$ cat a.d

template T(int N) {
int[N] array;
}

const int x = 2;
const int y = x * x;

mixin T!(y);


$ dmd -c a.d
a.d(9): Integer constant expression expected instead of x * x

Since y is decl-ed as const, its value is known at compile time.

no@where.com wrote:
> $ cat a.d
> 
> template T(int N) {
> int[N] array;
> }
> 
> const int x = 2;
> const int y = x * x;
> 
> mixin T!(y);
> 
> 
> $ dmd -c a.d
> a.d(9): Integer constant expression expected instead of x * x
> 
> Since y is decl-ed as const, its value is known at compile time.
> 
> 

Although constant propagation and constant folding are nice,
yet I don't think it is a bug if the compiler misses some
optimizing opportunities.

Consider an extreme case:
const double x = cast(double)rand();
const double y = cos(x)*cos(x)+sin(x)*sin(x);
y is obviously 1.0, but it's not that obvious to all the compilers
I have used.

$ cat a.d

template T(int N) {
int[N] array;
}

const int x = 2;
const int y = x * x;

int[y] why_here_it_works;
//mixin T!(y);

$ dmd -c a.d
$ ls a.o
a.o



In article <d4f8uv$16q$1@digitaldaemon.com>, zwang says...
>
>no@where.com wrote:
>> $ cat a.d
>> 
>> template T(int N) {
>> int[N] array;
>> }
>> 
>> const int x = 2;
>> const int y = x * x;
>> 
>> mixin T!(y);
>> 
>> 
>> $ dmd -c a.d
>> a.d(9): Integer constant expression expected instead of x * x
>> 
>> Since y is decl-ed as const, its value is known at compile time.
>> 
>> 
>
>Although constant propagation and constant folding are nice, yet I don't think it is a bug if the compiler misses some optimizing opportunities.
>
>Consider an extreme case:
>const double x = cast(double)rand();
>const double y = cos(x)*cos(x)+sin(x)*sin(x);
>y is obviously 1.0, but it's not that obvious to all the compilers
>I have used.

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

no@where.com schrieb am Sun, 24 Apr 2005 02:40:15 +0000 (UTC):
> $ cat a.d
>
> template T(int N) {
> int[N] array;
> }
>
> const int x = 2;
> const int y = x * x;
>
> mixin T!(y);
>
>
> $ dmd -c a.d
> a.d(9): Integer constant expression expected instead of x * x
>
> Since y is decl-ed as const, its value is known at compile time.

Added to DStress as
http://dstress.kuehne.cn/run/const_18.d
http://dstress.kuehne.cn/run/const_19.d

Thomas


-----BEGIN PGP SIGNATURE-----

iD8DBQFCbIcl3w+/yD4P9tIRAnE+AJ4/SElBVafvCIzmbM77EsXylJvY3ACfTph1
JnxP+LHqsHHF10O3YbUmEgI=
=jE16
-----END PGP SIGNATURE-----