This is the mail archive of the
binutils@sourceware.org
mailing list for the binutils project.
Re: patch for fixing illegal instruction (wldrb, wldrh, wstrb, wstrh)
On Fri, 22 Sep 2006, Bridge Wu wrote:
> Thanks your guide. I submitted such patch for the first time.
>
> I used as and objdump to get the following .d file. Is that ok? Seems
> still a little different from iwmmxt2.d you generated. I used the
> options shown in iwmmxt.d.
I advise studying the differences - which change a particular output text
into regular expressions for a range of possible output texts - since they
are there for a reason.
The addresses start with "0+" to avoid hardcoding dependence on the size
of an address. (Not really relevant for ARM but may well be relevant for
dual 32-bit/64-bit targets.)
The <.text+...> are converted to <[^>]*> to avoid dependence on the
particular section names used on a particular target.
Various spaces/tabs are converted to [ ]+ - that is [<SPACE><TAB>]+ -
since the exact formatting of the output is not what is being tested.
[r1] needs to be converted to \[r1\] so that literal [ and ] are matched.
> I run 'make check' in my building folder (other than binutils source
> folder), but failed to get final result. I attached the log. It seemed
> I need to set DEJAGNU. Could you tell me how to solve it? I installed
> runtest in /usr/bin. The version is Expect-5.43.0, Tcl-8.4 and
> Framework-1.4.4.
For binutils testing you don't need to set DEJAGNU. You should however
use "make -k check" so that all tests run even if some fail.
# In unmodified tree with new testcase added.
make -k check
grep ^FAIL `find . -name '*.sum'`
# Should show the new test as a FAIL.
# With source patch applied.
make -k check
grep ^FAIL `find . -name '*.sum'`
# Should have identical results except that the FAIL for the new test has
# gone.
--
Joseph S. Myers
joseph@codesourcery.com