Hacker News Reader

Heh, I never realized doubles can fit a googol. I knew they work up to 1e300-ish, but never made this explicit connection.

View on HN

Well, only approximately: as a double it is actually 10000000000000000159028911097599180468360808563945281389781327557747838772170381060813469985856815104

It still 'fits'. The box is just very weird and sometimes the object that comes out is slightly different in size, but at least it's predictable.

I'm usually extremely reluctant to use doubles past 2^52, when the minimum increment becomes larger than 1.0

In that case you should likely have used fixed point arithmetic to start with.

Doubles larger than 2^52 are no different from smaller numbers in context of scientific computing etc. which is what floating point is for.

E.g. in a particular numerical simulation you could bump above 2^52 by using units of nanometers or below by using units of kilometers... yet this is likely just a simple multiplicative factor and the result of the calculation does not change one bit because of what units of measurement you use.

Getting closer to 10^300 and it is another story.

“I would have liked to be able to represent (1 << 53) + 1 exactly, but all I have is JavaScript and importing a bignum library is a lot harder than just restricting the range” seems like an unfortunately common problem.

Reply on news.ycombinator.com