[splint-discuss] array element definition before use

[Simon Liddington]

I'm gettting this:

test_Database.c:3037:38: Array element fixed1ElemFields[0] used before
                            definition
  An rvalue is used that may not be initialized to a value on some execution
  path. (Use -usedef to inhibit warning)

The relevent code is (BTW, note the annoying use of casts to initialise values
of type dsdbm_fieldType_t (actually unsigned char i.e. 8 bit):

   dsdbm_fieldType_t fieldSizes[] = { (dsdbm_fieldType_t)1,
                                      (dsdbm_fieldType_t)5,
                                      (dsdbm_fieldType_t)8,
                                      (dsdbm_fieldType_t)9,
                                      (dsdbm_fieldType_t)16,
                                      (dsdbm_fieldType_t)17,
                                      (dsdbm_fieldType_t)20,
                                      (dsdbm_fieldType_t)32 };
  ....

   /*@reldef@*/ dsdbm_identifier_t fixed1ElemFields[sizeof(fieldSizes) /
                                                    sizeof(*fieldSizes)];

      for (.....)
      {
         for (sizeIndex = 0; sizeIndex < numFieldSizes ; ++sizeIndex)
         {
            for (.....)
            {
               for (.....)
               {
                  for (.....)
                  {
                     if (expression)
                     {
                        fixed1ElemFields[sizeIndex] = fieldId;
                     }
                  }
               }
            }
         }
      }

      ... code that uses fixed1ElemFields[]

I can tell splint the for loops execute with +forloopexec but I can't tell it
that "expression" is guaranteed to be true at least once for each value of
sizeIndex which means the array is fully defined at the end of these loops.

I don't want to use -usedef everywhere I use the array following this.

I also don't want to pre-initialise the array as I would have to change the 
initialiser each time I change fieldSizes[] (but it's probably my best bet 
for now).

The question is why doesn't reldef work?

Thanks,
Simon