Add C++ versions of iscanonical for ldbl-96 and ldbl-128ibm

  On 10/3/17, H.J. Lu <hjl.tools@gmail.com> wrote:
> On 10/3/17, H.J. Lu <hjl.tools@gmail.com> wrote:
>> On 10/3/17, Gabriel F. T. Gomes <gabriel@inconstante.eti.br> wrote:
>>> On 03 Oct 2017, Joseph Myers wrote:
>>>
>>>>> +/* In C++ mode, __MATH_TG cannot be used, because it relies on
>>>>> +   __builtin_types_compatible_p, which is a C-only builtin.  On the
>>>>> +   other hand, overloading provides the means to distinguish between
>>>>> +   the floating-point types.  The overloading resolution will match
>>>>> +   the correct parameter (regardless of type qualifiers (i.e.: const
>>>>> +   and volatile).  */
>>>>
>>>>(twice in this patch, and once already in math.h) is missing a second
>>>>close parenthesis to match the first open parenthesis, and should be
>>>>fixed.
>>>
>>> Thanks.  Pushed with these changes.
>>>
>>>
>>
>> On i686 wit GCC 7, I got
>>
>> test-math-iscanonical.cc: In function ‘void check_type()’:
>> test-math-iscanonical.cc:33:11: error: use of an operand of type
>> ‘bool’ in ‘operator++’ is deprecated [-Werror=deprecated]
>>      errors++;
>>            ^~
>> test-math-iscanonical.cc: In instantiation of ‘void check_type() [with
>> T = float]’:
>> test-math-iscanonical.cc:39:22:   required from here
>> test-math-iscanonical.cc:33:11: error: use of an operand of type
>> ‘bool’ in ‘operator++’ is deprecated [-Werror=deprecated]
>>      errors++;
>>      ~~~~~~^~
>>
>
> I am testing this:
>
> diff --git a/math/test-math-iscanonical.cc b/math/test-math-iscanonical.cc
> index aba68acb4f..8ced7a73b4 100644
> --- a/math/test-math-iscanonical.cc
> +++ b/math/test-math-iscanonical.cc
> @@ -20,7 +20,7 @@
>  #include <math.h>
>  #include <stdio.h>
>
> -static bool errors;
> +static int errors;
>
>  template <class T>
>  static void
>
>

This is what I checked in.
  

On 03 Oct 2017, H.J. Lu wrote:
>On 10/3/17, H.J. Lu <hjl.tools@gmail.com> wrote:
>> --- a/math/test-math-iscanonical.cc
>> +++ b/math/test-math-iscanonical.cc
>> @@ -20,7 +20,7 @@
>>  #include <math.h>
>>  #include <stdio.h>
>>
>> -static bool errors;
>> +static int errors;
>>
>>  template <class T>
>>  static void
>>
>>  
>
>This is what I checked in.

Thanks!
  

On Wed, 4 Oct 2017, H.J. Lu wrote:

> This is what I checked in.

This fix doesn't seem to be on 2.26 branch, but needs to go there as the 
original patch went there.

I don't think using an int count of errors and returning it from do_test 
is a good coding pattern, because if the count reaches 77 it will result 
in a spurious UNSUPPORTED result.  Of course in this particular test it 
can't reach 77, but a better pattern is either a boolean error state (set 
to true rather than using ++, given the warning quoted here), or a count 
but with do_test returning errors != 0.
  

On 10/04/2017 03:15 PM, Joseph Myers wrote:
> On Wed, 4 Oct 2017, H.J. Lu wrote:
> 
>> This is what I checked in.
> 
> This fix doesn't seem to be on 2.26 branch, but needs to go there as the
> original patch went there.
> 
> I don't think using an int count of errors and returning it from do_test
> is a good coding pattern, because if the count reaches 77 it will result
> in a spurious UNSUPPORTED result.  Of course in this particular test it
> can't reach 77, but a better pattern is either a boolean error state (set
> to true rather than using ++, given the warning quoted here), or a count
> but with do_test returning errors != 0.

Agreed.  Note that TEST_VERIFY allows the test to continue after a 
failure, and it also arranges for a non-zero exit status (even across 
fork, but currently not across dlopen).  It's usually a good alternative 
to such error variables.

Thanks,
Florian