Re: GNAT 4.0 strange behaviour

[James <james@recherche.enac.Fr>]

James wrote:
> jimmaureenrogers@worldnet.att.net wrote:
> 
>> James wrote:
>>
>>> This is a very small program compiled under linux debian with gnat 4
>>> based on gcc 4.0
>>> The output of the program is
>>>  1 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 1 0 0
>>> 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
>>> which is of course not right...
>>> If line labelled 1 is removed and replaced by line labelled 2, the
>>> output is correct.
>>> The output is also correct if part 1 is replaced by part 2.
>>> Any advice or place where I could send a bug report?
>>>
>>> with Text_Io;
>>> use Text_Io;
>>> with Interfaces;
>>> use Interfaces;
>>>
>>>
>>> procedure Toto is
>>>    type Intboard is new Unsigned_64;
>>>    for Intboard'Size use 64;
>>>    type Bit is new Integer range 0..1;
>>>    for Bit'Size use 1;
>>
>>
>>
>> You seem to think that Bit is an unsigned type.
>> It is not. It is a derived type from Integer, which is a
>> signed type.
>>
>> If you want an unsigned integer with a range of 0..1
>> you should declare:
>>
>> type Bit is mod 2;
>>
>> Ada unsigned integer types are defined using modular types.
>>
>> I am willing to bet the answer will improve if you use a
>> modular type.
>>
>> Strictly speaking, the subtypes used in part 1 below do not
>> match the subype used in Bitboard array index definition.
>>
>> The part 1 approach to indexing look suspiciuosly like C coding
>> logic. The part 2 approach is closer to a classical Ada.
>>
>> The most correct and maintainable approach to the looping
>> is:
>>
>> for I in A_B'Range loop
>>    A_B(I) := 1;
>>    Put(Bit'Image(A-B(I)));
>> end loop;
>>
>> Clearly your goal is to set each element of A_B to 1 and
>> then print a string representation of that element. The
>> approack shown above will correctly satisfy that goal even
>> if the index range of A_B should change sometime in the
>> future due to changes in system requirements.
>>
>>
>>>    type Bitboard is array (0..63) of Bit;
>>>    pragma Pack(Bitboard);
>>>    for Bitboard'Size use 64;
>>>
>>>    A_B : Bitboard;
>>>    A_I : Intboard;
>>>    for A_I'Address use A_B'Address;        --label1
>>> --   for A_B'Address use A_I'Address;      --label2
>>>
>>> begin
>>>    --part 1
>>>    for I in 0 .. 7 loop
>>>       for J in 0 .. 7 loop
>>>          A_B(I*8+J) := 1;
>>>          Put(Bit'Image(A_B(I*8+J)));
>>>       end loop;
>>>    end loop;
>>>
>>>    --part 2
>>> --   for I in 0 .. 63 loop
>>> --      A_B(I) := 1;
>>> --      Put(Bit'Image(A_B(I)));
>>> --   end loop;
>>> end Toto;
>>
>>
>>
>> Jim Rogers
>>
> Well, in fact in the beginning it was a boolean type. But I followed 
> your suggestion using modular types and the result is exactly identical...
> I must stress the fact that:
> 1) I don't use the overlay, I just declare it and never use it.
> 2) Part 2 is absolutely identical to part 1 from a semantical point of 
> view, I just use two loops in part 1 instead of one in part 2, but the 
> range is exactly the same (0..63). So the result should be the same if 
> the compiler was working properly according to me, whatever the size, 
> type or any other thing.
> 
> Any ideas?
Well, just a last word: everything is OK with gnat 3.4. So, it's a 
certainly a bug in gnat 4.0

Thanks