Re: -fdump-ada-spec

[Simon Wright <simon@pushface.org>]

Ludovic Brenta <ludovic@ludovic-brenta.org> writes:

>> Apparently one should use g++ rather than gcc, because it makes a better
>> job of it.
>>
>> GNAT GPL 2014 does translate (some) macros; it doesn't always get it
>> right, eg
>>
>>    #define M_SQRT2     1.41421356237309504880168872420969808   /* sqrt(2)        */
>>    #define M_SQRT1_2   0.707106781186547524400844362104849039  /* 1/sqrt(2)      */
>>    #define MAXFLOAT    0x1.fffffep+127f
>>
>>    // have to have _something_ besides the constants or no .ads gets
>>    // generated.
>>    extern float sinf(float);
>>
>> turns into
>>
>>    pragma Ada_2005;
>>    pragma Style_Checks (Off);
>>
>>    with Interfaces.C; use Interfaces.C;
>>
>>    package math_h is
>>
>>
>>       M_SQRT2 : constant := 1.41421356237309504880168872420969808;  --  ./math.h:1
>>       M_SQRT1_2 : constant := 8#.707106781186547524400844362104849039#;  --  ./math.h:2
>                                 ^^                                     ^
>
> Isn't that a blatant bug?  Either in the Ada or in the C source?

Yes; like I said, it doesn't always get it right.

And it doesn't understand hex floats either: 16#1.fffffep+127f# is *not*
a good translation of 0x1.fffffep+127f !

I have reported these problems to AdaCore, but as they have always said
the translation isn't expected to be perfect, so it'll be very low on
their priority list.