bricks/stm32: Force exit when shutdown requested.

[laurensvalk]

If a user catches the SystemExit exception indefinitely, there is currently not a way to power off the hub.

This PR makes the following program exit normally so we can still de-init pbsys and do a clean power off.

from pybricks.tools import wait

while True:
    try:
        wait(10)
    except SystemExit:
        print("nope!")

I'm not merging this yet because there may be a cleaner way to exit MicroPython using PYEXEC_FORCED_EXIT and / or various available hooks.

[dlech]

Does this work if you replace the wait() and print() with pass so the the event poll hook never runs?

[laurensvalk]

Does this work if you replace the wait() and print() with pass so the the event poll hook never runs?

while True:
    try:
        pass
    except:
        pass

Yes and no - this already cleanly exits on a single press because most of the time is not spent in the try part.

[laurensvalk]

This, however, does still get stuck:

# This is the same as time.sleep_ms
from pybricks.tools import wait

while True:
    try:
        while True:
            try:
                wait(10)  
            except SystemExit:
                print("nope!")
    except SystemExit:
        pass

[laurensvalk]

This, however, does still get stuck:

We can detect the situation just fine, nothing truly gets stuck, it's a matter of telling MicroPython to stop.

Raising a specific exception could work, but the user may still use a bare except.

[laurensvalk]

Another possibility: With the recent code refactoring, reintroducing the longjmp can be done now by confining it to just the main application:

Along with this to call it:

This costs about 200 bytes instead of the 500 bytes from before (?).

This freezes the hub though (watch dog), so this isn't quite right.

[dlech]

I had a thought as I was falling asleep last night that maybe we could do something like the long jump, but put it in the MicroPython code so that we could make use of the MicroPython "nlr" code to hopefully keep the code size down.

[laurensvalk]

I had a thought as I was falling asleep last night that maybe we could do something like the long jump, but put it in the MicroPython code so that we could make use of the MicroPython "nlr" code to hopefully keep the code size down.

That could be nice too.

pbsys_main_stop_program always gets called correctly, so we could jump from there to near the end of pbsys_main_run_program.

For generality, it would be nice to change:

-void pbsys_main_stop_program(void)
+void pbsys_main_stop_program(bool force_stop)

That way the micropython application code doesn't have to poke at pbsys, it just has to exit when it is told to.

But I'm having some trouble making this work, probably I'm making the same mistake as in the pbsys variant above. It always freezes and reboots.

[laurensvalk]

I think I've figured out how to make this work. When a forced exit is triggered we can pop the nlr to the top level that was used to start the main program or REPL, and then raise the SystemExit to make it jump to the end. Does that make sense? See updated PR.

Separately, I've added a commit to also use this when pressing the stop button in Pybricks Code. Thoughts? Until now the stop button could theoretically be used as an "input", but we have the I/O terminal for that as well (on most hubs). It seems nice if the stop button always makes it stop.

[laurensvalk]

Feel free to merge this in if it looks right, I may be offline for a bit.

[laurensvalk]

Upstream MicroPython is considering addressing this by adding an exception type that can only be caught in C, not with a bare except.