aot-compile-rpm or something that it invokes leaves buildroot paths to *.jar.so.debug. Example, ant-debuginfo-1.6.5-2jpp.2: $ strings /usr/lib/debug/usr/lib64/gcj/ant/ant-1.6.5.jar.so.debug | grep /var/tmp /var/tmp/ant-1.6.5-2jpp.2-buildroot/usr/lib64/gcj/ant I'm wondering what's that path used for if anything in the *.jar.so.debug - and if it is used, whether it's a security issue - or if it's expected/harmless. It does anyway trip check-buildroot from rpmdevtools (in Extras) - it finds the buildroot path in installed files and intentionally causes the build fail in those cases. For now, I'll exclude *.jar.so.debug from files being checked in check-buildroot so from that POV this is not urgent.
It's here: DW_AT_comp_dir : (indirect string, offset: 0x2c340): /var/tmp/ant-1.6.5-1jpp_11fc-buildroot/usr/lib64/gcj/ant whereas in libgcc we have: DW_AT_comp_dir : (indirect string, offset: 0xa0): /usr/src/debug/gcc-4.1.1-20060828/obj-x86_64-redhat-linux/gcc so, it looks as though when the debuginfo file is extracted from the shared object the directory name component is not rewritten. This is nothing to do with aot-compile-rpm.
Gary, does this issue apply to java-1.5.0-gcj? I'm trying to close out old java-1.4.2-gcj-compat bugs.
I'm not sure. Have a look in /usr/bin/aot-compile-rpm and find the line that says 'compiler.compile()'. If it says 'copy(tmpdir, srcdir, dstdir)' after it, the bug is fixed. If there's no copy then the bug is still there.
The copy line is there in Rawhide.