On 05/20/2011 11:37 PM, Pedro Alves wrote:
> Did you try setup_kfail?  See below.  The way you have things
> doesn't catch the internal error case because that is matched
> within gdb_test_multiple itself.  You could also check the return
> of gdb_test_multiple to see if an internal match happened, but
> that's more complicated than setup_kfail.
> 

I see.  It has been mentioned in the comment to proc gdb_test_multiple

# Returns:
#    1 if the test failed, according to a built-in failure pattern
#    0 if only user-supplied patterns matched
#   -1 if there was an internal error.

Thanks.

>> > x86_64-unknown-linux-gnu:
>> > KFAIL: gdb.base/varargs.exp: print find_max_float_real(4, fc1, fc2, fc3,
>> > fc4) (PRMS: gdb/12790)
>> > KFAIL: gdb.base/varargs.exp: print find_max_double_real(4, dc1, dc2,
>> > dc3, dc4) (PRMS: gdb/12790)
>> > FAIL: gdb.base/varargs.exp: print find_max_long_double_real(4, ldc1,
>> > ldc2, ldc3, ldc4) (GDB internal error)
>> > 
>     setup_kfail gdb/12776 "i?86-*-*"
>     setup_kfail gdb/12790 "x86_64-*-*"
>     setup_kfail gdb/12791 "arm*-*-*"
>     set test "print find_max_long_double_real(4, ldc1, ldc2, ldc3, ldc4)"
>     gdb_test_multiple $test $test {
> 	-re ".*= 4 \\+ 4 \\* I.*${gdb_prompt} $" {
> 	    pass $test
> 	}
>     }

This is exactly what I did when I was writing this patch.  However, the
wrong PR number is got in KFAIL result, like this,

KFAIL: gdb.base/varargs.exp: print find_max_long_double_real(4, ldc1,
ldc2, ldc3, ldc4) (PRMS: gdb/12791)

I was running test case on i686-pc-linux-gnu, but the PR number
displayed was arm's PR.  Source code of proc setup_kfail shows that PR
number is saved in kfail_prms unconditionally.  Of course, only the PR
in last call of setup_kfail is saved.  This is a limitation of
setup_kfail, IMO.  I gave up on this direction then.

In order to overcome this limitation, a new proc setup_kfail_for_target
is added in lib/gdb.exp, which only call setup_kfail if istarget returns
true.

> Maybe you can even convert the gdb_test_multiple's to gdb_test that way.
> (You could also put the setup_kfails in a procedure to not need to 
> repeat them everywhere).

Since we are using setup_kfail, we can surely convert gdb_test_multiple
to gdb_test.  I don't put setup_kfails in a proc, because, KFAILs on x86
are different from KFAILs on arm/x86_64.

In my new patch, the internal-error on x86_64 is KFAIL'ed.

KFAIL: gdb.base/varargs.exp: print find_max_long_double_real(4, ldc1,
ldc2, ldc3, ldc4) (GDB internal error) (PRMS: gdb/12790)

-- 
Yao (齐尧)