Memory leak when using Input?

[Nick-Aus]

A simple loop with Input.begin() and (optionally) Input.end() will crash the Edge Control board after a repeatable number of cycles. 1134 in my test. The heap fills up until it borks. Input.begin() does a Wire.begin() behind the scenes. Adding a Wire.end() to the test loop stabilizes the heap size and it runs ok "forever". Wire calls TWI but I can't immediately see in either library what's causing this behavior.

[manchoz]

Hi @Nick-Aus,
Input.begin() is meant to be run only one time in setup(): do you have a different scenario?

[Nick-Aus]

Hi @manchoz,
Not really - I guess I assumed that since the 5V rail was being shut down after using Input (for a 10-minute sleep before the next reading), that it would need to be initialized again before reuse. If this is not the case, then I am fine with just running Input.begin() once in setup. I will try it out!

[manchoz]

In that case, I would recommend to end() everything before going to sleep and then begin() everything again at wake-up.

Low-power scenarios needs to be managed carefully.

Let me know how is going

[Nick-Aus]

That is how my code was written and it hangs after several days operating. I shortened up the delays, added memory reporting and found the issue as described in my first post. I will refactor to add a Wire.end() before going to sleep. Unfortunately inadvertently calling end() twice also causes a hang, so I have to understand what the various libraries are doing too. Thanks for your comments!

[Nick-Aus]

Just some more info on this after a bit more playing -
After Wire.begin() is called, calling it again causes additional heap to be used up.
Unfortunately Wire.begin() is called from Expander.begin(), and Input.begin() calls Expander.begin(). So doing the following (as per the examples) -
Wire.begin(); Expander.begin(); ... Input.begin();
... will actually cause heap to be used that is not reclaimed by any use of end(). So many begin/read/end/sleep cycles will eventually cause a crash.

[Nick-Aus]

Just to close this out with a bit more info -

I got it to work without leaking memory with the following approach:

Input.begin(); // which calls Expander.begin() which calls Wire.begin() so don't call these separately
Input.enable();
// -- read input(s) as required
Input.disable();
Input.end();
Wire.end(); // frees the memory allocation from the Wire.begin()

This can be run in a loop (in my case, every 10 minutes with sleep in between) without eventually crashing due to memory overflow. It feels it would be nicer if Wire.begin() deleted any previous memory allocation before blindly creating new, but perhaps my scenario is an unusual case.

[manchoz]

Hi @Nick-Aus,
I will definitely check if it is possible to manage a better Wire initialization and de-initialization inside the classes that require it.

Unfortunately, I think I will need to manage the current status of Wire inside an Arduino_EdgeControl class or global or similar because it is currently not possible to detect the status of Wire by querying itself, due to the current Wire API.

Stay tuned and thanks so much for your effort and debugging.