Re: AdaCore Community 2018 Bugs

[Simon Wright <simon@pushface.org>]

Roger <rogermcm2@gmail.com> writes:

> On Sunday, August 12, 2018 at 5:33:35 PM UTC+10, Simon Wright wrote:
>> Roger <rogermcm2@gmail.com> writes:
>> 
>> > As it now occurs even with GCC 6.1.0 I'm thinking it most likely has
>> > to do with recent OSX changes.
>> 
>> Since it occurs on Debian Stretch, GCC 6.3.0, it can't be just macOS.
>> 
>> The actual error reported can change with successive compilations
>> (sometimes it says something like 'undeclared variable ""', seems to
>> settle on the CE), a surefire symptom of bad memory handling.
>> 
>> > This problem doesn't occur for compile only.
>> > It occurs when I Build All from GPS
>> 
>> The error is a crash in the compiler, doesn't matter how you get there.
>
> I Have cut most of the code out of the test program but the crash
> still occurs.  However, I am making progress on tracking back to the
> minimal code where the crash first occurs.

The problem is in the packages Morph.Mesh_Morph_Value_Pointers,
Mesh_Morph_Weight_Pointers.

You say

   package Mesh_Morph_Value_Pointers is new Interfaces.C.Pointers
     (Interfaces.C.unsigned, API_Morph_Value, API_Morph_Values_Array,
      API_Morph_Value'(others => <>));

but

   type API_Morph_Value is new Interfaces.C.unsigned;

so others => <> isn't a legal expression for an API_Morph_Value; you
should say

      API_Morph_Value'(0));


There is still a bug, which ought to be fixed, but it's that GNAT has
failed to recognise the actual problem and has got fatally confused.