Hello Joaquin, I just came up with a slightly different question in more or less same direction. Can I safely erase data from the index view in a range based for loop. The index container contains only ordered indices (no random access). I saw a suggestion that an iterator is being invalidated and therefore pre-incrementing it for the next step is fine. But what happens if an element erased via its key itself in a range-based for-loop? Would the next increment work fine? Many thanks, Ovanes On Sun, Sep 28, 2014 at 4:03 PM, Aaron Levy wrote:

28.09.2014, 19:06, "Joaquin M Lopez Munoz" :

...

Aaron Levy writes:

...

If I erase elements in a multi_index_container, how are iterators invalidated? 1. For other than random_access index, only iterators to the deleted element? 2. For random_access index, potentially any iterator? ...

For all indices, on every ocassion, only the iterators to the deleted element are invalidated.

...

What are the data structures used internally for this container and its indexes.

* random-access indices use this:

http://bannalia.blogspot.com/2008/08/stable-vectors.html

Aah! Random access and stable, I should have guessed stable_vectors.

And the pointers to the implementation are impressive.

Thanks. _______________________________________________ Boost-users mailing list Boost-users@lists.boost.org http://lists.boost.org/mailman/listinfo.cgi/boost-users

Ovanes Markarian writes:

Hello Joaquin, I just came up with a slightly different question in more or less same direction. Can I safely erase data from the index view in a range based for loop. The index container contains only ordered indices (no random access).

No: the definition of range-based for given in [stmt.ranged] actually *guarantees* that you'd produce undefined behavior (the iterator is post-incremented). Joaquín M López Muñoz Telefónica