defining big types on avr-gcc

[Andrew Haley]

On 08/03/2010 04:26 PM, Massimiliano Cialdi wrote:
> I use avr-gcc 4.4.4:
> 
> Using built-in specs.
> Target: avr
> Configured with: ../gcc-4.4.4/configure --target=avr
> --enable-languages=c,c++ --disable-nls --disable-libssp --with-dwarf2
> --prefix=/usr/local/cross-gcc-avr-4.4.4 --disable-threads --with-gnu-as
> --with-gnu-ld
> Thread model: single
> gcc version 4.4.4 (GCC)
> 
> 
> 
> and my target is an ATMega32
> 
> I tried to define a type (only define, without allocate any variable):
> 
> 
> 
> typedef struct dummy
> {
>   uint8_t data[9000];
> } dummy_t;
> 
> typedef struct big_dummy
> {
>   dummy_t bigdata[4];
>   dummy_t singledata;
> } big_dummy_t;
> 
> 
> 
> I never allocate any big_dummy_t object. I use big_dummy_t only as
> placeholder (I use offsetof() macro), only to calculate offsets of
> object placed into a big extern serial flash memory.
> 
> I get the error from gcc:
> 
> error: size of array 'bigdata' is too large
> 
> it is 36000 bytes, bigger than 32KB.
> 
> after many tests I found a possible solution:
> 
> typedef struct dummy
> {
>   uint8_t data[9000];
> } dummy_t;
> 
> typedef struct big_dummy
> {
>   union
>   {
>     dummy_t bigdata[0];
>     struct
>     {
>       dummy_t bigdata0;
>       dummy_t bigdata1;
>       dummy_t bigdata2;
>       dummy_t bigdata3;
>     };
>   };
>   dummy_t singledata;
> } big_dummy_t;
> 
> now if I want the offset of a field I can do as follows:
> 
> uint16_t p = (uint16_t)&(((big_dummy_t*)0)->bigdata[3]);
> 
> This works, but if I wand offset of singledata:
> 
> uint16_t p = (uint16_t)&(((big_dummy_t*)0)->singledata);
> 
> I get a warning: 'integer overflow in expression'
> 
> but if I change in this way:
> big_dummy_t *p = (big_dummy_t*)0;
> uint16_t p = (uint16_t)&(p->singledata);
> 
> I get NO warning!!!
> 
> This looks incredible to me. I wonder why.
> Is it a bug?

I don't think it is.  gcc uses ssize_t (or, to be precise, a signed
integer type beased on sizetype) internally when constructing the size
of an array.  Given that array offsets can be negative or positive.
this seems reasonable enough.

Andrew.