Jeremy Maitin-Shepard ha escrito:
Joaquín Mª López Muñoz <joaquin@tid.es> writes:
[snip]
Why not use an MPL sequence of indexes in the first place?
Well, this has been previously discussed before. The rationale is that I'd like to isolate common users from MPL as much as possible. Imagine the tutorial somehow pointed to the MPL reference for a forward compile-time sequence: IMHO, many readers would scare away. Which percentage of *users* of Boost are acquainted with such a complex library as MPL? MPL is an extremely powerful library, but I honestly don't expect it to to be part of the common knowledge shared by the C++ community.
But aren't you just effectively aliasing or reproducing mpl::list in your "index_list"?
Yes, the point is that users need not even know what MPL is.
I do not see that as a very good solution. If you want to use a type users might be more familiar with, you could allow them to use boost::tuple to specify the index types.
Very first versions of the container (about a year ago) did use boost::tuple, but: 1. Implementation is much harder, as tuples are not MPL sequences, a feature the lib internally relies in for building the container out of the indices. 2. More importantly, tuples do not feel as compile-time specifications devices: they are rather run-time objects, meant to be constructed and passed around. Index specifiers are purely syntactic constucts, they need not even be copyable. The "indexed_by" renaming enforces this syntactic feel, that's why I like it.
As I said in a previous message though, what I would really like to see would be allowing specifying the indices directly as parameters to the container template, as a convenience syntax.
I answered that mail, were the reasons given there not convincing to you? Joaquín M López Muñoz Telefónica, Investigación y Desarrollo