displayio.Shape does not change after calling set_boundary() again

[FoamyGuy]

I have this script:

import displayio
import board
import time

WIDTH = 100
HEIGHT = 100
display = board.DISPLAY

shape = displayio.Shape(WIDTH, HEIGHT, mirror_x=False, mirror_y=False)
for y in range(HEIGHT):
    shape.set_boundary(y, 0, 30)

palette = displayio.Palette(2)
palette[0] = 0x0000FF
palette[1] = 0x00FFFF

tile_grid = displayio.TileGrid(shape, pixel_shader=palette)
group = displayio.Group()
group.append(tile_grid)
display.show(group)

#time.sleep(1)
while True:
    for y in range(HEIGHT):
        shape.set_boundary(y, 0, 80)
    time.sleep(1)
    for y in range(HEIGHT):
        shape.set_boundary(y, 0, 30)
    time.sleep(1)

If I run this code as-is I will see the 80px wide shape drawn on the screen but it will never change to 30px wide even though are changing the shape with new calls to set_boundary() inside the main loop.

If I un-comment the time.sleep(1) on the line before the main loop then I will see the 30px wide shape drawn on the screen, but never the 80px one.

The shape seems to get "stuck" at a certain size and does not change visually with further changes. Calling time.sleep() seems to have some impact on which size the shape is shown at, but it's "stuck" either way, never changing visibly.

I tested this on PyPortal and PyGamer both versions 5.3.1 and 6.0.0.alpha2.

[anecdata]

Forcing the show by manually doing

    display.show(None)
    display.show(group)

after each for loop gets the shapes toggling.

Still dependent on the sleep value... no sleep toggles blockily. 1 or 0.1 seconds are fine. 0.01 seconds freezes again. Interesting behaviors.

[tannewt]

We'll need to add dirty tracking to Shape to make this work. Similar to this: https://github.com/adafruit/circuitpython/blob/main/shared-module/displayio/Bitmap.c#L153-L164

[kmatch98]

I’m starting to look into this issue. @tannewt, I will appreciate if you can confirm I am heading in the right direction:

shared_module\Group.c has a function called displayio_group_get_refresh_areas that seeks out the refresh areas of Vectorio, TileGrid or Group display elements. Also there is a corresponding function displayio_group_finish_refresh that Is called after a refresh is completed, that also has entries for the same three display structures.

I suspect that in these two Group.c functions, a new entry should be added for the Shape element and the associated shape_get_refresh_areas, shape_finish_refresh and internal data structure for storing the Shape dirty rectangle tracking.

One thing I’m not sure of is why the time.sleep() has an impact on the behavior. Is there something else related to sleep that is affecting the behavior that should be investigated and resolved?

[tannewt]

I have no idea why sleep impacts it.

Group shouldn't need to change because Shape acts similar to Bitmap. TileGrid manages it's refresh area. You can see one TODO about it here: https://github.com/adafruit/circuitpython/blob/main/shared-module/displayio/TileGrid.c#L502

You'll also want to do something similar to this: https://github.com/adafruit/circuitpython/blob/main/shared-module/displayio/TileGrid.c#L533

[kmatch98]

@FoamyGuy and @tannewt I replicated your observation about the screen drawing related to time.sleep on an NRF52840 Itsy Bitsy.

Here is one hypothesis on why sometimes the drawing updates. In your example code, perhaps the Shape boundary gets updated to the 80 size before the screen refresh timing hits. So, even though there is no dirty rectangle tracking, if the display refresh cycle hits after the screen was drawn at 30 and then at 80, then it will only show the 80 size.

I fiddled with the delay and actually observed something in-between drawn:

display.refresh(target_frames_per_second=60, minimum_frames_per_second=60)
display.show(group)
time.sleep(1/100)

On an NRF52840 ItsyBitsy

[FoamyGuy]

fixed by #3374