On 24 January 2013 22:31, Gabriel Dos Reis <gdr@integrable-solutions.net> wrote:
On Thu, Jan 24, 2013 at 8:02 PM, Ed Meyer <eem2314@gmail.com> wrote:
On the other hand I think it is a bad idea to write code that depends on
a particular implementation or rounding behaviour
How *concretely* to do you propose to avoid that? Are you suggesting
something to the programmers or to the implementers?
He's saying, I think, that we shouldn't expect calls like pow(3.0, 2)
to have no round-off error for any implementation of pow, so we should
code around that possible round-off error. I myself think that since
GNU's C++11 implementation is the only one we've seen with round-off
error here, that this can reasonably be considered a bug in libstdc++.