Re: [boost] Fw: Is there interest for iterator based conversion library? attempt #2

From: Alexander A Potocki (sane_at_[hidden])
Date: 2006-07-20 09:07:28

Shunsuke Sogame wrote:
> Your code looks similar to Boost.Serialization Data Flow.
> http://www.boost.org/libs/serialization/doc/dataflow.html

Yes, something is similar. As Dataflow as my library provide the set of 
basic transformation primitives and give possibility to compose them to 
create a complex compound converter.

But there are several distinctions.

1 Create and instantiation of conversion entity in my library takes place in 
the same time. Dataflow at first create compound iterator type and then 
instantiate it. Such instantiation is simple and not clear.

2 My library provides as input as output conversion iterators, but Dataflow, 
it seems, provides only input iterator. Taking in mind "outputness" or 
"inputness" of iterator  I distinguish transformation algorithms for 
concrete conversion. For output iterator it's one, for input iterator it's 
slightly (or not slightly) other.

3. Shifting operation is not clear at all in Dataflow.

> But, we have now Boost.Range.
> Assume 'range_copy(rng,it)' just calls
> 'std::copy(boost::begin(rng),boost::end(rng),it)'.
>
>     range_copy(make_u32_to_u8_iterator_range(src), it);
>
> There is no typedef's.
> You can look into <boost/regex/pending/unicode_iterator.hpp>, which
> shows such implementations.

It's not clear for me. range_copy  is a good method, but 
make_XXX_to_YYY_iterator_range will be implemented using the same typedefs. 
And what is more how many make_ZZZ functions will be provided? I guess there 
are very many combinations of possible XXX and YYY.