Synchronize read & write of TaskGroupWithContext's err variable

[thekondor]

Context

That is not stated whether Wait() & Submit() for TaskGroupWithContext are allowed for usage from different execution threads. As end user I expected that it is.

Problem

// thread 1
group.Submit(func() error {
   return errors.New("failed")
})

// thread 2
group.Wait()

leads to race condition because of non-synchronized (write & read) operations to .err.

The problem is detectable in 100% cases by a built-in race checker.

PR

The change introduces usage of mutex to synchronize access to this variable at the above mentioned code flows.

[alitto]

Thanks a lot for this contribution @thekondor! 🙌. I'll merge this and release a patch version to include it.
Very nice PR description 🙂

[NOMORECOFFEE]

You have to return g.ctx.Err(), otherwise we will get a data race.

ctx, cancel := context.WithTimeout(ctx, time.Second)
defer cancel()

g, gctx := pool.GroupContext(ctx)
g.Submit(func() error { select { case <-gctx.Done(): return gctx.Err() }})
err := g.Wait() // err == nil

[alitto]

Oh, good catch @NOMORECOFFEE!. If you don't mind, please open a PR with suggested changes and I can review it later. Otherwise I can send one. Thank you! 🙌

[alitto]

I just released v1.8.3 with a fix for this, please check it out and let me know if you find anything. Thank you!