[Cython] iterators / libcpp container definitions cleanup, any clues?

[Yury V. Zaytsev]

Hi,

I've been looking into libcpp container definitions and found that they
are kind of messy, in particular, iterators are defined as nested
classes for every container.

This leads to code duplication, and for certain containers some
overloaded methods seem to be missing or conversely declared, but not
provided.

I was trying to have a stab at cleaning this up and offering patches,
but I'm confused as to how to proceed.

At first, I thought that I could turn concepts into a hierarchy of
inherited classes in `iterator.pxd` like _ForwardIterator ->
_BidirectionalIterator -> _RandomAccessIterator and inherit container
iterators from those classes.

It seems that this is a bad idea for several reasons:

1) Apparently, there are some problems with inheritance, overload
resolution and type deduction, e.g. such that Cython deduces
`_RandomAccessIterator` as a result type for a + operation instead of
`vector[int].iterator` (which inherits from _RandomAccessIterator).

2) Cython thinks these are real classes and uses them in explicit
template arguments in generated C++ code.

The combination of (1) and (2) causes a lot of fail.

The only other possibility that crossed my mind so far was to use the
old "include" for code reuse and create files like
_RandomAccessIterator.pxi, _RandomAccessIteratorConst.pxi, etc. and
include their contents in the container definitions.

Sounds terrible, and I guess this wouldn't get merged, even if I get
that to work. Now, are there better approaches that I've overlooked?

-- 
Sincerely yours,
Yury V. Zaytsev