gnuplot-bugs

[Gnuplot-bugs] Bug with Use of Floor() Function

[Victor B. <ban...@mc...>]

All, 

In trying to plot a discrete signal, I get the following result.
Plotting floor(10*x) / 10 should, theoretically, yield the same plot as
floor(10*x) * 0.1, but the former yields the plot for floor(x).  If,
instead of carrying out the division floor(y*x)/y, I do
floor(y*x)*(y**(-1)), the plot is as expected.

It seems that if division is used, the coefficient of x cancels out.  If
multiplication with the inverse of the coefficient is used, the plot is
fine.  Is this intentional?

Thanks, 

Victor Bandur

Re: [Gnuplot-bugs] Bug with Use of Floor() Function

[Hans-Bernhard B. <br...@ph...>]

Victor Bandur wrote:
> All, 
> 
> In trying to plot a discrete signal, I get the following result.
> Plotting floor(10*x) / 10 should, theoretically, yield the same plot as
> floor(10*x) * 0.1, 

By which theory, exactly?

You're missing one aspect of computer-based arithmetic here: the 
existence of different types of numbers.  gnuplot's floor() function 
yields an actual integer, not a floating-point number with an integer 
value.  So dividing it by another integer will behave differently from 
multiplying it with the floating-point inverse of that integer.

To get the same as floor(10*x)*0.1, try  floor(10*x)/10.0

 > but the former yields the plot for floor(x).

Only for positive arguments.

> Is this intentional?

Yes.