[dmd-internals] What is the point of runnable/testdate.d?

[Brad Roberts]

On Sunday, August 14, 2011 7:13:35 PM, Walter Bright wrote:
> On 8/14/2011 1:14 PM, Brad Roberts wrote:
>> Changing tests is dangerous.  Without going back to the original code to make sure the original bug is still caught and
>> tested for by the moved code or the altered to be a even more reduced test case, there's a real risk of a reduction in
>> the quality of the test suite.  I'd personally LOVE it if not a single phobos import existed in any of the dmd test
>> suite.  Possibly even true for druntime imports in the dmd test suite.  However, it's not clear that the benefits
>> justify the risk.
>>
> 
> Some library tests are probably infeasible to do as unit tests, such as:
> 
> 1. tests that involve a whole bunch of modules working together (integration tests)
> 
> 2. ones that are far more exhaustive than would be suitable for inclusion in the library source code.
> 
> 3. The regexp tests, for another example, are under a different license than Phobos', so I think it wo
uld be
> inappropriate to fold them into std.regexp.d.
> 
> 4. Tests that are cobbled up from some interesting application that may be completely unsuitable as being part of the
> library source code.

None of that justifies them being part of the dmd test suite.  They're 
justifications for not being in unittest blocks inside the specific 
files in phobos.