Fix crash when writing to a Barrier more than once (#637)

* Fix crash when writing to a Barrier more than once

* Less confusing now

Author: pepeiborra

src/Development/IDE/Core/Shake.hs

@@ -497,8 +497,15 @@ newSession IdeState{shakeExtras=ShakeExtras{..}, ..} systemActs userActs = do
     let runInShakeSession :: forall a . Action a -> IO (IO a)
         runInShakeSession act = do
               res <- newBarrier
-              let act' = actionCatch @SomeException (Right <$> act) (pure . Left)
-              atomically $ writeTQueue actionQueue (act' >>= liftIO . signalBarrier res)
+              let act' = do
+                    -- work gets reenqueued when the Shake session is restarted
+                    -- it can happen that a work item finished just as it was reenqueud
+                    -- in that case, skipping the work is fine
+                    alreadyDone <- liftIO $ isJust <$> waitBarrierMaybe res
+                    unless alreadyDone $ do
+                        x <- actionCatch @SomeException (Right <$> act) (pure . Left)
+                        liftIO $ signalBarrier res x
+              atomically $ writeTQueue actionQueue act'
               return (waitBarrier res >>= either throwIO return)
 
     --  Cancelling is required to flush the Shake database when either