[dmd-internals] runnable/sdtor.d

[Brad Roberts]

On 6/2/2011 9:22 AM, Andrei Alexandrescu wrote:
> On 6/2/11 11:04 AM, Brad Roberts wrote:
>> So, is d (and thus dmd) supposed to deal with that case better?  I assume no since there's no case 1:, and that the test
>> should be changed to:
>>    if (0) switch(1) { A51 a; default: }
> 
> In this case a should never be destroyed even on if (1) as no control flow reaches it.
> 
> (On another vein, I'm surprised that that ever compiled. In C and C++ I recall it's illegal to have a label followed by
> no code.)
> 
>> Of course, the opposite is also interesting in terms of 'in what cases is a actually constructed/destructed':
>>    if (0) switch(1) { default: A51 a; }
> 
> Diff looks good to me, and the last line should indeed call two destructors.
> 
> 
> Andrei

I just double checked, neither gcc or g++ accept an empty default statement.  Obviously dmd does.

Adding two more tests:

   A51_a = 0; if (1) switch(1) { A51 a; default: } assert(A51_a == 1); // should be 0, right?
   A51_a = 0; if (1) switch(1) { default: A51 a; } assert(A51_a == 1);

So.. two bugs.. oh Walter..