--- In boost@[...].. , Kresimir Fresl <fresl@[...]..>  wrote:
> 
>  jhrwalter wrote:
> 
>  [...]
>  > (but I'm unsure, if valarray<> is a legal array_type of
>  > numerics::vector currently ;-).
> 
>  It is ... with few minor modifications (at least with g++ 3.0.4).
> 
>  std::valarray<> does not have `iterator' & `const_iterator'
>  types.

Yes, I remember, that I stopped seeing std::valarray as an option,
when I realized that and tried to imagine, how to work around it
under MSVC ;-)

>  Therefore `iterator_type' & `const_iterator_type'
>  in numerics::vector<> can't be defined. But they are not
>  needed if `indexed_iterator<>' is used instead of nested
>  iterator classes.
> 
>  So, I first added
> 
>  #ifndef NUMERICS_USE_INDEXED_ITERATOR
>  typedef typename A::const_iterator const_iterator_type;
>  typedef typename A::iterator iterator_type;
>  #endif
> 
>  in numerics::vector<>.

I imagine, that we could introduce here some (array_type_?)traits
classes, which define the iterators for valarray via partial
specialization.

>  But to compile some simple examples with macro
>  NUMERICS_USE_INDEXED_ITERATOR defined,
>  I had to add few typename's in `vector_pr.h':
> 
>  line 184:
>  NUMERICS_TYPENAME vector_type::iterator::iterator_category>
iterator;
>  instead of:
>  vector_type::iterator::iterator_category> iterator;
> 
>  line 186:
>  NUMERICS_TYPENAME
vector_type::const_iterator::iterator_category> 
>  const_iterator;
> 
>  line 664:
>  NUMERICS_TYPENAME vector_type::iterator::iterator_category>
iterator;
> 
>  and line 666:
>  NUMERICS_TYPENAME
vector_type::const_iterator::iterator_category> 
>  const_iterator;

Yep, one of my usual problems: I'll add these, too. Thanks.

>  Now this works:
> 
>  #include <valarray>
>  #include <boost/numerics/ublas/vector.h>
> 
>  typedef numerics::vector<double, std::valarray<double> > vector;
>  vector v1(10), v2(10), v3(10);
>  v3 = v1 + v2;

Ok, if all agree, that it would be useful to have std::valarray as
another array_type option, I'll try to add some traits logic to
support it at least under the more standard conforming compilers.

>  BTW, did you measure what is more efficient: indexed_iterator<>
>  or iterator classes nested in numerics::vector<>?

Nested pointer based iterators result usually in tighter inner loops
than something like indexed_iterator<>  on the compilers I've seen so
far. They are only slower for small vector & matrix sizes, so that we
switch to assignment via subscription in such cases.

> 
>  Sincerely,
> 
>  fres

Regards

Joerg