https://sourceware.org/bugzilla/show_bug.cgi?id=33177
--- Comment #17 from Rainer Orth <ro at CeBiTec dot Uni-Bielefeld.DE> --- > --- Comment #16 from Alan Modra <amodra at gmail dot com> --- > Created attachment 16607 > --> https://sourceware.org/bugzilla/attachment.cgi?id=16607&action=edit > revised patch > > This one makes PROVIDE (_etext = .) and similar in scripts work for you by > setting up symbols early, and makes those ABI globals STT_OBJECT and SHN_ABS. > Plus assorted other tidies. Great, we're almosst there now: the symbol versioning test now PASSes just fine. However, there are differences in the symbols themselves. Since nm even with its sorted output is tedious to compare, I've now used Solaris elfdump instead, extracting only the relevant parts. I'm attaching the current ABI testsuite patch just for reference. What I've done then is: using solaris-exe and solaris-lib.so (a simple executable resp. unversioned shared object), once linked with ld and once with gld: for t in .symtab .dynsym; do for f in solaris-exe solaris-lib.so; do elfdump -s -N $t $f |grep '^[ ]*\['|awk '{printf "%s\t%s\t%s\t%s\n", $4, $5, $6, $9}'|sort -k +4 > $f$t; done; done then diffing the output and extracting the ABI symbols only. The only difference left is with ld only _GLOBAL_OFFSET_TABLE_ is protected, all others use default visiblity. -- You are receiving this mail because: You are on the CC list for the bug.
