This is the mail archive of the
mailing list for the GDB project.
Re: RFA: Delete `weird.exp:p v_comb' test
Fernando Nasser <email@example.com> writes:
> Jim Blandy wrote:
> > 2001-11-01 Jim Blandy <firstname.lastname@example.org>
> > * gdb.stabs/weird.exp: Delete "p v_comb" test. It assumes that
> > pointers are 32 bits long, and that offsets of relocs are always
> > stored in the data (REL-style), and not in the reloc entry itself
> > (RELA-style).
> If you think it should go I will take your word for it. Please check it
> Just a question: what were we supposed to be testing with this test
> (and that we won't be testing anymore)?
We were testing GDB's ability to make sense of old-ABI C++ objects
built by hand-written assembly code, not compiler-generated code.
Effectively, it tests GDB's ability to understand that specific C++
ABI, not just the one the compiler happens to be using at the time.
It simultaneously tests GDB's ability to debug unlinked .o files.
The rest of the file also exercises the latter. I think the C++ tests
do a sufficient job of the former. And it would be difficult to make
the test portable.
I'm open to suggestions as to better ways to handle it. Simply
xfailing it seems like avoiding the issue; if people tell me what they
want, I will do it, but whatever decisions there are to make, I want
to get them over with. That test has been failing (and being ignored)
at least since 1997.