Testing array ptr for offset 0...

[Era Scarecrow via Digitalmars-d-learn]

On Thursday, 26 May 2016 at 22:15:42 UTC, ag0aep6g wrote:
> On 05/26/2016 11:13 PM, Era Scarecrow wrote:
>>    void scan(ref Data[][] data, Condition cond) {
>>      foreach(i; ...) {
>>        if (cond)
>>          data[i] ~= ...
>>      }
>>    }
>
> Sorry, I'm still lost. Why can't you do whatever you're doing 
> in opOpAssign directly there, or in a free function? Does the 
> pseudo-array contain any additional data? Would a wrapper 
> around a built-in array not work?

  Well a few things at work. First the array is originally a 
static array, so I pass it as a slice; But if you pass it as a 
slice you get the entire length, and modifying the length 
requires either changing length or appending, both which may 
cause allocation (something I want to avoid). By making my own 
compatible structure I can manage the length and pointer directly 
and precisely.

  Second I would prefer the length to start at 0 and grow UP TO 
the size of the static array. If it was just a single array this 
wouldn't be an issue, but I'm working with a couple hundred of 
these. Plus I don't want to add more data to the memory or 
structure than necessary to treat it as an array after the fact. 
I shouldn't have to.

  Truthfully the same result can be done with shrinking it 
afterwards, but I was trying to see if I could make it any faster 
by managing the pointer and length directly and avoiding the 
large temporaries. So far any speed gains are so minimal it's 
worth dropping.


> Oh, I see. Yeah, the answer is garbage. Seeing the spec section 
> that Adam D. Ruppe linked, the length field actually comes 
> first.
>
> That CTFE errors out when you move the pointer to a variable 
> screams that something is wrong. The code seems to be seriously 
> misinterpreted in CTFE.
>
> I've filed an issue:
> https://issues.dlang.org/show_bug.cgi?id=16081

  Glad I found a bug! :) And all because [].ptr.offsetof wasn't an 
option...

  I just wanted to guarantee it was the correct order (in case it 
gets changed in the future). But ultimately it isn't worth 
pursuing.