std.string and ranges

[Andrei Alexandrescu]

Frits van Bommel wrote:
> Andrei Alexandrescu wrote:
>> dsimcha wrote:
>>> Is std.string by any chance going to see any ripple effects from 
>>> ranges?  I
>>> want to write some patches to make it work with CTFE, which would be 
>>> useful
>>> for some compile time mixin generation, but I don't want to waste 
>>> time working
>>> on a module that Andrei is going to completely revamp anyway.
>>
>> Yes, std.string must be revamped too. You may want to hold onto your 
>> work for a little while more.
> 
> Then I suppose the next question should be:
> will a char[] be a range of char, or a range of dchar? (Let's keep wchar 
> out of it)

Glad you asked. Given type C[] in { char[], wchar[], dchar[] }, here are 
the ranges that can be used to iterate them:

a) byCharacter: a bidirectional range with immutable elements. The 
element type is always dchar, and the range goes over the full 
characters in the string. If C == dchar, the range has random access.

b) byCharacterWritable: a more expensive bidirectional range that still 
has element type dchar but holds a reference to the original string such 
that it can go back and change it. (This is tricky because writing back 
a dchar into the string may change its length.) The language needs the 
alias this feature and solid proxy objects to be able to implement this. 
If C == dchar, the range has random access and fast writes.

c) byNaturalWidth: a random-access range that iterates the raw elements.

> On the one hand, char is the logical option since that's what the array 
> actually contains. Also, this better allows for mutability (since 
> encoding dchars can lead to not having enough space to store 
> replacements in the current position).
> 
> On the other hand, dchars would probably be more useful in many situations.
> 
> Maybe an adapter could be written, taking a char[] (or any range of 
> chars/wchars) and acting as a range of dchars?
> Perhaps also similar adapters to convert to chars and wchars?

Quite so. The important things are (1) only bidirectional, not random, 
access is affordable when reading one dchar at a time from char[] or 
wchar[]; (b) proxy objects are important in assuring proper mutability.


Andrei