Re: [lmi] strtod("inf") and MSVC

[Vadim Zeitlin]

On Thu, 31 May 2012 17:37:31 +0000 Greg Chicares <address@hidden> wrote:

GC> You anticipated the case-sensitivity issue:

 Yes (I'm replying to this as this part actually dates from my original
patch from 2011-06-27 email), the patch was really the smallest change
needed to make LMI work when compiled with MSVC, it was not supposed to be
exhaustive.

GC> and one thing led to another, and now I have a patch that converts all
GC> infinities for como, and I'd guess for msvc too; but does it look
GC> right, and is there a less awful way to write it that isn't slower?

 I suspect that writing comparisons char-by-char, i.e. (warning: untested
pseudo-code follows)

        (nptr[0] == 'i' || nptr[0] == 'I') &&
        (nptr[1] == 'n' || nptr[1] == 'N') &&
        (nptr[2] == 'f' || nptr[2] == 'F') &&
        (nptr[3] == '\0' ||
                (nptr[3] == 'i' || nptr[3] == 'I') &&
                ...
                (nptr[7] == 'y' || nptr[7] == 'Y'))

might be actually the fastest way to do it but I didn't benchmark it so I'm
not sure about this.

 But IMHO getting rid of duplicate "-foo" and "foo" comparisons would make
the code more clear so why not write it like this (same warning as above):

        bool negative = false;
        if(*nptr == '-')
            {
            negative = true;
            nptr++;
            }

        if
            (
               std::strcmp(nptr, "inf") == 0
            || std::strcmp(nptr, "INF") == 0)
            {
            nptr += 3;
            }
        else if
            (
               std::strcmp(nptr, "infinity") == 0
            || std::strcmp(nptr, "INFINITY") == 0)
            {
            nptr += 8;
            }
        else
            {
            throw std::invalid_argument("Numeric conversion failed.");
            }

        if(endptr)
            {
            *endptr = const_cast<char *>(nptr);
            }

        return negative
            ? -std::numeric_limits<T>::infinity()
            : std::numeric_limits<T>::infinity()
            ;

?

 Granted, it's not a huge improvement but it still looks a bit more clear
to me.

 Regards,
VZ