Re: [Boost-users] case insensitive less_than string comparison

From: Pavol Droba (droba_at_[hidden])
Date: 2006-03-03 16:11:39

Hi,

On Fri, Mar 03, 2006 at 02:42:14PM +0100, Olaf van der Spek wrote:
> Pavol Droba wrote:
> > The functionality is in cvs. I have added lexicographical_compare (also in i variant)
> > and comparison predicates is_(i)less, is_not_(i)greater.
> > 
> > Check it out.
> 
> struct is_not_greater
> {
>      template< typename T1, typename T2 >
>          bool operator()( const T1& Arg1, const T2& Arg2 ) const
>      {
>          return Arg1>=Arg2;
>      }
> };
> 
> >=, isn't that is_not_less?

Oh, this is what happens when the regression tests are not complete.
Thanks for notice, you are right of couse. The operator should be reversed.

> 
> struct is_iless
> {
>      is_iless( const std::locale& Loc=std::locale() ) :
>          m_Loc( Loc ) {}
> 
>      template< typename T1, typename T2 >
>          bool operator()( const T1& Arg1, const T2& Arg2 ) const
>      {
>          #if defined(__BORLANDC__) && (__BORLANDC__ >= 0x560) &&
> (__BORLANDC__ <= 0x564) && !defined(_USE_OLD_RW_STL)
>              return std::toupper(Arg1)<std::toupper(Arg2);
>          #else
>              return std::toupper(Arg1,m_Loc)<std::toupper(Arg2,m_Loc);
>          #endif
>      }
> }
> 
> What about 'A' vs 'a'? Shouldn't 'A' < 'a'?

Hmm, interesting point. But it is not fully correct. Take this example:

str1="Aab"
str2="aAa"

in my opinion following should hold str1>str2, however if 'A' < 'a' then the comparison 
would yield the oposite.
Such a comparison make sense only if both strings are equal (except the case).

Current implementation of ilexicographical_compare does not perform this extra step
since it tries to be compliant with std::lexicographical_compare.

> 
> I'm also wondering how expensive std::toupper is. Would it make sense to
> do a == first to avoid the std::toupper calls in certain cases?
> 

Well, this is a more complicated point. Small optimization you are proposing
does not really make any difference. I have read a paper (I don't remember where),
where author proposes to cache the results of tolower during the comparison. 
This approch is complex and it effectively disables the possibility to 
use user-defined comparison predicates. So I decided to go for a simple solution.

Best regards,
Pavol