Re: Turning off test cases (was CVS Bind on MSVC failure)

From: Gennadiy E. Rozental (rogeeff_at_[hidden])
Date: 2001-11-29 13:08:08

The proper way would be to separate test module on to two. 
Compilation is the "only point of failure" type. So no need to create 
huge test modules. In an extreme case it could be one file per 
feature. But in a reallity it is convinient to combin test cases that 
are related and will either compile all together or fail all together.

Gennadiy.

--- In boost_at_y..., Darin Adler <darin_at_b...> wrote:
> On 11/29/01 6:47 AM, "Peter Dimov" <pdimov_at_m...> wrote:
> 
> >> ** Metrowerks CodeWarrior
> >> mwcc -o boosttmp.exe -maxerrors 10 -cwd source -I- -
Ic:/boost/site
> >> c:/boost/site/libs/bind/bind_test.cpp
> >> ### mwcc Compiler:
> >> #    File: ..\site\libs\bind\bind_test.cpp
> >> # ----------------------------------------
> >> #     193:      bind<void>(Y(), i, _1, 9, 4)(k);
> >> #   Error:                                 ^
> >> #   illegal use of 'void'
> > 
> > This particular test case never worked with MW. Dave originally 
had it
> > #ifdef-ed out but this is cheating, isn't it? ;-)
> > 
> > How should we proceed in cases like this? #ifdef the test case 
just to see a
> > 'pass'? Or keep the test as-is and be honest with users?
> 
> The downside of leaving the test as-is is that this single problem 
causes
> the entire bind_test to fail. This means that when we make other 
changes to
> bind, the regression test doesn't catch the problems.
> 
> Because of this, I think it's reasonable to turn off parts of the 
tests even
> if that's misleading. I especially like Beman's suggestion of using
> Boost.Config to remind ourselves of the particular shortcoming of 
the
> compiler that requires turning off part of the test. Adding things 
to
> Boost.Config does require a bit of work, though. And we don't have a
> particular BOOST_NO_XXX for this particular problem, do we?
> 
>     -- Darin