Re: [Boost-users] A few ICL questions.

Subject: Re: [Boost-users] A few ICL questions.
From: Joachim Faulhaber (afojgo_at_[hidden])
Date: 2015-11-18 13:06:23

Hi Stephen,

2015-11-17 14:06 GMT+01:00 Stephen Hewitt <shewitt.au_at_[hidden]>:

> I'm new to ICL and have a few questions and a problem.
> Here is my test code which demonstrated (Windows, MSVC 2012):
>
>> // "icl_question.cpp"
>>
>>
>>> #include "precomp.hpp"
>>
>> #include <iostream>
>>
>> #include <utility>
>>
>> #include <functional>
>>
>> #include <boost/cstdint.hpp>
>>
>> #include <boost/icl/closed_interval.hpp>
>>
>> #include <boost/icl/interval_map.hpp>
>>
>>
>>> typedef boost::icl::closed_interval<boost::uint16_t> interval_t;
>>
>>
>>> // Based on the in "boost/icl/functors.hpp".
>>
>> template <typename Type> struct inplace_assign
>>
>>     : public boost::icl::identity_based_inplace_combine<Type>
>>
>> {
>>
>>     typedef inplace_assign<Type> type;
>>
>>
>>>     void operator()(Type& object, const Type& operand) const
>>
>>     {
>>
>>         object = operand;
>>
>>     }
>>
>> };
>>
>>
>>> // Why is the 7 template parameter not a template template parameter?
>>
>> // The code in "boost/icl/detail/design_config.hpp" does not define the
>>
>> // ICL_USE_INTERVAL_TEMPLATE_TYPE macro (I assume intentionally) which, if
>>
>> // defined, would make it so.
>>
>> typedef boost::icl::interval_map
>>
>>                         <
>>
>>                         boost::uint16_t,
>>
>>                         const char*,
>>
>>                         boost::icl::partial_absorber,
>>
>>                         std::less,
>>
>>                         inplace_assign,
>>
>>                         boost::icl::inter_section,
>>
>>                         interval_t
>>
>>                         > map_t;
>>
>>
>>> int main(int argc, char *argv[])
>>
>> {
>>
>>     using namespace std;
>>
>>
>>>     map_t mem;
>>
>>     // The commented line two down is what I intend, but it asserts at
>>> runtime.
>>
>>     // This one works.
>>
>>     mem += make_pair(interval_t(0, 0xfffe), "RAM");
>>
>>     //mem += make_pair(interval_t(0, 0xffff), "RAM");
>>
>>     mem += make_pair(interval_t(0xa000, 0xbfff), "Basic ROM");
>>
>>
>>>     cout << hex << mem << endl;
>>
>>
>>>     return 0;
>>
>> }
>>
>>
> The first question is about the aggregate on collision functionality. I
> would just like to replace. I have implemented *inplace_assign *which
> does this but it seems strange that this isn't included. Am I missing
> something here (I fear I may well be)?
>

Interval_maps of the icl are not only containers but also aggregation
machines. So the default of aggregation functors are "addition" of
associated elements and its inversion "subtraction". Other functors are
provided for other kinds of aggregation. The functor inplace_identity
results, when instantiated in the behavior of std::maps that keep their
values on insertion of values for already inserted keys. You will get the
same semantics using the member function insert. The functions to just
"set" or "overwrite" values is available as member function set. See
http://www.boost.org/doc/libs/1_59_0/libs/icl/doc/html/boost_icl/function_reference/insertion.html
.

Adding the "set" or "replace" functionality to the library via functor
would be an addition that makes sense. But you can use the icl pretty well
without it. Also you can add the functor yourself to your code base when
you would like to use it as you just demonstrated.

>
> The second question is why* interval_map*'s 7th template parameter (
> *Interval*) is not a template template parameter. The code in
> "boost/icl/detail/design_config.hpp" does not define the *ICL_USE_INTERVAL_TEMPLATE_TYPE
> *macro (I assume intentionally) which, if defined, would make it so.
>

It needs to be a type. Otherwise the library would not be able to use user
defined non-generic interval types. See also
http://www.boost.org/doc/libs/1_59_0/libs/icl/doc/html/boost_icl/examples/custom_interval.html

>
> The third question is about a runtime error I'm getting in the above code.
> The comment in *main *shows the scenario. Basically if I use *0xffff *as
> the end of the first range I get an ASSERT at runtime whereas 0xfffe works
> fine.
>

You can use icl interval containers with limited integral numeric
domain-types safely on intervals using values from ++std::min<T>() to
--std::max<T>(): Two values less. So if you want to manage a domain of
values like in your example, you'd need to use a larger value type. It is
possible to make the code handle those corner cases correctly (yet not
trivial, since it is generic code). For achieving that the most frequently
called functions would all need to make additional checks making the code
ugly and more slowly. And yes, this information should be added to the
documentation.

>
> Finally, the documentation seems a little sparse on the function of the
> *Section* template parameter. Any clarification on that would be
> appreciated.
>

May be you'll find these sections helpful.

http://www.boost.org/doc/libs/1_59_0/libs/icl/doc/html/boost_icl/semantics/collectors__maps_of_sets.html
http://www.boost.org/doc/libs/1_59_0/libs/icl/doc/html/boost_icl/semantics/quantifiers__maps_of_numbers.html

Cheers,
Joachim

>
> _______________________________________________
> Boost-users mailing list
> Boost-users_at_[hidden]
> http://lists.boost.org/mailman/listinfo.cgi/boost-users
>

-- 
Interval Container Library [Boost.Icl]
http://www.joachim-faulhaber.de