Re: Using a real parser generator to parse numbers (and dates)

On 02/29/2012 02:57 PM, Stephan Bergmann wrote:
> On 02/29/2012 11:49 AM, Lionel Elie Mamane wrote:
> > 2) I add a const_iterator interface to rtl::OUString. Instead of
> > coding it manually to use iterateCodePoints, I again reuse the
> > power of boost. But I have in my git history a manually-coded
> > version based on iterateCodePoints if for some reason it is
> > desirable (I don't think it is).
> >
> > That's also more generally a more "C++ natural" interface to
> > OUStrings, so we can use it throughout our codebase instead of
> > iterateCodePoints.
> >
> > I don't define an iterator interface, as this could be *extremely*
> > slow: writing one character in the middle of a string is O(n)
> > (linear in the length of the string) when replacing a
> > surrogates-encoded codepoint by a non-surrogate encoded one and
> > vice-versa.
> [...]
> > 5) I've stuck the LibreOffice<->Spirit integration in
> > svl/source/numbers/spirit/, but possibly we could move it to a more
> > general location (sal/rtl?) so that it can be reused in other parts
> > of LibreOffice.
>
> Note that the stable sal interface historically stays clear of boost,
> because of differences in the various boost versions available in the
> various environments. Also, the SDK does not include any boost headers,
> so client code that does not even itself make use of boost could stop to
> compile.

(Also, is boost::u16_to_u32_iterator from 
boost/regex/pending/unicode_iterator.hpp supposed to be a stable interface?)

Stephan