hack: Bad process #280
hack: Bad process #280
Conversation
|
Also, when reproducing, disconnect any UART sensors. These also still broadcast events (will be fixed in #275). So it is possible we've had this issue for longer but the continuous source of UART broadcast event broadcasts was masking it? |
|
If this isn't working and we have ruled out overflowing the event queue, then it kind smells like possibly a memory corruption bug to me. One way to possibly debug that could be to keep turning off non-essential drivers (PBDRV_CONFIG) and subsystems (PBIO_CONFIG) until this starts working to narrow it down to what is interfering with proper operation. |
|
Thank you. This seems to be a debug print issue rather than a process issue. I went back to apply this patch to previous released and noticed it started occurring since b460d6d, which is plausible, because running the event loop from a process is a bad idea. All the more reason to finish up direct UART access for debugging 😄 |
This should not be merged, but serves to document and reproduce an odd bug that I am seeing locally.
Context
I was working on a major revision of the port subsystem, and was running into issues with etimers and polling processes.
After some time, I was able to boil it down to a very small example that also builds from the current master branch. This makes it seem independent from the current work (which is not submitted here).
To reproduce
Build either of the two commits from this PR. If everything is working, you should see
Hello Worldprinted in Pybricks Code every second after connecting.evcheck, it doesn't work at first but it starts when kicked externally. For example, start the REPL (state changes trigger a event broadcast).Caveats
This may be due to unrelated undefined behavior somewhere else. It seems to vary from build to build. At one point, adding comments made it (dis)appear, presumably because that changes the LC value of the switches. Similarly, experimenting with variations of
etimer_setsuch asetimer_resetdon't make the difference, even if they sometimes appear to make it work, but presumably that is the same as above.This is one reason to create the CI build --- to see if it reproduces there too. EDIT: It does. I suppose that is a good thing.