Avoid potential hang in connectionproxytest

[reuk]

At JUCE, we run the validator in our CI pipelines. We are frequently seeing that the validator hangs and causes the build to time-out. In a recent build on Linux, we saw that the build hung after printing [ConnectionProxy: Send message on 2nd thread], never printing [Succeeded], so this test seems to be flaky.

I suspect that the use of condition_variable::wait is the cause of the hang. If the second thread starts quickly and calls notify_all before the first thread calls wait, then the wait may never be unblocked.

Note that it's also possible for wait to wake spuriously. If this were to happen before the second thread called proxy.notify, it's possible that the test EXPECT_NE (*notifyResult, kResultTrue); would fail.

For these reasons, it's generally recommended to use the form of wait that takes a predicate argument, so that the predicate is checked before starting to wait, and on each spurious wake.

There are a couple of other issues with this test:

• Even though notifyResult is atomic, it should only be modified while holding the condition_variable's lock, according to cppreference. This isn't happening at the moment. The condvar is using m1, but notifyResult is modified under m2.

  Even if the shared variable is atomic, it must be modified while owning the mutex to correctly publish the modification to the waiting thread.

• There's no need to use notify_all here. Only one thread is waiting, so notify_one is sufficient.

[reuk]

Ping, just in case this got lost over the festive period.

[reuk]

There's been another point release since I opened this issue.

This is a real bug that causes problems for people running the validator regularly, especially on CI.

It would be great if this fix could be added in the next version of the SDK. Thanks!

[ygrabit]

Hi, It will be add to the next update, Thanks.