Raising the bar on Phobos unittest coverage

[monarch_dodra]

On Monday, 18 March 2013 at 20:30:32 UTC, Jonathan M Davis wrote:
> On Monday, March 18, 2013 12:09:47 Walter Bright wrote:
>> On 3/18/2013 11:12 AM, Jonathan M Davis wrote:
>> > On Monday, March 18, 2013 17:34:12 Martin Nowak wrote:
>> >> Can someone look into std.file's unittests. They use 60% 
>> >> (32s/50s) of
>> >> the unittest RUNtime on my machine.
>> > 
>> > My first guess would be that you're running a different OS 
>> > from Walter and
>> > that OS-specific code counts as not being run when you run 
>> > it on a
>> > different OS.
>> Code that is statically compiled out is not counted as 
>> executable code by
>> the coverage analyzer.
>
> Then it was a bad guess. Good to know (and better behavior that 
> way really).

Is it though? It basically means that if you don't instantiate a
template, it statically doesn't get compiled, so it is not
counted:

main.d:
//--------
         |void foo(T)()
         |{
         |   int i = 0;
         |   ++i;
         |   ++i;
         |   ++i;
         |   ++i;
         |   ++i;
         |   ++i;
         |   ++i;
         |   ++i;
         |   ++i;
         |   ++i;
         |   ++i;
         |   ++i;
         |   ++i;
         |   ++i;
         |   ++i;
         |   ++i;
         |   ++i;
         |   ++i;
         |   ++i;
         |   ++i;
         |}
         |
         |void main()
         |{
        1|    int i;
         |}
main.d is 100% covered
//--------

main.d 100% covered my ass.

I also call BS on std.algorithm's 95% coverage. I looked at the
.lst. It is *roughly* covered, but not 95% covered.

We need to be able to differentiate code that is stripped out due
to "version" blocks (which legitimately doesn't count for the
current build), and code that simply didn't get compiled because
it WASN'T COVERED.