This is the mail archive of the
mailing list for the binutils project.
Re: PATCH: Don't align text/data/bss sections in x86 assembler
- From: Roland McGrath <mcgrathr at google dot com>
- To: "H.J. Lu" <hjl dot tools at gmail dot com>
- Cc: "binutils at sourceware dot org" <binutils at sourceware dot org>
- Date: Mon, 3 Jun 2013 17:01:17 -0700
- Subject: Re: PATCH: Don't align text/data/bss sections in x86 assembler
- References: <20130531173241 dot GA30259 at intel dot com>
On Fri, May 31, 2013 at 10:32 AM, H.J. Lu <email@example.com> wrote:
> Roland, there are some x86 NACL test failures prior to my change. I
> don't know if they are real problems or not.
The ld-size tests all fail in a NaCl configuration for two reasons. (This
is not new. It has been this way since you added those tests and I'm not
regularly checking the x86 results because only the arm-nacl target is
currently in production use.)
First, they use -melf_i386 et al. In a --target=i686-nacl configuration,
elf_i386 is not supported, only elf_i386_nacl (and likewise for x86-64 and
x32 variants). So "#target: i?86-*-*" does not go with "#ld: -melf_i386"
unless you do the option_regsub hack. My preference is to do that hack
rather than disable these tests for i?86-*-nacl*. Then we get to...
Second, their regexps require certain specific addresses that presume the
exact layout details of the common layout, which are different for NaCl
targets. Is it really necessary for these tests to match the particular
addresses? It looks to me like they'd still be testing everything they
actually want to test if they ignored the specific addresses in the objdump
output (the contents and order will still match for NaCl targets). If you
think that's OK, I can send a change.
The other two failures I see now (in i386.exp) look like they are in fact
related to your change. e.g. you updates tlsbindesc.rd but not
tlsbindesc-nacl.rd. Please use a --target=x86_64-nacl build to run the
test suite and fix those up so that i386.exp has no failures.