That's the races out of our test suite!

[andydotxyz]

Working perfectly locally, let's see what CI says!

Progresses #4654

Checklist:

• Tests included.
• Lint and formatter run with no errors.
• Tests all pass.

[andydotxyz]

Oh man, so many races not detected on my local setup :(

[dweymouth]

InMemoryPreferences will need to avoid using a goroutine here

[andydotxyz]

InMemoryPreferences will need to avoid using a goroutine here

Good catch thanks. Fixed

[andydotxyz]

OK, some platforms are now not detecting any races, so I have enabled it on the rest and working through what I hope is the last few!!!

[dweymouth]

Use new RunFromGoroutine API here - https://github.com/fyne-io/fyne/blob/develop/app/settings_desktop.go#L48

[coveralls]

coverage: 59.759% (+0.4%) from 59.329%
when pulling 7d5fb07 on andydotxyz:fix/racedetection
into 1bfc24c on fyne-io:develop.

[coveralls]

coverage: 59.754% (+0.4%) from 59.329%
when pulling 367a6a6 on andydotxyz:fix/racedetection
into 1bfc24c on fyne-io:develop.

[coveralls]

coverage: 59.764% (+0.4%) from 59.329%
when pulling 895953c on andydotxyz:fix/racedetection
into 1bfc24c on fyne-io:develop.

[coveralls]

coverage: 59.246% (-0.1%) from 59.374%
when pulling bcb6742 on andydotxyz:fix/racedetection
into f2ba049 on fyne-io:develop.

[dweymouth]

@Jacalz It seems we may need to remove or update the new pool_test. I think sync.Pool doesn't make a guarantee that an object released into the pool is immediately available for re-use, which is why it's failing occasionally here

[andydotxyz]

@Jacalz It seems we may need to remove or update the new pool_test. I think sync.Pool doesn't make a guarantee that an object released into the pool is immediately available for re-use, which is why it's failing occasionally here

Good point, the contract includes this on Get:

// Get may choose to ignore the pool and treat it as empty.
// Callers should not assume any relation between values passed to Put and
// the values returned by Get.

So I have removed the Set/Get test - but the Get on empty and the replacement New tests should both be OK.

[Jacalz]

This looking really good. Some greps of suspicious goroutines that I found:

• fyne/internal/driver/mobile/driver.go

  Line 313 in 1bfc24c

  go f() // don't block main, we don't have window event queue

• fyne/internal/driver/mobile/driver.go

  Line 419 in 1bfc24c

  go func() {

• Might be nothing but just FYI that it runs in a goroutine:

  fyne/app/app.go

  Line 71 in 1bfc24c

  go a.lifecycle.RunEventQueue()

[Jacalz]

This sleep looks suspicous:

fyne/app/preferences.go

Line 52 in 1bfc24c

time.Sleep(time.Millisecond * 100) // writes are not always atomic. 10ms worked, 100 is safer.

[Jacalz]

Sorry if those are just false positives. Figured it was better to report things that I thought looked strange. I also see some potential races in the flatpak dialog code that I added but I can tackle that later on.

[Jacalz]

Should we get #5382 merged so this can be tested more easily?

[dweymouth]

This looking really good. Some greps of suspicious goroutines that I found:

• fyne/internal/driver/mobile/driver.go

  Line 313 in 1bfc24c

  go f() // don't block main, we don't have window event queue

• fyne/internal/driver/mobile/driver.go

  Line 419 in 1bfc24c

  go func() {

• Might be nothing but just FYI that it runs in a goroutine:

  fyne/app/app.go

  Line 71 in 1bfc24c

  go a.lifecycle.RunEventQueue()

Good finds, those will all need some work because they all result in user callbacks, or widget functions, being called directly from goroutines which is against the new model

[Jacalz]

That's it merged. Rebasing this should make it possible to click around in fyne_demo and test out any potential races. I think I got one when changing application scale in the settings menu earlier but it might have been fixed by this PR already, I don't know :)

[andydotxyz]

This sleep looks suspicous:

fyne/app/preferences.go

Line 52 in 1bfc24c

time.Sleep(time.Millisecond * 100) // writes are not always atomic. 10ms worked, 100 is safer.

Thanks - fixed. I left the delay as there was a bug reported about file write delay but then it joins back to main now