This is a duplicate of bug #32980.
This is a long standing bug.
If you have interest in providing a fix, please post a patch there.

Oh that's very disappointing, but thanks for clarifying Dimitri.

Is there a plan to fix the problem ?


I do wonder though if a trick is being missed here.  I agree that a single precision library will manifest the problem, but it's not really the CAUSE of the problem in my view.

When it comes down to it a good high res display hardware has perhaps 7680x4320 pixels, and that defines the plot resolution, not single or double precision.  The job of the plotting subsystem is to map the graphical data to that limited display range by normalising it (subtract the mean offset and scaling).  Given the limited hardware resolution, single precision arithmetic has sufficient resolution to correctly do that.

However the PLOT function takes in double precision data and passes it to OpenGL, which apparently uses single precision.  If the data is not appropriately normalised (subtract the mean offset and scale) at the double to single conversion then detail will be lost at that stage, and that I suspect is where the problem lays. 

I'd suggest that the PLOT function is not appropriately normalising the data before it is converted to single precision.  The normalisation should only need to ensure the detail fits into a single precision number, the normalisation to the display hardware resolution should be handled in the graphics library itself.  The actual value can still be shown by axis labels.

The issue is really that the PLOT function accepts double precision data, but does not correctly map double precision data to the display, which should be possible.

I've encountered a bug in the Octave PLOT function, when plotting some scientific data.  I've created below a stripped down demo code to demonstrate the problem with synthetic data.

The demo code produces two figures with comparable data, one showing the problem, one not.  The data produces a straight line graph in fig1, but in fig2 produces an error and displays a staircase waveform instead.  Also the xgrid is non-uniform & poorly labelled, and the y-axis labelling is poorly positioned.

The problem appears to be triggered when the data on one axis has a moderately large mean value, but small dynamic range.  And it occurs whether it is x data or y data.

I've checked this on the following platforms, all with the same result:

Linux Oct 6.3 (Flatpack)
Linux Oct 5.2 (Native)
Win10 Oct 6.1

I've plotted the same data with Python (code attached) and it plots correctly with no error.  Unfortunately I no longer have access to Matlab, but I would expect it to plot correctly also.

I suspect the problem may be that the plot function is not properly considering the graph limits, to plot the data with the correct resolution.

Please could a fix be made.

Regards, Trevor



OCTAVE DEMO CODE

function retval = plterr()

  close all;

  x=[0:0.001:1];
  y=[0:0.001:1];
  k=700000;

  h1=figure;      % This plots OK
  plot(x, y);
  grid on;

  x=x+k;          % Add offset to data

  h2=figure;      % This plots INCORRECTLY
  plot(x, y);
  grid on;

end

PYTHON DEMO CODE

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Sat Aug 28 13:22:20 2021

@author: trevor
"""

import numpy as np
import matplotlib.pyplot as plt

x = np.linspace(0,1,num=1000)+700000
y = np.linspace(0,1,num=1000)

plt.figure(num=0,dpi=120)
plt.plot(x, y)