Re: Implementing NSString getBytes

[Luboš Doležel]

On 06/09/2013 04:30 PM, Fred Kiefer wrote:
> I like the idea of adding this functionality. You actual implementation
> wont do as you already noticed. We also cannot use iconv()
> unconditionally. I remember back in time when I first added a call to
> iconv() to GNUstep we also had to introduce the configure checks for it.
> And of course provide a fall back implementation for that rare case
> where it isn't available. You should also have a second look at
> GSFromUnicode() to notice it is trying a much better job at providing a
> lossy conversion.

Oh yes, I saw that. It makes me wonder, though, why bother supporting 
the non-iconv case at all, since that may cause trouble to apps that 
expect conversion to work. But hey, it's already done.

> You already pointed out the difference that you need to the current
> implementation of GSFromUnicode(). You don't want your function to
> allocate extra memory, you need it to return the index up to which it
> was able to convert characters and you need a different return code that
> states whether any characters where converted. Although the last could
> possibly be determined by checking the returned index.
> Now this sounds like a strict extension of GSFromUnicode() and I think
> that in the long run these two functions should share their
> implementation code. But we also need to make sure that GSFromUnicode()
> is rock solid and fast, maybe we first keep them separate.
>
> Maybe it is best if you first write your new function by extracting all
> the code from GSFromUnicode() that doesn't handle memory management and
> add the different return values there. Next we could think about
> rewriting GSFromUnicode() on top of that. Adding the memory management
> and using some pointer arithmetic. And then we test whether this is fast
> enough.

I was expecting you to say something like that :-)

I don't like the current GSFromUnicode() implementation - probably 
because of the GROW() macro and the non-iconv case mess.

It could be more readable to write a minimal iconv implementation for 
builds without iconv, GSFromUnicode() could then only have a single 
branch of code.

--
Luboš Doležel