Re: Convention for defining/calling type-based test cases

[Martin Sebor]

Eric Lemings wrote:

I've noticed a lot of test employing the following convention:

That's a workaround for compilers that can't handle this kind
of thing (see, for example, the EXPLICIT_ARG.cpp config test).
Some older compilers that did grok the explicit argument list
weren't able to overload on it. I don't think we have any of
these compilers on our matrix anymore so we can probably stop
using the workaround and maybe even delete the config test.

Btw., some time ago, we briefly discussed compiling a list
of essential C++ features that we should be able to assume
of every compiler. Seems that explicit function template
argument lists should be on that list.

See the thread Re: list of required C++ features:
  http://stdcxx.markmail.org/message/zuawhacu2cihto4y

Martin

template < typename T >

test_case_X (T /*unused*/)
{
    ...
}

static void

run_test ()
{
    test_case_X (char ());
}

The type argument is typically only used for implicit binding and unused

with the test case function itself.  So couldn't the test case function
be written without the argument and the type specified explicitly at the
point of call?  Example:

template < typename T >
test_case_X ()
{
    ...
}

static void
run_test ()
{
    test_case_X <char> ();
}

This makes it a little more evident that the test case function only
tests the type itself and not instances of the type.

Brad.