Re: Bug: Accessing return value of type static array with length 1 or 2 by index.

[John Colvin]

On Saturday, 20 April 2013 at 13:37:55 UTC, John Colvin wrote:

On Saturday, 20 April 2013 at 12:23:20 UTC, deed wrote:

import std.stdio : writeln;

template Template (uint n, T)
{
   T[n] statArr()
   {
       T[n] arr;
       return arr;
   }

   T[] dynArr()
   {
       T[] dynArr = new T[n];
       return dynArr;
   }
}

void main()
{
   alias statArr9 = Template!(9, int).statArr;
   alias statArr3 = Template!(3, int).statArr;
   alias statArr2 = Template!(2, int).statArr;
   alias statArr1 = Template!(1, int).statArr;
   alias statArr0 = Template!(0, int).statArr;

   // Fine

statArr9().writeln(); // Writes [0, 0, 0, 0, 0, 0, 0, 0, 0]

   statArr3().writeln();       // Writes [0, 0, 0]
   statArr2().writeln();       // Writes [0, 0]
   statArr1().writeln();       // Writes [0]
   statArr0().writeln();       // Writes []

   // 2 bugs
   statArr9()[0].writeln();    // OK : Writes 0
   statArr3()[0].writeln();    // OK : Writes 0

//statArr2()[0].writeln(); // BUG: Internal error: ..\ztc\cgcs.c 344 //statArr1()[0].writeln(); // BUG: Internal error: ..\ztc\cgcs.c 344 //statArr0()[0].writeln(); // OK : Error: array index 0 is out of bounds

                               //      getArr0()[0 .. 0]


   alias dynArr9 = Template!(9, int).dynArr;
   alias dynArr3 = Template!(3, int).dynArr;
   alias dynArr2 = Template!(2, int).dynArr;
   alias dynArr1 = Template!(1, int).dynArr;
   alias dynArr0 = Template!(0, int).dynArr;

   dynArr9()[0].writeln();     // OK: Writes 0
   dynArr3()[0].writeln();     // OK: Writes 0
   dynArr2()[0].writeln();     // OK: Writes 0
   dynArr1()[0].writeln();     // OK: Writes 0

//dynArr0()[0].writeln(); // OK: core.exception.RangeError:

                               //     Range violation

   // Other types
   //Template!(2, bool).statArr()[0].writeln();    // BUG
   //Template!(2, byte).statArr()[0].writeln();    // BUG
   //Template!(2, ubyte).statArr()[0].writeln();   // BUG
   //Template!(2, short).statArr()[0].writeln();   // BUG
   //Template!(2, ushort).statArr()[0].writeln();  // BUG
   //Template!(2, int).statArr()[0].writeln();     // BUG
   //Template!(2, uint).statArr()[0].writeln();    // BUG
   //Template!(2, long).statArr()[0].writeln();    // BUG
   //Template!(2, ulong).statArr()[0].writeln();   // BUG
   //Template!(2, float).statArr()[0].writeln();   // BUG
   //Template!(2, double).statArr()[0].writeln();  // BUG

Template!(2, real).statArr()[0].writeln(); // OK : Writes nan

   //Template!(2, ifloat).statArr()[0].writeln();  // BUG
   //Template!(2, idouble).statArr()[0].writeln(); // BUG

Template!(2, ireal).statArr()[0].writeln(); // OK : Writes inan

   //Template!(2, cfloat).statArr()[0].writeln();  // BUG

Template!(2, cdouble).statArr()[0].writeln(); // OK : Writes nan+nani Template!(2, creal).statArr()[0].writeln(); // OK : Writes nan+nani

   //Template!(2, char).statArr()[0].writeln();    // BUG
   //Template!(2, wchar).statArr()[0].writeln();   // BUG
   //Template!(2, dchar).statArr()[0].writeln();   // BUG

   struct Sint     { int  a; }
   struct Sreal    { real a; }

Template!(2, Sint).statArr()[0].writeln(); // OK : Writes Sint(0) Template!(2, Sreal).statArr()[0].writeln(); // OK : Writes Sreal(nan)

}

Found in both dmd 2.061 and 2.062 for 32- and 64-bit. Other types only tested with dmd 2.062.

I can only replicate one of these bugs in dmd git master x64, the cfloat one. It segfaults during the initialisation of the array.

dmd calls _memset64 to do the initialisation. It seems to (partially) forget that we're in x64 and tries to pass the value to set to (float.nan) on the stack.

The result: A complete mess. Not enough arguments, in the wrong places.

_memset64 gets the right destination pointer but gets the length of the array (2) as the value to write and then tries to initialise RDX number of values. But RDX was never set. Segfault.

Sorry should specify this was on linux x64