Re: mxIs* functions now return bool

[Rik]

On 09/11/2016 07:36 AM, John W. Eaton wrote:
> On 09/11/2016 01:29 AM, Rik wrote:
>> 9/10/16
>> I've been working on improving the MEX interface in Octave.  One of the
>> things I noticed is that the predicate tests in Matlab are all returning a
>> bool, whereas in Octave they return an int.  A predicate test might be
>> something like mxIsScalar or mxIsComplex.  Does anyone see a problem with
>> changing the function prototypes to return bool?
>>
>> extern OCTINTERP_API int mxIsComplex (const mxArray *ptr);
>> =>
>> extern OCTINTERP_API bool mxIsComplex (const mxArray *ptr);
>
> I suspect this is a relatively recent change in Matlab (for some value of
> relatively recent).  I'm pretty sure they used to return int.
>
>> One of the things I don't understand is that the C language didn't have a
>> bool type until C99, and even then it was necessary to #include <stdbool.h>
>> in order to access it.  However, I accidentally coded mxIsScalar to return
>> bool and it has compiled correctly.  The only thing I see at the top of
>> mexproto.h is
>>
>> #if defined (__cplusplus)
>> #  include <cstdlib>
>> extern "C" {
>> #else
>> #  include <stdlib.h>
>> #endif
>>
>> When I run 'mex -v ...' I can see that the compiler is gcc, not g++, so I
>> don't think the first branch is being taken.  Should I worry and add the
>> include for stdbool.h when compiling with a straight C compiler?
>
> If add -save-temps to the compiler command that is used and then look at
> the generated .i  file can you see where bool is defined?

That was a help.  It turns out that we are defining it ourselves in mex.h

#if ! defined (__cplusplus) && ! defined (__bool_true_false_are_defined)
#  if ! defined (bool)
typedef int bool;
#  endif
#  if ! defined (true)
#    define true 1
#  endif
#  if ! defined (false)
#    define false 0
#  endif
#endif

I'm not sure how many people are using a C compiler older than 17 years
(pre-C99), but it seems that there is a gnulib module for stdbool so I
think we can add that to the modules list and then unconditionally include
stdbool.h in the C side of Octave.

>
> If I compile the simple program "bool x;" with gcc, it errors on the
> undefined bool type.  So I think there must be some place that bool is
> being defined.
>
> But I don't think it would hurt to include stdbool.h along with
> stdlib.h.  If that causes trouble, I'm sure we'll hear about it.

Yes, and I thought if we're really going to change APIs we should do it on
a big change like the 4.2.0 release.

--Rik