On 5/3/2011 3:54 AM, Thomas Heller wrote:
On Tuesday, May 03, 2011 10:33:50 AM Vladimir Batov wrote:
If we decide to go ahead building/extending a set of converters (based on this lib. or any other lib), then I'd probably favor the user choosing which converter he/she wants by specifying
#include<boost/convert/string-to-bool-fast.hpp> or #include<boost/convert/string-to-int-spirit.hpp>
Something along these lines. That way stream-based converters would co-exist with some quick ones. Just a thought.
Wouldn't it be very tedious to incorporate the locales and formaters when using these optimizations? Have you really thought this through? Wouldn't it render all the "nice" operator overloading tricks useless?
Again you claim your library to be extendable while it's not really. You have all those nice features so that you can easily use std locales and formatters (But i suspect these only work well with stdlib like streams). Once you incorporate any custom non stream based converter, all bets are off. Did I miss something obvious here?
I'd be happy with Vladimir's proposal of optional optimizations but there should also be a single convenience header that represents the "best" conversion. Preferably one that provides optimal performance for common string->numeric and numeric->string without invoking spirit's headers. See Antony Polukhin's patches for lexical_cast: http://boost.2283326.n4.nabble.com/lexical-cast-version-of-lexical-cast-in-1... Antony's patch can do locale-sensitive integral conversion (preprocessor disable-able) without invoking streams at all, so it's certainly possible to extend and maintain the extra features. Of course, it does add a lot of lines of code (essentially reimplementing and improving strtol), but it's worth it IMO. -Matt