Re: [Boost-users] [string_algo] split input parameter not const?

On Tue, Dec 13, 2005 at 01:00:05PM -0600, Thore Karlsen wrote:

On Tue, 13 Dec 2005 19:24:49 +0100, Pavol Droba wrote:

...

...

split() is declared as follows:

template SequenceSequenceT & split(SequenceSequenceT & Result, RangeT & Input, PredicateT Pred, token_compress_mode_type eCompress = token_compress_off);

Is there a reason why Input isn't const? It would be nice to be able to construct an iterator range in the call to split(), like this:

typedef iterator_rangestring::const_iterator range; vector<range> inputs; split(inputs, range(Begin, End), is_any_of("\r\n"), token_compress_on);

It's the same for find_all(), ifind_all(), and probably others, so there might be a good reason for it that I'm missing.

...

The const is not missing there. Because the function is templated, const is added to the calculated type.

Example:

std::string str("Hello world"); const std::string cstr("Good bye world");

split(res, str, pred); // RangeT is substituted to std::string split(res, cstr, pred); // RangeT is substituted to const std::string

So you can use the code you have described without problems.

Hmm.. I can use a const value just fine, but I can't use a temporary. At least not in VC++ 8.0 with the warning level set to 4 (I compile all my code at warning level 4). I get a warning C4239:

warning C4239: nonstandard extension used : 'argument' : conversion from 'boost::iterator_range<IteratorT>' to 'range &' with [

IteratorT=std::_String_const_iterator ] A non-const reference may only be bound to an lvalue

I see your problem now. The restriction of temporaries is important in find algorithms. Since they return 'a reference' (in the form of iterator_range) into the input, passing a temporary there is dangerous. split can be used this way too: vectorstring::iterator > result; string str("hello world"); split(result, str, is_space); I see no universal solution here until r-value reference will be added to the language. Until then I prefer the 'safer' one (given the fact, that it is not very restrictive and can be easily workarounded) Regards, Pavol