Temporal breaks context error handling

[Hades32]

ctx, cancel := context.WithTimeout(cctx, 2*time.Second)
defer cancel()
wf := tClient.GetWorkflow(ctx, wfID, "")
err := wf.Get(ctx, &res)
if errors.Is(err, context.Canceled) || errors.Is(err, context.DeadlineExceeded) {
	fmt.Println("timeout")
} else {
	fmt.Println("no timeout")
}

Expected Behavior

Should print "timeout"

Actual Behavior

Prints "no timeout"

[cretz]

Will have to investigate. With gRPC, our underlying protocol, often the context deadline is relayed server side and IIRC it's actually gRPC that chooses to "break context error handling" by returning a failed gRPC status with DeadlineExceeded.

If that is indeed the case, I am not sure us changing gRPC native behavior is best for users who already rely on gRPC errors (both in the high level client and in the low level direct gRPC service client that is also made available).

[Hades32]

That sounds plausible, as when I dumped the error I got some status fields.

I don't think it's a good idea to expose those gRPC details though. I shouldn't have to care about the underlying protocol if I don't use the service interfaces directly.

In my workaround code I'm now using https://github.com/uber-go/multierr to ensure both, the original error as well as a proper context.DeadlineExceeded are contained in the err