On 12/05/2014 10:36 AM, Luis Machado wrote:
> On 12/01/2014 06:30 PM, Sergio Durigan Junior wrote:
>> On Monday, December 01 2014, Luis Machado wrote:
>>
>>> This test assumes the typeid symbols are always available before
>>> actually starting the inferior, which is not true for architectures
>>> that place such symbols under relocatable sections.
>>>
>>> The following patch fixes this by conditionalizing the execution of
>>> such tests on the accessibility of the typeid symbols before the
>>> inferior is running.
>>>
>>> Regression-tested on ppc32/64.
>>
>> Hey Luis!
>>
>> Thanks for the patch.  Just a somewhat minor comment.
>>
>>> diff --git a/gdb/testsuite/gdb.cp/typeid.exp
>>> b/gdb/testsuite/gdb.cp/typeid.exp
>>> index 9963a8a..7469b2b 100644
>>> --- a/gdb/testsuite/gdb.cp/typeid.exp
>>> +++ b/gdb/testsuite/gdb.cp/typeid.exp
>>> @@ -25,20 +25,35 @@ if {[prepare_for_testing $testfile.exp $testfile
>>> $srcfile {debug c++}]} {
>>>
>>>   proc do_typeid_tests {started} {
>>>       global hex
>>> +    global gdb_prompt
>>> +    set symbol_found 1
>>>
>>> -    # We might see the standard type or gdb's internal type.
>>> -    set type_re "(std::type_info|struct gdb_gnu_v3_type_info)"
>>> +    # Try to access one of the symbols to make sure it is
>>> available.  Some
>>> +    # architectures put the symbols on relocatable sections, which
>>> means
>>> +    # they will not be accessible before the inferior is running.
>>> +    send_gdb "print 'typeinfo for int'\n"
>>> +    gdb_expect {
>>> +    -re "No symbol \"typeinfo for int\" in current
>>> context.*$gdb_prompt" {
>>> +        set symbol_found 0
>>> +    }
>>> +    -re ".*$gdb_prompt" {
>>> +    }
>>> +    }
>>
>> Any particular reason for not using gdb_test_multiple here (and
>> everywhere else)?  This "send_gdb...gdb_expect" dialect is not used
>> anymore in the testsuite, AFAIR.
>>
>
> It looks a bit more natural when you are aiming at tests that should not
> expose PASS/FAIL. But gdb_test_multiple can be used that way as well,
> though with a somewhat strange empty testname parameter.
>
> Works the same though.
>
> I've updated the patch and fixed a previous gotcha in the logic.
>
> Ok?
>
>

Of course, now actually attaching the patch itself!