Proposal for SentinelInputRange

[Walter Bright]

On 2/28/2013 1:19 AM, Peter Alexander wrote:
> On Thursday, 28 February 2013 at 01:12:19 UTC, Walter Bright wrote:
>> Motivation:
>> 1. easy conversion of C strings to ranges
>
> This is already easy.
>
> struct CString
> {
>      @property char front() { return *p; }
>      void popFront() { ++p; }
>      @property bool empty() const { return !*p; }
>      char* p;
> }
>
> (Yep, rubbish implementation, missing save, constructors etc., but you get the
> idea).

Notice the double read of *p. This is what SentinelInputRange is trying to fix.


>> 2. necessary for a fast implementation of a lexer
>
> This is very domain specific. Is a whole new range concept in the standard
> library necessary?

I've posted in this thread several use cases.


> My worry here is that we are setting a precedent for adding new range concepts.
> If the only justification needed is that is saves a single operation in some
> niche area of computing then we are opening the door to a LOT of different range
> concepts.

There are many cases where speed is a very big deal.


> Are there any other REAL uses for this other than in one line of a lexer
> implementation? I think inclusions into the standard library should require at
> least several distinct and realistic use cases.

I've posted several - including one in Phobos.