gpio: Implement debounced WaitForEdge logic

[lutzky]

Some progress on #5, #9

This does, however, introduce sleeps into tests. This can be avoided by using https://github.com/jonboulle/clockwork, but I wanted to get early feedback before I dig into that.

[codecov-commenter]

Codecov Report

Merging #12 (6352fe3) into main (de2ac48) will increase coverage by 0.0%. The diff coverage is 100.0%.

@@          Coverage Diff          @@
##            main     #12   +/-   ##
=====================================
  Coverage   98.3%   98.3%           
=====================================
  Files         31      31           
  Lines       2699    2708    +9     
=====================================
+ Hits        2652    2661    +9     
  Misses        44      44           
  Partials       3       3           

Impacted Files	Coverage Δ
gpio/gpioutil/debounce.go	100.0% <100.0%> (ø)

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update de2ac48...6352fe3. Read the comment docs.

[lutzky]

CI failure appears unrelated to my change.

[maruel]

gohci

[maruel]

About CI failure: I discovered that my nice hack to run a go module at a specific commit works if the commit exists on any branch but not on PRs. I need to update the github actions on every of the repository which sucks and I haven't found time to fix this.

The code looks good.

[maruel]

(My brain is fried tonight, will recheck tomorrow)

[lutzky]

Thanks. Progressing a bit further, I've switched from time-based testing to using clockwork, making use use of it for one of the tests so far.

Pros:

• Much faster test runs (0.002s instead of 1.003s)
• Tests should not be flaky (well, with 1sec timeouts flakiness would likely be very limited, but still)

Cons:

• Adds an external dependency, which is disallowed (thereby also breaking without a go.sum file)
• Higher code complexity

WDYT?

[maruel]

Other updates:

• I've committed https://github.com/periph/conn/commit/376b141048edd82e8d3710f8ad0944057fbb3554 to fix failures specifically with PRs. You'll want to rebase
• raspberrypi-2f34 still fails but you can ignore that, it'll be fixed on the next host release.
• You'll want to integrate the last commit in branch https://github.com/periph/conn/tree/gosum to make the checks pass; https://github.com/periph/conn/commit/458744453656063835ba4b5252a41f255a6bd62b

[lutzky]

I've rebased on top of gosum, I hope this is indeed what you meant.

Moving Clock into gpiotest.Pin was much cleaner than I expected, I must've confused myself pretty badly there. Hopefully this will make converting the rest of the tests easier.

[maruel]

Oops I think I found out why canary test fails, will fix ASAP.

[maruel]

Actually, can you replace the go get ./... in the "Test on foo" in .github/workflows/test.yml with go get -t ./... and try again?

[lutzky]

That doesn't seem to be it... go.sum is... missing somehow?

[maruel]

\o/

It's because devices only depends on gpiotest in its unit tests, do go get ./... didn't get the clockwork dependency.

[lutzky]

Ah! I guess I was looking at a previous run of the checks. I'll adapt the one remaining test that uses time.Sleep.

[maruel]

Thanks!

[maruel]

Thanks for bearing with me! Should I release right away.

[lutzky]

Oh, I figured I'd do the other test that does time.Sleep as clockwork as well, but I guess I can do that separately.

[maruel]

Oh yes please do it as a separate change. I'll fix the other repositories to do go get -t ./...