the same Makefile and configure options into the
sfwnv consolidation, and create packages, the resulting Ruby cannot load
the Openssl extension.
ie.,
root at dn17 nightly-nd >/usr/ruby/1.8/bin/ruby -e "require 'openssl'"
/usr/ruby/1.8/lib/ruby/1.8/i386-solaris2.11/openssl.so: ld.so.1: ruby:
fatal: relocation error: file
/usr/ruby/1.8/lib/ruby/1.8/i386-solaris2.11/openssl.so: symbol
ENGINE_load_openssl: referenced symbol not found -
/usr/ruby/1.8/lib/ruby/1.8/i386-solaris2.11/openssl.so (LoadError)
from /usr/ruby/1.8/lib/ruby/1.8/openssl.rb:17
from -e:1:in `require'
from -e:1
Now, the symbol ENGINE_load_openssl shows up as undefined, irrespective
of which build of openssl.so I look at. So I assumed that vanilla Ruby
build "knew" where to look for this symbol, while the sfwnv Ruby build
did not.
So I used LD_DEBUG=symbols to watch where ENGINE_load_openssl was being
loaded from. And it so happens that this symbol is not loaded at
all(going by the contents of the LD_DEBUG_OUTPUT file), in the vanilla
Ruby build . . . but in the Ruby built from sfwnv, the loader tries hard
to resolve it, and finally gives up(and fails).
So is there a property in an so file to tell the linker not to load
symbols that are not used? Or any other ideas about what could be
different compiler/linkers wise, about the sfwnv openssl.so(or Ruby)
that is causing this load to fail?
Thanks
