Re: Complex compiler/linker tests

[Liviu Nicoara]

Hi Ralf,

I am trying to understand what you mean by intuitive? How is it 
intuitive for example a test case which tests for the existence of a 
header file?

I believe there is hardly anything intuitive about the test case I 
presented. A test case, by definition, tests a marginal use-case of the 
language and in the absence of the rest of the application (from which 
the test-case has been carved) some parts or all of it may not make sense.

The example is useful to me in that it allows me to detect a particular 
behaviour of a C++ compiler/linker which allows me to craft my code 
accordingly. In what regards portability I would expect the [portable] 
infrastructure I use to provide me the means to detect the unportable 
language constructs or non-compliant compiler behaviour across the 
platforms I am porting on.

The example I provided is representative for a real-world situation: 
Microsoft Visual C++ 6.0/7.0/7.1 and Visual Age for C++ 5.x/6.x/7.0 both 
violate ODR in certain situations. If you know a better or simpler way 
to detect the behaviour in my example I would be happy to learn about 
it, but just dismissing the example as not being representative for a 
real-world situation is not very helpful.

These being said, I appreciate you're taking the time to respond.

Thanks.

Liviu

Ralf Wildenhues wrote:
> Hi Liviu,
>
> * Liviu Nicoara wrote on Wed, Jun 29, 2005 at 06:12:36PM CEST:
>
> > One trivial example would be a test to detect whether or not a compiler 
> > collapses static locals in inline functions occurring in both library 
> > and user program. It would require a library and a program, e.g.:
>
>
> Ouch.  While I see why you need more than one translation unit
> here, the example seems very unfortunate to me: it's a very ugly
> and non-intuitive way of coding IMVHO.  Could you provide a useful
> example, i.e. one for which it would make sense to adapt a piece
> of intended-to-be-portable software to?
>
> To put it another way: with that example, you did not convince me
> of the usefulness.  IME the autoconf developers are less easily
> convinced than I am.  :)
>
> Regards,
> Ralf