From: Rob Stewart (stewart_at_[hidden])
Date: 2004-10-20 15:32:33
From: "Erik Wien" <wien_at_[hidden]>
> Robert Ramey wrote:
> > a) the standard library has std:::basic_string<T> where T is any type
> > char, wchar_t or whatever.
> Yes. The problem with unicode is that it is not really possible to represent
> a character as an atomic value. A single glyph could in extreme cases be
> made up of 3 (or even more) 32 bit code units (UTF-32), and therefore
> defining a good T, is nigh on impossible.
Could the character type be a class that can hold one or more
data members of some representation type plus a pointer to
overflow data? Then, an abstract character can be represented
completely within the character type if the encoding is
sufficiently simple, and if the encoding is more complex, the
additional data is put on the free store.
For example, if most characters can be represented with a single
representation type instance, then the class would contain one
data member of that type plus a pointer to the rest, if any.
Performance analysis can indicate how best to implement such a
class, but it could have from one to N data members of the
representation type, where N is the maximum number of
representation type values needed to represent all abstract
characters. Differing choices of N and the representation type
will give different performance characteristics for a given
Unicode string. Those values might be tuned for general purpose
use or they might be exposed via template parameters.
Granted, a simple character is enlarged by an unused pointer and
it may be that using N objects of the representation type takes
no more space, thereby obviating the conditional code checking
for a non-null pointer. Nevertheless, it's an idea to consider,
if only for a minute. ;-)
-- Rob Stewart stewart_at_[hidden] Software Engineer http://www.sig.com Susquehanna International Group, LLP using std::disclaimer;
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk