Re: [boost] Re: Type Traits and Test Library

From: John Maddock (john_at_[hidden])
Date: 2004-01-24 07:53:26

> > Do you have any specific compile time test tools in mind? Does it worth
> > reusing?

[ code snipped ]

The rationale for doing it the current was is as follows:

1) I can build all of the type traits tests as one big test program in my
IDE and run all the tests at once - this is a heck of a lot quicker than
running all the individual tests.

2) The test failures are reported at runtime - this allows multiple failures
to be fixed all at once, where as a compile time failure would typically
report only one failure (the first on).

3) The information printed out on failure is quite extensive, and generally
more informative than some compiler's error messages, especially those that
don't report template instantiation back-traces.

4) The end user gets information on which cases fail with their compiler
(all of them, not just the first to cause a compiler error), and can decide
if these are important or not for them.

5) The tests can handle "desired values", things we would like to be
auto-detected but can't currently on most compilers: is_POD, has_trivial_*
etc.

Finally, the unit test lib seems to be rebuilt every single damn time by
cygwin bjam *even when no headers have changed at all*, frustrating :-(

John.