Proposal for quantize_normal_tics() algorithm replacement

[Marek P. <ma...@du...>]

Dear gnuplot developers,


// first, many thanks for this marvelous sw package, I have been using it 
for school, work and fun ~10 years, and I am constantly astonished by the 
quality and the fact, how clever subset of functionality it has.


Now, I would like to consult with you an algorithm for automatical tic 
size choice, as present in quantize_normal_tics() last years. I have been 
in need of some user interface plotting for a cardiologic real-time 
analysis software, written in Athena/Xt GUI. I proposed an algorithm for 
auto tics, then, I looked into gnuplot for a comparison.

Since my own algorithm seems to me to be possibly slightly better, I would 
like to share the idea. The code snippet is:


float tic_quant[] = {5.0, 2.0, 1.0};
int n_min = 5, n_max = 10;

float quantize_tics(float y_min, float y_max)
{
   float f, d;
   int i, m, n;

   f = log10((y_max - y_min)/n_max);
   for (i = 0; i < length(tic_quant); i++) {
     m = ceil(f - log10(tic_quant[i]) - FLT_EPSILON);
     d = tic_quant[i]*pow(10, m);
     n = ceil(y_max/d) - floor(y_min/d);
     if (n >= n_min)
       break;
   }
   return d;
}


The tic size is always tic_quant[i]*10^k, k \in Z. Number of intertic 
spaces (intervals) n obeys strictly n_min <= n <= n_max. The algorithm can 
be tuned by variation of tic_quant[] and n_min, n_max. However, it should 
be assured, that for chosen n_min, n_max the algorithm will always find a 
solution.

It should be noted, that the function uses both bounds of desired axis 
range, not only the interval between. The purpose is to assure n_min, 
n_max, what can not be achieved, when only the interval is known. Thus, it 
can not serve as a drop-in replacement of quantize_normal_tics(), or it 
can, but the n_max may be overshot by one.

What do you think about this way? If you recommend it, I can prepare a 
patch to prove the idea.


Best regards,
Marek