testsuite expectations

Jan Beulich jbeulich@novell.com
Fri Nov 17 10:17:00 GMT 2006


I'm testing a patch allowing executables not participating in any dynamic
linking to also have an .eh_frame_hdr section (I actually need this for
the linux kernel, but it's not just there that this is not working, as outlined
in an earlier mail).

In the course of that I needed to move the creation of the .eh_frame_hdr
section out of _bfd_elf_link_create_dynamic_sections. This in turn now
results in that section getting attached to the first non-shared input
object (of the same file format as the output) rather than simply the first
one of the same format. I don't think this by itself presents a problem,
since .eh_frame sections possibly resulting in entries being created in
.eh_frame_hdr cannot originate from shared inputs.

However, this also changes the section indices assigned by bfd, and
since ia64 (didn't check targets other than that and x86 so far) has
hidden dependencies on the section numbering (in how it calculates
the hash value for local symbols referenced from relocations), the
output now produced differs for at least one test case from the one
previously generated, as the order of linkage table (.got) entries
depends on the traversal order of the hash table entries. I would
generally think that test case expectations should not depend on
such internals (they would obviously also fail if any new section was
added internally by the linker [without this resulting in a new output
section), but before relaxing the expectation pattern accordingly
I'd like to confirm my assumption here is in sync with everybody else.

Thanks, Jan



More information about the Binutils mailing list