Re: How can I get this data into the .data section of the binary?

["Luke A. Guest" <laguest@archeia.com>]

On 16/06/2020 15:14, Niklas Holsti wrote:
> On 2020-06-16 14:31, Luke A. Guest wrote:
>> Hi,
>>
>> I'm trying to get some static data tables into the data section rather
>> than be elaborated at runtime. I can see no reason why this particular
>> set of types, records and aggregates cannot go into the data section.
> 
>    [snip]
> 
>>     Pixel_Format_Unknown     : constant Pixel_Format_Names :=
>>       Pixel_Format_Names'(Planar        => True,
>>                           Planar_Format => Planar_Pixels'
>>                             (others => Ada.Characters.Latin_1.NUL));

Strangely enough, with preelaborate, that's the only one which gets put
into rodata:

Disassembly of section
.rodata.sdl__video__pixel_formats__pixel_format_unknown:

0000000000000000 <sdl__video__pixel_formats__pixel_format_unknown>:
   sdl__video__pixel_formats__pixel_format_unknown : constant
   0:   00 00                   add    %al,(%rax)
        ...

Disassembly of section .bss.sdl__video__pixel_formats__A44s:

0000000000000000 <sdl__video__pixel_formats__A44s>:
   0:   00 00                   add    %al,(%rax)
        ...

Disassembly of section
.bss.sdl__video__pixel_formats__pixel_format_index_1_lsb:

0000000000000000 <sdl__video__pixel_formats__pixel_format_index_1_lsb>:
   sdl__video__pixel_formats__pixel_format_index_1_lsb : constant
   0:   00 00                   add    %al,(%rax)

> Several years ago we had a problem like this, with a large constant
> array that we wanted to have only once in RAM. This was with the XGC Ada
> compiler, but there we found a work-around: using positional rather than
> named association for the large array aggregate.
> 
> I have no idea if this will help you -- the types in our case were much
> simpler -- but you might try it with a small subset of your package:
> 
>    Pixel_Format_Unknown     : constant Pixel_Format_Names :=
>      (True, (NUL, NUL, NUL, NUL));
> 
> (assuming "use Ada.Characters.Latin_1").
> 

Nope, that doesn't work.