pressing Ctrl-C breaks `time.sleep()` until reboot

[xobs]

Description

Pressing Ctrl-C to interrupt time.sleep() will cause time.sleep() to instantaneously return on NRF until the board is reset.

Steps to reproduce

In the REPL, run:

import time
time.sleep(100)

While it's sleeping, press Ctrl-C. The REPL will come back. Then run:

time.sleep(100)

The REPL will return instantly.

Workaround

Performing a soft- or hard-reboot re-initializes the sleep timer and gets it working again.

[tannewt]

Is this on both nRF52840 and nRF52833? Please let us know what boards you've found this on. Thanks!

[dhalbert]

I see this on Feather nRF52840.

[dhalbert]

This is not just nRF. I see the same thing on Metro M0.

[dhalbert]

This broke somewhere between 3.1.2 and 4.1.2.
EDIT: narrowed to between 4.0.2 and 4.1.0-beta.1. Now I'm bisecting (had to install python2 as python to build)

[dhalbert]

Commit causing the issue is ea760c0: "Turn on MICROPY_OPT_COMPUTED_GOTO for 5x CPU-bound speedup". So probably there's something wrong with how MP_STATE_VM(mp_pending_exception) is handled when MICROPY_OPT_COMPUTED_GOTO is turned on in the VM.

When time.sleep(100) is ctrl-c'd in a good build, you see

>>> time.sleep(100)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
KeyboardInterrupt: 

In a bad build, you just see

>>> time.sleep(100)   # ctrl-c typed
>>>

[dhalbert]

This problem does not occur in current micropython even when MICROPY_OPT_COMPUTED_GOTO = 1.

I compared our py/vm.c with micropython's, because MICROPY_OPT_COMPUTED_GOTO is only significant in that file. There are differences in exception handling based on the flag.

Really interestingly, this micropython commit, which is a minor speedup, micropython@869024d, changes

    #define DISPATCH() break

to

    #define DISPATCH() goto dispatch_loop

If I make that same change in our py/vm.c, then I can reproduce the problem even when MICROPY_OPT_COMPUTED_GOTO = 0. I haven't been able to track down other micropython changes yet that are causing this problem not to occur on micropython, though I will go back a year or two at a time in micropython in ports with computed goto turned on and see what happens. It's fishy in the previous comment that the interrupt doesn't get printed.

[dhalbert]

Aha: #1092

[kevinjwalters]

Is this a dupe of #2691 ?

#2689 features more control-c / interrupt woes.