Re: [Bug-apl] Almost One

[Juergen Sauermann]

Hi Mike,

I see. I have increased the number of digits to 17, but only for ⎕TF (because
otherwise a few of my automated testcases would break). SVN 674.

/// Jürgen

On 09/04/2015 05:42 PM, Mike Duvos wrote:

One small correction.  2^53 is 9e15, which is 16 digits, so writing stuff rounded to 17 decimal places should exactly reproduce it for IEEE 64 bit floats.  I don't know why 18 popped into my head. 

So just let the maximum ⎕PP be 17 instead of 16, and all will be well.

On Fri, Sep 4, 2015 at 8:26 AM, Mike Duvos <address@hidden> wrote:

I see the problem.  You limit ⎕PP to 16, which isn't enough digits.  1E¯16 is 1 to 16 places.  I didn't notice this, because you just set it to 16 if I set it to 18.

You need to allow a larger ⎕PP to exactly reproduce floating data written out and read back in.