Reduced test case  from std.math. *ps is 0x800F_FFFF_FFFF_FFFF, yet
the if() branch is not taken.
This one looks as though it might be the root cause of many failures.

void check(double x)
{
ulong *ps = cast(ulong *)&x;
if (*ps & 0x8000_0000_0000_0000) {}
else assert(x>=0);
}

void main()
{
  check(-double.min_normal*double.epsilon);
}

Works when I try it. Do you have the latest dmd?

Don Clugston wrote:
> Reduced test case  from std.math. *ps is 0x800F_FFFF_FFFF_FFFF, yet
> the if() branch is not taken.
> This one looks as though it might be the root cause of many failures.
>
> void check(double x)
> {
> ulong *ps = cast(ulong *)&x;
> if (*ps & 0x8000_0000_0000_0000) {}
> else assert(x>=0);
> }
>
> void main()
> {
>   check(-double.min_normal*double.epsilon);
> }
>
> 

Aargh, it was an incorrect reduction. Here's an larger case which
fails with the latest DMD, Phobos, and druntime.
(Reducing test cases is a bit painful through a remote connection).
-------------
double nextUpw(double x) @trusted nothrow
{
    ulong *ps = cast(ulong *)&x;

    if ((*ps & 0x7FF0_0000_0000_0000) == 0x7FF0_0000_0000_0000) {
        if (x == -x.infinity) return  -x.max;
        return x;
    }
    if (!(*ps & 0x8000_0000_0000_0000L))  {
         assert(x>=0);
    }
    return x;
}

void main()
{
  nextUpw(-double.min_normal*(1.0 - double.epsilon));
}
=========

On 3 January 2011 23:37, Walter Bright <walter at digitalmars.com> wrote:
> Works when I try it. Do you have the latest dmd?
>
> Don Clugston wrote:
>>
>> Reduced test case ?from std.math. *ps is 0x800F_FFFF_FFFF_FFFF, yet
>> the if() branch is not taken.
>> This one looks as though it might be the root cause of many failures.
>>
>> void check(double x)
>> {
>> ulong *ps = cast(ulong *)&x;
>> if (*ps & 0x8000_0000_0000_0000) {}
>> else assert(x>=0);
>> }
>>
>> void main()
>> {
>> ?check(-double.min_normal*double.epsilon);
>> }
>>
>>
>
> _______________________________________________
> dmd-internals mailing list
> dmd-internals at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/dmd-internals
>

Good, I can reproduce this one.

BTW, you'll need a 64 bit machine sooner or later! I bought one for about $300 last month (recycling keyboard, mouse, display, and hard disk from an older machine).

Don Clugston wrote:
> Aargh, it was an incorrect reduction. Here's an larger case which
> fails with the latest DMD, Phobos, and druntime.
> (Reducing test cases is a bit painful through a remote connection).
> -------------
> double nextUpw(double x) @trusted nothrow
> {
>     ulong *ps = cast(ulong *)&x;
>
>     if ((*ps & 0x7FF0_0000_0000_0000) == 0x7FF0_0000_0000_0000) {
>         if (x == -x.infinity) return  -x.max;
>         return x;
>     }
>     if (!(*ps & 0x8000_0000_0000_0000L))  {
>          assert(x>=0);
>     }
>     return x;
> }
>
> void main()
> {
>   nextUpw(-double.min_normal*(1.0 - double.epsilon));
> }
> =========
>
> On 3 January 2011 23:37, Walter Bright <walter at digitalmars.com> wrote:
> 
>> Works when I try it. Do you have the latest dmd?
>>
>> Don Clugston wrote:
>> 
>>> Reduced test case  from std.math. *ps is 0x800F_FFFF_FFFF_FFFF, yet
>>> the if() branch is not taken.
>>> This one looks as though it might be the root cause of many failures.
>>>
>>> void check(double x)
>>> {
>>> ulong *ps = cast(ulong *)&x;
>>> if (*ps & 0x8000_0000_0000_0000) {}
>>> else assert(x>=0);
>>> }
>>>
>>> void main()
>>> {
>>>  check(-double.min_normal*double.epsilon);
>>> }
>>>
>>>
>>> 
>> _______________________________________________
>> dmd-internals mailing list
>> dmd-internals at puremagic.com
>> http://lists.puremagic.com/mailman/listinfo/dmd-internals
>>
>> 
> _______________________________________________
> dmd-internals mailing list
> dmd-internals at puremagic.com
> http://lists.puremagic.com/mailman/listinfo/dmd-internals
>
>
>