Hi all, I am trying to build some mainstream packages on a non-mainstream Linux architecture that prepends an underscore to symbol names (e.g., "foo" in C is "_foo" to the linker). Many packages will limit the number of symbols they export with -export-symbols-regex e.g., -export-symbols-regex "mysyms_*". This doesn't work for my architecture even though libtool appears to understand that a symbol prefix is needed to some degree.
The generated libtool I get has the following definitions: global_symbol_pipe="sed -n -e 's/^.*[ ]\\([ABCDGIRSTW][ABCDGIRSTW]*\\)[ ][ ]*_\\([_A-Za-z][_A-Za-z0-9]*\\)\$/\\1 _\\2 \\2/p'" export_symbols_cmds="\$NM \$libobjs \$convenience | \$global_symbol_pipe | \$SED 's/.* //' | sort | uniq > \$export_symbols" The output of global_symbol_pipe looks something like this: T _foo foo The sed expression in export_symbols_cmds will then reduce this to: foo This is the symbol name that is used in the linker version script. But ld doesn't know anything about unprefixed symbol names so I end up with no global symbols in the resulting binary. It seems to me like it might be more appropriate for the sed expression in export_symbols_cmds to pick out the second word rather than the third, something like: export_symbols_cmds="\$NM \$libobjs \$convenience | \$global_symbol_pipe | \$SED 's/[^ ]* \\([^ ]*\\) .*/\\1/' | sort | uniq > \$export_symbols" Or would that be incorrect? Thanks, (Please cc me, I'm not subscribed) _______________________________________________ http://lists.gnu.org/mailman/listinfo/libtool
