Re: [boost] Better testing; was type_traits_test

From: Beman Dawes (beman_at_[hidden])
Date: 2000-11-02 16:36:26

David Abrahams wrote:

 >Many new enhancements for MSVC6.4 have been added to type_traits 
recently,
 >but many of them fail to compile due to a compiler bug (the short-circuit
 >||/&& evaluation in constant expression problem). I am going to check in 
a
 >fix, but considering that the regression test reports success for
 >type_traits_test even without the fix, I wonder if we're even trying to
 >test
 >the new features. It seems like a mistake to me that we're publishing 
code
 >without tests (to say nothing of code which won't compile).

Agreed.

I'm starting to modify the regression test driver program to execute as 
well as compile the test programs.  The status table will report two 
entries per cell:

   yes/yes == compile and test both returned 0
   yes/no  == compile OK but test did not return 0
   yes/--  == compile OK but test not run; need better test program
   no/--   == compile failed, test not run; no point since compile failed

The "no" entries are bold so they stand out.

While this gets the ball rolling, it isn't totally effective without 
modifying the test programs to always use the return code to indicate 
success or failure.

That in turn would be helped by macros to use instead of assert to pass the 
failure return code.  One macro to exit on failure, another to set global 
state to failure but continue program execution.

I know this isn't as good as DejaGnu, but it is very easy to get started.

Comments?

--Beman