add_scatter_series crashes when references to numpy ndarray are passed

[Dragoner0]

Version of Dear PyGui

Version: 2.1.0
Operating System: Windows 11

My Issue/Question

While using add_scatter_series with references to a numpy ndarray (see code below) an exception is raised. This can be avoided by copying the relevant data before or creating a list.

Exception has occurred: SystemError returned a result with an exception set
File "...", line 32, in
dpg.add_scatter_series(sindatax, sindatay, label="0.5 + 0.5 * sin(x)", parent="y_axis", tag="series_tag")
SystemError: returned a result with an exception set

Is it the expected behavior that an exception is raised here?

To Reproduce

Run the script below.

Expected behavior

add_scatter_plot should work when references to ndarrays are used.

Standalone, minimal, complete and verifiable example

import dearpygui.dearpygui as dpg
from math import sin
import numpy as np

dpg.create_context()


sindata = np.zeros((500,2))
for i in range(0, 500):
    sindata[i,0] = i / 1000
    sindata[i,1] = 0.5 + 0.5 * sin(50 * i / 1000)


with dpg.window(label="Tutorial", tag="win"):
    with dpg.plot(label="Line Series", height=400, width=400):

        dpg.add_plot_axis(dpg.mvXAxis, label="x")
        dpg.add_plot_axis(dpg.mvYAxis, label="y", tag="y_axis")

        sindatax = sindata[:,0]
        sindatay = sindata[:,1]

        #  === WORKAROUND / FIX No. 1 === 
        #sindatax = np.copy(sindata[:,0])
        #sindatay = np.copy(sindata[:,1])

        #  === WORKAROUND / FIX No. 2 === 
        #sindatax = sindata[:,0].tolist()
        #sindatay = sindata[:,1].tolist()

        # Crashes here
        dpg.add_scatter_series(sindatax, sindatay, label="0.5 + 0.5 * sin(x)", parent="y_axis", tag="series_tag")

dpg.create_viewport(title='Custom Title', width=800, height=600)
dpg.setup_dearpygui()
dpg.show_viewport()
dpg.start_dearpygui()
dpg.destroy_context()

[v-ein]

Good catch! Yes, the exception is expected and should actually give you a message about ndarray being non-contiguous. For me, however, your script segfaults 😂, and this is a problem we need to fix (luckily, the fix is trivial).

Here's what happens.

• np.zeros((500,2)) gives you a 2D array whose memory layout is like this: x[0, 0], x[0, 1], x[1, 0], x[1, 1], x[2, 0], ... That is, it's "a list of pairs" rather than "a pair of lists".
• sindata[:,0] creates a on-the-fly mapping into that array without actually copying memory anywhere. Since this mapping accesses every other item, the resulting array is non-contiguous in memory.
• When you pass this non-contiguous array into add_scatter_series, DPG tries to copy data from the array. It supports tuples, lists, and anything that supports the buffer protocol. NumPy arrays fall into the latter category. However, since the array is non-contiguous, NumPy rejects the attempt to acquire buffer access - the buffer must be contiguous.

The solution for you would be to swap indexes and use (2, 500) instead of (500, 2).

I'm going to open another issue to address the segfaults that I was getting. Unfortunately this appears to be quite a widespread problem in DPG code, affecting not only scatter series but a number of other APIs.