Hi,

Please do not proceed at the expense of error handling. I believe using itoa and ftoa is fine, but using atoi and atof is definitely not:

cout << atoi("123X") prints 123 cout << atoi("X") prints 0

lexical_cast<int> throws in both cases, which I think is 100% the correct behaviour.

Would strtol() and friends be suitable from a performance point of view? They at least provide some error reporting. -Ossama

This implementation of itoa is rather fast and open-source, based on a lookup up table of groups of four digits: https://sourceforge.net/projects/itoa "Fast implementation of the old deprecated itoa(), lltoa() etc... plus some C++ wrapping. Theoretically controversed but practically very useful." The source is visible here, with a humble attempt to integrate it into Boost : http://itoa.cvs.sourceforge.net/viewvc/itoa/itoa/ Thorsten Ottosen wrote:

Darren Garvey skrev:

...

2008/12/3 David Abrahams <dave@boostpro.com>

...

Someone was asking me today about the inefficiency of lexical_cast for its common itoa/atoi-like usage, and it occurred to me that it could *easily* be optimized to handle the common cases by dispatching to itoa/atoi ftoa/atof where available, and even sprintf. Seems like a great idea to me.

Thoughts?

You may find the following interesting:

http://lists.boost.org/Archives/boost/2007/11/131001.php

I'd quite like to know if Johan got these changes into the library. Any news on this?

I haven't looked at the implementation, but I know Matthew Wilson had a series of articles in CUJ about fast string to integer conversions. IIRC, he was doing much better than the functions from the C library:

http://www.ddj.com/cpp/184403874

I guess it might be possible to optimize the floating point casts too, albeit it might be somewhat harder.

-Thorsten _______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost

This is about 10% faster on a Pentium: { for(const char* p = str_value; size; ++p, --size) { Jonathan Brannan wrote:

we have found that atoi/strtol/strtoll/strtod are slow compared to the optimized output (gcc -O2/3) of a very simple * 10 loop. The C number converters work for a large varity of bases, which makes them slower than needed for base 10.

we use something like:

template<typename TInteger> bool validated_to_unsigned_int( const char* str_value, size_t size, TInteger& result) { size_t i = 0; for(const char* p = str_value; i < size; ++p, ++i) { char digit = *p; if (digit >= '0' && digit <='9') { result = result * 10 + digit - '0'; } else { return false; } } return true; }

with wrappers for signed and ones that throw exceptions. We have a few converters that are faster, but only without error checking (base 16 conversion with lookup table works for small numbers, for example) _______________________________________________ Unsubscribe & other changes: http://lists.boost.org/mailman/listinfo.cgi/boost