[boost] Re: How does boost::function deal with default arguments?

From: David Abrahams (dave_at_[hidden])
Date: 2003-12-26 18:51:20

Douglas Paul Gregor <gregod_at_[hidden]> writes:

> On Fri, 26 Dec 2003, David Abrahams wrote:
>> Douglas Paul Gregor <gregod_at_[hidden]> writes:
>>
>> > On Thu, 25 Dec 2003, Jim Apple wrote:
>> >
>> >> Douglas Paul Gregor wrote:
>> >>
>> >> > This should not compile.
>> >>
>> >> Compiled for me with g++ 3.3.1, Intel 8 on linux.  Here are some diffs
>> >> to note this compiler error:
>> >
>> > Thanks, I'll add it to the testsuite.
>>
>> I'm a little bit concerned about having this is the Boost.Function
>> test suite.  It isn't really a Boost.Function-specific problem at all;
>> the compiler bug doesn't affect the usability of Boost.Function,
>
> It does affect the usability of Boost.Function, because users of
> these compilers may try it in some small test case and think that
> Boost.Function (unlike normal function pointers) can handle default
> arguments.

Sure, but it also affects Boost.Python, and probably quite a few other
libraries, in a similar way.  I don't think every library which takes
specific function pointer types as arguments should test for this bug.

People could also use type traits on template-dependent arguments
without `typename' with vc6 and be misled into thinking that you don't
need typename when using the type traits.

>> and it interacts with *any* code which does function
>> pointer/reference assignment.  Tests like this one, which really
>> just look for specific compiler bugs, belong in the config library
>> test suite.  We should also have a corresponding config macro.
>
> Why put it into the config library test suite? Do you have any use for
> this as a macro?

No, but I don't know any other place where we just test for compiler
bugs.

I don't feel very strongly about this, but it does seem like it's just
going to make the library look broken when it's actually fully
functional.

-- 
Dave Abrahams
Boost Consulting
www.boost-consulting.com