//
// bs3.d
// 2025-05-05 stvo
//
// $ dmd -g -checkaction=context -unittest -main -run bs3.d
// bs3.d(25): [unittest] [4, 3, 3, 4] != [4, 3, 2, 1]
// 1/1 modules FAILED unittests
//

auto foo (ubyte [4] s)
{
   auto u = s;
   ubyte [4] tmp = u;
   u[0] = tmp [3];
   u[1] = tmp [2];
   u[2] = tmp [1];
   u[3] = tmp [0];
   return u;
}

unittest {
   enum ubyte [4] u = [1,2,3,4];
   enum r = foo (u);
   auto s = foo (u);
   assert (r == s);
}

According to the spec [1] the above code conforms to the CTFE restrictions and contains no pointers. It also does not contain "implementation defined behavior". Thus the result of CTFE and RTFE should be identical but they differ.

Suggestions appreciated!

[1] https://dlang.org/spec/function.html#interpretation
20.22 Compile Time Function Execution (CTFE)

On Monday, 5 May 2025 at 17:15:57 UTC, kdevel wrote:

even worse:

auto foo(ubyte[4] s){
    ubyte[] u = s;
    ubyte[4] tmp = u;
    u[0] = tmp[3];
    u[3] = tmp[0];
    return u;
}

unittest{
    import std;

    enum ubyte[4] u = [1, 2, 3, 4];
    enum r = foo(u);
    auto s = foo(u);
    r.writeln;
    s.writeln;
    assert(r!=s);
}

           2.105.3: Success with output:
-----
1 modules passed unittests
[1, 2, 3, 4]
[4, 2, 3, 1]
[216, 88, 171, 59]
-----

           2.106.1: Success with output:
-----
1 modules passed unittests
[1, 2, 3, 4]
[4, 2, 3, 1]
[136, 48, 143, 115]
-----

Since      2.107.0: Success with output:
-----
1 modules passed unittests
[1, 2, 3, 4]
[4, 2, 3, 1]
[56, 227, 42, 198]
-----

what fun

On Monday, 5 May 2025 at 17:15:57 UTC, kdevel wrote:

//
// bs3.d
// 2025-05-05 stvo
//
// $ dmd -g -checkaction=context -unittest -main -run bs3.d
// bs3.d(25): [unittest] [4, 3, 3, 4] != [4, 3, 2, 1]

This is a bug, please report it.

-Steve

On Tuesday, 6 May 2025 at 01:29:35 UTC, Steven Schveighoffer wrote:

//
// bs3.d
// 2025-05-05 stvo
//
// $ dmd -g -checkaction=context -unittest -main -run bs3.d
// bs3.d(25): [unittest] [4, 3, 3, 4] != [4, 3, 2, 1]

I would like to yield to someone with a GH account.

On Monday, 5 May 2025 at 17:15:57 UTC, kdevel wrote:

auto foo (ubyte [4] s)
{
   auto u = s;
   ubyte [4] tmp = u;

When I put shared in front of that definition the unittest passes.

   u[0] = tmp [3];
   u[1] = tmp [2];
   u[2] = tmp [1];
   u[3] = tmp [0];
   return u;
}

I appreciate if sb. with a GH account could file the issue with the original code as bug.