Declaration ring_iterator friend operator+=() about sg14

[TBBle]

Looks like we've stumbled over FAQ: Why do I get linker errors when I use template friends?.

[TBBle]

However, I believe the suggested resolution causes every operator+= instantiation to be friends with every ring_iterator instantiation, even if the template arguments don't match. We probably want to use the more-verbose solution from Making New Friends idiom page. (i.e. pre-declare template operator+= and friend it using template type deduction).

[TBBle]

Also, already logged as #77 but never apparently acted upon.

[martinmoene]

Yeah that's the one.

The compile-time (as opposed to link-time) error message (for my incantation of ring_span) with GNU 5.2 was:

prompt>g++ -std=c++17 -O2 -Wall -Wextra -Wno-unused-parameter -o ring-span-lite.t.exe -I../include/nonstd ring-span-lite.t.cpp ring-span.t.cpp   && ring-span-lite.t.exe
In file included from ring-span-lite.t.hpp:12:0,
                 from ring-span-lite.t.cpp:7:
../include/nonstd/ring_span.hpp:518:46: warning: friend declaration 'nonstd::ring_iterator< <template-parameter-1-1>, <anonymous> >::type& nonstd::operator+=(nonstd::ring_iterator< <template-parameter-1-1>, <anonymous> >::type&, int)' declares a non-template function [-Wnon-template-friend]
     friend type& operator+=(type& it, int i) noexcept;
                                              ^
../include/nonstd/ring_span.hpp:518:46: note: (if this is not what you intended, make sure the function template has already been declared and add <> after the function name here)

[martinmoene]

@TBBle Thanks for your notes.

What are reasons to not declare ring_iterator's member operator+=() and operator-=() non-friend like:

type & operator+=( int i ) noexcept
{
    m_idx += i;
    return *this;
}

and operator+() and operator-() in the sg14 (eventually std) namespace like:

template< class Ring, bool is_const >
ring_iterator<Ring,is_const> operator+( ring_iterator<Ring,is_const> it, int i ) noexcept
{
    it += i;
    return it;
}

[TBBle]

I'd defer that question to the paper editors, @hatcat and @Quuxplusone.

Side-note, I've just confirmed this warning on g++ 6.3.0. I intend under #96 to try and get better test coverage in-place so that things like this can't easily slip past.

[TBBle]

Yeah, the free-function operators should be in the sg14 namespace. Almost certainly an oversight, it breaks the build on VS2017 which can't find them, and hence can't use std::distance and hence std::inner_product from the unit tests.

Edit: Different issue. We're actually missing operator-(ring_iterator<Ring,is_const>,ring_iterator<Ring,is_const>).

[TBBle]

Binary Operators in the Stackoverflow C++ Operator mega-answer agrees with @martinmoene's approach:

For the binary arithmetic operators, do not forget to obey the third basic rule operator overloading: If you provide +, also provide +=, if you provide -, do not omit -=, etc. Andrew Koenig is said to have been the first to observe that the compound assignment operators can be used as a base for their non-compound counterparts. That is, operator + is implemented in terms of +=, - is implemented in terms of -= etc.

According to our rules of thumb, + and its companions should be non-members, while their compound assignment counterparts (+= etc.), changing their left argument, should be a member. Here is the exemplary code for += and +, the other binary arithmetic operators should be implemented in the same way:

[TBBle]

My current hack-tree is https://github.com/TBBle/SG14/tree/SG14-build-fix-hacks, which when merged with #95, gets everything building on the platforms I happen to be testing on (four Windows x64 compilers). I simply took the Make New Friends idiom in order to get it to compile. The above-described approach is better.

[Quuxplusone]

Can this issue be closed as "fixed," at this point?