obj.erase(it) fails on map about trial.protocol

[vinipsmaker2]

I forgot to fill more details. it is the value returned from obj.key_begin() and obj is dynamic::variable

[breese]

Good catch.

The resolution gives me some headache though. There are two solutions we can use:

1. Add a key_iterator erase(key_iterator) member function.
2. Add an explicit conversion from key_iterator to const_iterator, e.g. key_iterator::base() -> const_iterator.

The first solution enables the user to loop over all elements and erase some of them. That is:

for (auto it = v.key_iterator(); it != v.key_end(); /* increment done below */)
{
  if (should_we_erase_element(*it))
    it = v.erase(it); // erase returns a key_iterator
  else
    ++it;
}

The second solution has the advantage of being more general, so it can be used for other functions, like insert(). However, the example above cannot be used.

I am leaning towards the second solution because:

• We already have the looping problem with iterator erase(const_iterator) which was introduced in C++11 (via proposal N2350.)
• There is precedence for the explicit iterator conversion with reverse_iterator::base().

[vinipsmaker]

Why can't the example compile with solution #2?

[breese]

Because it has type key_iterator and erase returns type iterator.

There is no conversion from iterator to key_iterator because the latter needs an extra member variable.

[breese]

Regarding your comment about parsing the entire input to a dynamic::variable, you will be pleased to know that I am working on a json::parse and json::format that works on dynamic::variable without the need for Boost.Serialization.

[vinipsmaker]

We already have the looping problem with iterator erase(const_iterator) which was introduced in C++11 (via proposal N2350.)

Okay. I've read N2350 (except for the proposed wording chapter).

I think if we apply the same solution here, the proper solution would be to add key_erase to match the begin/key_begin split. key_erase would receive a const_key_iterator and return a key_iterator.

I wouldn't mind to have a overloaded erase function instead typing the key_ thou.

Regarding your comment about parsing the entire input to a dynamic::variable, you will be pleased to know that I am working on a json::parse and json::format that works on dynamic::variable without the need for Boost.Serialization.

Surely I'll enjoy that.

[breese]

I have added a key_iterator::base() function to convert a key_iterator into a const_iterator in db775c4, so now v.erase(kit.base()) is doable.

Regarding the key_erase() function, I would rather have such functionality as algorithms, e.g. like the dynamic::key::count() and dynamic::key::find() algorithms, to prevent a proliferation of the variable interface.

[vinipsmaker]

Okay. Let's leave the issue open for some time in case others want to pop in and add their comments.

I have added a key_iterator::base() function to convert a key_iterator into a const_iterator in db775c4, so now v.erase(kit.base()) is doable.

This will suit me for now. Probably I'll send a PR in the following weeks with one dynamic::key::erase. The question of a sweet spot here between convenience and a clean interface is a tricky one. Not many ready guidelines we can adhere to. But also, it's the problems we always face with novel design (otherwise they wouldn't be novel).