Re: [boost] [Review] ITL review starts today, February 18th

Subject: Re: [boost] [Review] ITL review starts today, February 18th
From: Phil Endecott (spam_from_boost_dev_at_[hidden])
Date: 2010-03-01 09:26:21

Barend Gehrels wrote:
>> You are just storing the number of people living on any particular 
>> date.  Try changing your code to record all of the names, e.g. 
>> something like:
> Yes, I did this in my review. The first version of my program listed 
> like this:
>     typedef boost::itl::set<std::string> itl_set_string;
>     typedef boost::itl::interval_map<int, itl_set_string> itl_map;
>     (...)
>     itl_set_string person;
>     person.insert(p.indi.name);
>
>     int death_date = p.indi.death.date.hasdate ? p.indi.death.date.year 
> : 9999;
>     livetimes.add(std::make_pair( 
> boost::itl::interval<int>::rightopen(p.indi.birth.date.year, 
> death_date), person));
>     (...)
>     itl_set_string const& who = it->second;
>     stream << when.first() << " - " << when.upper() << " : " << 
> who.size() << std::endl;

> Also this first test did have no noticeable delay. It is subsecond, and 
> I did also write the names of all these persons in this test which 
> should not be timed, so it is not problematic.

Because you're using only the years of birth and death, this is limited 
to O(N_YEARS * N_PEOPLE).  With N_YEARS ~ hundreds and N_PEOPLE ~ 
thousands, I can believe this would run in a short time (modern 
computers are fast).

If you have time, I would be interested to know how the performance 
changes if you use a higher-resolution date type, e.g. integer number 
of days.

Regards,  Phil.