Re: [Boost-users] Multi_array: A "proper" iterator over a N>1 Multi_array?

Upon reflection I realized that a) My code was broken b) origin() returns an element* and not any smarter iterator, so that layout is important c) I would propose as a solution a call that returns an iterator pointing to the first element of a multi array and when incremented goes through all the elements one by one all the way to the end something like for (iter=marray.element_begin(); iter !=marray.element_end(); iter ++) { *iter+=1; } would increment all the elements of the multi_array marray by 1. This should work for views and independent of number of dimensions and layout. I think it would be easily implementable as a skip iterator where the increment operator increments a *element by a number that's the product of the appropriate strides. Dereferencing semantics is the same as element*. Now everything rests of what "appropriate" means, but I guess whoever implemented how to increment any index in a view knows how to pull this off as well -- that is Ron, any comments? I think this would be a powerful extension. Once this is in place there would be no need for specialized algorithms -- for_each, transform and even binary_search would do exactly what people want them to do, like in for_each(marray.element_begin(); marray.element_end(), _1++) equivalent to the loop above. Another possibility that reminds me of the statistical language R is that we could have a call as_vector that flattens an multi array into a 1D view. Then begin_element above would be equivalent to begin() on as_vector(marray). Antonio On 4/25/06, Bruno Martínez wrote:

On Tue, 25 Apr 2006 23:11:26 -0300, Greg Link wrote:

...

I would agree with you there - a 'foreach' would do what I need as well, but as I'm not 100% familiar with the way libraries such as boost are implemented, I worry that making your own control structure is much more difficult than making an accessor/modifier pair. I can't even imagine the syntax needed to convert the following into a theoretical boost-defined 'foreach'

I didn't mean new sintax, but a std::for_each work-alike.

...

double previous = 0; double accumulator = 0; for(iterator_t myIterator = m_array.data(); myIterator != m_array.end (); myIterator++) { accumulator += (*myIterator); (*myIterator) *= previous accumulator = (previous > 7) ? (do_function(accumulator)) : (0); }

You can do that with Boost.Lambda:

#include #include #include #include <algorithm>

int main() { double previous = 0; double accumulator = 0; double d[3] = {1, 2, 3};

using boost::lambda::_1; using boost::lambda::if_then_else_return; using boost::lambda::var; using boost::lambda::bind; std::for_each(&d[0], &d[3], ( accumulator += _1, _1 *= previous, var(accumulator) = if_then_else_return(var(previous) > 7, bind(&do_function, accumulator), 0))); }

I spend some time today writing a std::for_each-like function for multiarrays. This works correctly for subviews, but isn't as convenient as an iterator. You can't iterate through two multiarrays at the same time, for example. I only tested with MSVC8, so there may be some typenames missing. Here it is:

template struct for_each_impl { F f; for_each_impl(F f) : f(f) {}

void operator()(MA& ma) { std::for_each(ma.begin(), ma.end(), for_each_impl(f)); }

};

template struct for_each_impl { F f; for_each_impl(F f) : f(f) {}

void operator()(Ref r) { f(r); } };

template void for_each(MA& ma, F f) { for_each_impl impl(f); impl(ma); }

And a test:

typedef boost::multi_array array_type; typedef array_type::index index; array_type myarray(boost::extents[3][4][2]); typedef array_type::index_range range; array_type::array_view<3>::type myview = myarray[ boost::indices[range(0,2)][range(1,3)][range(0,4,2)] ]; for_each(myview, _1 = 56);

Hope that helps.

Bruno

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