EPaperDisplay.time_to_refresh is too low to ensure safe refresh

[FoamyGuy]

CircuitPython version

Adafruit CircuitPython 9.0.0 on 2024-03-19; Adafruit MagTag with ESP32S2

Code/REPL

import time
import keypad
import board

display = board.DISPLAY
buttons = keypad.Keys((board.BUTTON_A, board.BUTTON_B, board.BUTTON_C, board.BUTTON_D), value_when_pressed=False)

while True:
    event = buttons.events.get()
    if event:
        if event.released:
            print(f"btn {event.key_number}")

            try:
                display.refresh()
            except RuntimeError as e:
                print("Caught Runtime error, probably refreshed too soon.")
                print(e)
                print(f"sleeping for: {display.time_to_refresh}")
                time.sleep(display.time_to_refresh)
                display.refresh()

Behavior

If you runthis code and then press two mag tag buttons within quick succession:

The second button press will cause a refresh too soon error which gets caught by the except clause, then we sleep for the value in display.time_to_refresh and then try to refresh again.

Often this ends up just raising another refresh too soon error:

code.py output:
btn 0
Caught Runtime error, probably refreshed too soon.
Refresh too soon
sleeping for: 1.367
Traceback (most recent call last):
  File "code.py", line 15, in <module>
RuntimeError: Refresh too soon

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "code.py", line 21, in <module>
RuntimeError: Refresh too soon

Code done running.

I found that if I add an extra 0.6 seconds to the sleep then I consistently no longer get the error. Granted this is a bit of a cludge for the workaround, I've only tested on two different devices, it's possible that others have different "magical numbers" required for the additional wait time.

Description

No response

Additional information

No response

[elpekenin]

AFAICT the C code works in milliseconds yet it gets converted to seconds when "passing" it to Py, which probably causes a rounding error which is the real problem making the code fail

I think a viable solution would be to do a breaking change so that the property returns in milliseconds and call sleep_ms instead of sleep

On a side note, maybe there should be some (opt-in) code that automatically flushes the screen (if time_elapsed_last_draw >= timeout && screen.dirty) to prevent users from manually messing with the time

[tannewt]

I think it'd be simpler to add a bit of time to the milliseconds number before it gets divided to float.