search

dotnet / iot

This repo includes .NET Core implementations for various IoT boards, chips, displays and PCBs.
MIT License
2.12k stars 574 forks source link

Revert "Add libgpiodv2 support (#2184)" #2287

Closed pgrawehr closed 3 months ago

pgrawehr commented 4 months ago

This reverts commit 3fb06eb5fefbea23306cd85d67bfcff7b987065b. Could be the reason for recent test failures

update: 1/1 green so far https://github.com/dotnet/iot/pull/2287#issuecomment-1971740169

Microsoft Reviewers: Open in CodeFlow
pgrawehr commented 4 months ago

Marked DNM, as this is just as a test whether the change causes the increased test failure rate.

krwq commented 4 months ago

green before rerun

krwq commented 4 months ago

/azp run

azure-pipelines[bot] commented 4 months ago
Azure Pipelines successfully started running 1 pipeline(s).
krwq commented 3 months ago

@huesla we think something in the libgpiodv2 PR has regressed our CI and it's red very frequently now (and since revert makes it green we think it's the culprit), for example see: https://helixre107v0xdeko0k025g8.blob.core.windows.net/dotnet-iot-refs-pull-2148-merge-6ddc551227bf4d07ae/System.Device.Gpio.Tests.dll/1/console.758758bc.log?helixlogtype=result - the errors seem a bit random, any chance you could help us pinpoint the issue? (i.e. run tests in a loop and try to repro perhaps)

huesla commented 3 months ago

@krwq I try find the issue, but I need a similar setup. What hardware/OS does the CI run tests on?

huesla commented 3 months ago

And is it always the SysFsDriver tests? How did tests behave before adding libgpiod2, 100% green?

pgrawehr commented 3 months ago

@huesla The tests where not 100% green, but maybe 90%, now they're at 50% or less. It seems to mostly be the sysfs driver, but that could be a coincidence. I suspect that there's a situation where now a pin or GpioController is not properly disposed, which is known to cause problems when switching between sysfs and libgpiod. (see eg #1992) The error can also happen if a previous test run exited unexpectedly, as the devices are not rebooted between tests.

The hardware in use are Raspberry Pi 3B's with a custom test print attached. The schematic can be found here.. For the GPIO tests, you should only need the connection between pins 5 and 6. I haven't done that in a while, but running the test suite should be as easy as running

dotnet test src\System.Device.Gpio\System.Device.Gpio.sln

on the Pi.

huesla commented 3 months ago

@pgrawehr @krwq I think I found the issue, I have created #2291 that should fix it. I tested it also on a RPI3.

pgrawehr commented 3 months ago

Closing this, as issue seems to be resolved with #2291