On 9 May 2018 at 02:44, John Carter <john.carter@xxxxxxxxxxxxx> wrote:
> I tried playing with __builtin_constant_p and __builtin_choose_exp but they
> have a very different notion of what a compile time constant is. ie. It's
> something a literal or something declared const, _not_ something that is
> known at compile time.
Are you sure? __builtin_constant_p should be precisely something that
is known at compile time, whether declared const or not.
>
> Even weirder.... The following compiles without warnings, but.....
>
> #include <stdlib.h>#include <stdio.h>#include <assert.h>
> static char __rangeCheckVariable;
> #define asshurt(exp) \
> do { \
> char __rangeCheck[2]="a"; \
> __rangeCheckVariable = \
> __rangeCheck[(exp) ? 0 : 10]; \
> } while(0)
> void bunc( int i){
>
> asshurt( i< 4);
> assert( i<4);
> printf( "%d\n",i);}
> void func( int i){
> bunc( i);}
> int main( int argc __attribute__((unused)), char * argv[]
> __attribute__((unused))){
You know you can just declare main with no arguments?
int main(){
> int i = 10;
>
> func( i);
>
> return 0;}
>
> Removed the line assert( i< 4); and you get
>
> gcc -g3 -ggdb3 -O3 -W -Wall -Wextra -c d.c ;objdump -S d.o
> d.c: In function ‘main’:
> d.c:11:22: warning: array subscript is above array bounds [-Warray-bounds]
> __rangeCheck[(exp) ? 0 : 10]; \
> ^
> d.c:17:4: note: in expansion of macro ‘asshurt’
> asshurt( i< 4);
> ^
Now I think what Mason said applies: the extra code for assert(i<4)
prevents GCC from inlining bunc, and so the optimizer doesn't see that
the argument to bunc will make (i<4) false.
If you make bunc static, or add __attribute__((always_inline)) then
you get a warning even with the assert line present.
