Loss of accuracy over time - use monotonic_ns?

[tgs]

Good morning!

I'm working on a project that will be continuously powered for months at a time. I'd like to use this (very fine) debouncer implementation, but I think it will lose accuracy over time. As a test, I ran this code on my ItsyBitsy M0 Express for a day or so:

import time


while True:
    print("At", time.monotonic(), "- ", end="")
    before = time.monotonic()
    time.sleep(0.01)
    print(time.monotonic() - before)

    time.sleep(1)

So it's sleeping for a hundredth of a second, and seeing what the time.monotonic() difference is over this time.

At 87214.7 - 0.0
At 87215.7 - 0.03125
At 87216.7 - 0.0

At 24 hours or so, most of the hundredths of a second are too fine to be resolved by the floating point number returned by time.monotonic(), and when it does catch one, it thinks it's a 32nd of a second. So I think you'll have to press a button for longer and longer before the debouncer will acknowledge it.

To fix the issue, I'm thinking about something like this:

diff --git a/adafruit_debouncer.py b/adafruit_debouncer.py
index b4702ff..5567e82 100644
--- a/adafruit_debouncer.py
+++ b/adafruit_debouncer.py
@@ -53,6 +53,14 @@ _DEBOUNCED_STATE = const(0x01)
 _UNSTABLE_STATE = const(0x02)
 _CHANGED_STATE = const(0x04)
 
+if hasattr(time, 'monotonic_ns'):
+    INTERVAL_FACTOR = 1_000_000_000
+    MONOTONIC_TIME = time.monotonic_ns
+else:
+    INTERVAL_FACTOR = 1
+    MONOTONIC_TIME = time.monotonic
+
+
 class Debouncer(object):
     """Debounce an input pin or an arbitrary predicate"""
 
@@ -69,7 +77,7 @@ class Debouncer(object):
         if self.function():
             self._set_state(_DEBOUNCED_STATE | _UNSTABLE_STATE)
         self.previous_time = 0
-        self.interval = interval
+        self.interval = interval * INTERVAL_FACTOR
 
 
     def _set_state(self, bits):
@@ -90,7 +98,7 @@ class Debouncer(object):
 
     def update(self):
         """Update the debouncer state. MUST be called frequently"""
-        now = time.monotonic()
+        now = MONOTONIC_TIME()
         self._unset_state(_CHANGED_STATE)
         current_state = self.function()
         if current_state != self._get_state(_UNSTABLE_STATE):

There are a few TODOs on that patch:

• Could preserve backwards compatibility for reading debouncer.interval by using a separate attribute to store interval * INTERVAL_FACTOR
• Could preserve the ability for client code to write debouncer.interval by making it into a property with a getter and setter, and doing some bookkeeping
• I've never used the micropython const system, but I'm sure it would help with speed. I just don't know if it works inside if statements or anything. I would just test this and see how it works.
• I don't know if there's anything helpful to do for systems too small to have monotonic_ns, like the Trinket M0. If we could access the ticks_ms CircuitPython internal variable, we could handle it, but we can't see that from Python land :(

So does that sound like something you're interested in? I'm happy to work on the patch and bring a PR if so.

[dastels]

Please do. I'm interested to see how this turns out. I do wonder if GC is interfering with the timing. Try your experiment with a gc.collect() after the first print statement.

[tgs]

Here's another version of the experiment:

>>> times = set()
>>> for _ in range(100):
...     times.add(time.monotonic())
...     
>>> times
{254643.0, 254643.0}

And here's the output of the requested gc.collect() version, also changed to run more often:

At 254843.0 - 0.0
At 254843.0 - 0.0
At 254843.0 - 0.0625
At 254843.0 - 0.0625
At 254843.0 - 0.0
At 254843.0 - 0.0
At 254843.0 - 0.0
At 254843.0 - 0.0
At 254843.0 - 0.0
At 254843.0 - 0.0
At 254843.0 - 0.0
At 254844.0 - 0.0
At 254844.0 - 0.0625
At 254844.0 - 0.0625
At 254844.0 - 0.0625
At 254844.0 - 0.0625
At 254844.0 - 0.0625
At 254844.0 - 0.0625
At 254844.0 - 0.0625
At 254844.0 - 0.0625
At 254844.0 - 0.0625
At 254844.0 - 0.0
At 254844.0 - 0.0
At 254844.0 - 0.0
At 254844.0 - 0.0
At 254844.0 - 0.0
At 254844.0 - 0.0625
At 254845.0 - 0.0625

I'm done with a first draft of the patch! I'll submit it in a sec

[kattni]

Resolved!