Merge pull request #268 from well-typed/edsko/issue-221

Client-side timeouts

dev/interop.md

@@ -240,12 +240,7 @@ $ docker stop $(docker ps -q)
 >
 > The only exception is `timeout_on_sleeping_server` for Java: it seems that the
 > server does not conform to the gRPC specification here, and simply closes the
-> connection without sending `DEADLINE_EXCEEDED` to the client. The Java
-> _client_ does not notice this because (it seems) it imposes a _local_ timeout
-> also, and doesn't even _connect_ to the server: the test even passes without
-> the server running at all. (And indeed, the other clients don't seem precise
-> enough here: the `grapesy` server was passing this test before it implemented
-> timeouts _at all_.)
+> connection without sending `DEADLINE_EXCEEDED` to the client.
 
 It is also possible to only run one specific test case, for example:
 

grapesy/interop/Main.hs

@@ -15,13 +15,9 @@ import Interop.Server (runInteropServer)
 
 main :: IO ()
 main = do
-    -- TODO: <https://github.com/well-typed/grapesy/issues/123>
-    --
-    -- Under some circumstances we can see a "Thread killed" exception when a
-    -- test fails. By setting this handler, we can at least confirm that this
-    -- exception is coming from an uncaught exception in a thread.
-    setUncaughtExceptionHandler $ \err ->
+    setUncaughtExceptionHandler $ \err -> do
       hPutStrLn stderr $ "Uncaught exception: " ++ show err
+      hFlush stderr
 
     -- Ensure we see server output when running inside docker
     hSetBuffering stdout NoBuffering