Fix race condition during matplotlib rendering

[Tom-Willemsen]

Description of work

This fixes a race/conflict which I believe occurs between calls to draw() and print_figure().

I believe the two things racing against each other are:

• Updates triggered from the python side (possible because our backend is non-blocking)
• Updates triggered by websocket messages sent from the GUI

To reproduce blank files being saved

• Put the following script (based on MPL example which draws a 2d heatmap in similar way to refl alignment bluesky plans) in c:\scripts\mpl_stress.py

Details

import numpy as np


def mandelbrot_set(xmin, xmax, ymin, ymax, xn, yn, maxiter, horizon=2.0):
    X = np.linspace(xmin, xmax, xn).astype(np.float32)
    Y = np.linspace(ymin, ymax, yn).astype(np.float32)
    C = X + Y[:, None] * 1j
    N = np.zeros_like(C, dtype=int)
    Z = np.zeros_like(C)
    for n in range(maxiter):
        I = abs(Z) < horizon
        N[I] = n
        Z[I] = Z[I]**2 + C[I]
    N[N == maxiter-1] = 0
    return Z, N


def stress():
    import time

    import matplotlib.pyplot as plt

    import matplotlib
    from matplotlib import colors

    xmin, xmax, xn = -2.25, +0.75, 3000 // 2
    ymin, ymax, yn = -1.25, +1.25, 2500 // 2
    maxiter = 200
    horizon = 2.0 ** 40
    log_horizon = np.log2(np.log(horizon))
    Z, N = mandelbrot_set(xmin, xmax, ymin, ymax, xn, yn, maxiter, horizon)

    # Normalized recount as explained in:
    # https://linas.org/art-gallery/escape/smooth.html
    # https://web.archive.org/web/20160331171238/https://www.ibm.com/developerworks/community/blogs/jfp/entry/My_Christmas_Gift?lang=en

    # This line will generate warnings for null values, but it is faster to
    # process them afterwards using the nan_to_num
    with np.errstate(invalid='ignore'):
        M = np.nan_to_num(N + 1 - np.log2(np.log(abs(Z))) + log_horizon)

    dpi = 72
    width = 10
    height = 10*yn/xn
    fig = plt.figure(figsize=(width, height), dpi=dpi)
    ax = fig.add_axes((0, 0, 1, 1), frameon=False, aspect=1)

    # Shaded rendering
    light = colors.LightSource(azdeg=315, altdeg=10)
    M = light.shade(M, cmap=plt.colormaps["hot"], vert_exag=1.5,
                    norm=colors.PowerNorm(0.3), blend_mode='hsv')
    ax.imshow(M, extent=[xmin, xmax, ymin, ymax], interpolation="bicubic")
    ax.set_xticks([])
    ax.set_yticks([])

    # Some advertisement for matplotlib
    year = time.strftime("%Y")
    text = ("The Mandelbrot fractal set\n"
            "Rendered with matplotlib %s, %s - https://matplotlib.org"
            % (matplotlib.__version__, year))
    ax.text(xmin+.025, ymin+.025, text, color="white", fontsize=12, alpha=0.5)

    plt.show()
    from time import time
    plt.savefig("c:/instrument/mpl/" + str(time()) + ".png")

• Load the script with g.load_script("mpl_stress.py", check_script=False)
• Create directory c:\instrument\mpl
• In PyDEV scripting console in GUI run for _ in range(500): stress()
• Check for difference in behaviour between this branch and main:
  • On main some of the generated .png files in C:\Instrument\mpl will be blank / misrendered. On my machine, this occured for around 5% of files on average.
  • On this branch, all generated .png files should be identical and should all show exactly the same image. None should be blank/misrendered.
• Tip: look at file sizes or thumbnails in windows explorer to easily spot 'blank' images

To reproduce plot not appearing until resized

• Run in the python scripting console inside the GUI:

import matplotlib.pyplot as plt
for i in range(1000):
    plt.plot(range(i, i*2))
    plt.show()
    plt.savefig("c:/instrument/mpl/" + str(i))

• On main, you may get a blank plot appearing in the GUI (until the plot is resized)
  • You may also see partially-rendered plots in saved files, for example.
• On this branch, you should always see some plot in the GUI, and save files should be rendered consistently correctly.

To test

Closes #90

Acceptance criteria

List the acceptance criteria for the PR

---

Code Review

• Is the code of an acceptable quality?
• Are there unit tests in place? Are the unit tests small and test the a class in isolation?
• Have the changes been documented in the release notes. If so, do they describe the changes appropriately?

Functional Tests

• Do changes function as described? Add comments below that describe the tests performed.
• How do the changes handle unexpected situations, e.g. bad input?
• Has developer documentation been updated if required?