Nonidem2.0 -- merge non-idempotent branch once it is fixed

haskell/DEVLOG.md

@@ -594,6 +594,42 @@ Haven't seen this one before, a compile failure on the cfa package:
           ...
     ld: symbol(s) not found for architecture x86_64
 
+
+[2014.01.24] {Failure after recent nonidem merge}
+-------------------------------------------------
+
+    Failed to install lvish-1.2.0.0
+    Last 10 lines of the build log ( /ffh/ryan/cloud_drive/working_copies/lvars/lvars/haskell/.cabal-sandbox/logs/lvish-1.2.0.0.log ):
+     Passed  5           64           69
+     Failed  0           1            1
+     Total   5           65           70
+     [*] test-framework exiting with: ExitFailure 1
+     [*] GC finished on main thread.
+     [*] Main thread exiting.
+    Test suite test-lvish: FAIL
+    Test suite logged to:
+    dist/dist-sandbox-38f74c87/test/lvish-1.2.0.0-test-lvish.log
+    0 of 1 test suites (0 of 1 test cases) passed.
+
+Specifically, the failure is:
+
+    ThreadKilled exception inside child thread, ThreadId 79270 (not propagating!): "worker thread"
+      v4: [Failed]
+    Bad test outcome--exception: PutAfterFreezeExn "Attempt to change a frozen LVar"
+
+It's proving a bit hard to reproduce, however.
+
+Also... test after i9h is apparently deadlocking on -N4.  Hmm.. why
+are timeouts not working? 
+  (And I'm getting plenty of the blocked-indefinitely errors in
+  various papers... these are probably the same failure but sometimes
+  the GC turns it into an exception.  Not sure why timeouts aren't 
+  working consistently on this branch.  I see them sometimes.)
+
+Ah, ok, I can get failures on AddRemoveSetTests 
+
+
+
 [2014.01.24] {More scheduler debugging}
 ----------------------------------------
 
@@ -669,6 +705,8 @@ default, and I got this:
     ERROR: Final continuation of Par computation was duplicated, in spite of GET_ONCE!
 
 
+
+
 [2014.10.24] {Criterion Microbenchmarking}
 ------------------------------------------
 
@@ -829,3 +867,38 @@ because this is a loop IN the par monad rather than the IO monad....
 	    True -> lvl1_rcQD `cast` ...
 	  }; }
 
+
+
+
+[2014.10.26] {Debugging nonidem branch}
+----------------------------------------
+
+On the main (2.0) branch we're having some trouble with duplicated
+gets inspite of `-fgetonce`.  I wanted to try (finally) merging the
+nonidem branch instead.  I've done most of the merges to bring it up
+to speed.  However, not only does it still have errors like this:
+
+  "Bad test outcome--exception: Final continuation of Par computation
+   was duplicated, in spite of GET_ONCE!"
+
+But, it also has some divergences.  Timeouts on the following:
+
+      v3b: [Failed]
+    ERROR: <<timeout>>
+
+      i3c: [Failed]
+    Got the wrong exception, expected one of the strings: ["Attempt to change a frozen LVar"]
+    Instead got this exception:
+      "<<timeout>>"
+
+      v3d: [Failed]
+    ERROR: <<timeout>>
+
+      v3e: [Failed]
+    ERROR: <<timeout>>
+
+      v8a: [Failed]
+    ERROR: <<timeout>>
+
+      v8b: [Failed]
+    ERROR: <<timeout>>

haskell/lvish/Control/LVish.hs

@@ -119,8 +119,9 @@ import           Control.LVish.Types
 import           Control.LVish.Internal as I
 import           Control.LVish.Basics as B
 import           Control.LVish.Logical
-import qualified Internal.Control.LVish.SchedIdempotent as L
-import           Control.LVish.SchedIdempotentInternal (State)
+import qualified Internal.Control.LVish.Sched as L
+import           Control.LVish.SchedQueue (State)
+
 import           Control.Par.EffectSigs
 import           Control.LVish.Logging (OutDest(..))
 import           Data.LVar.IVar 

haskell/lvish/Control/LVish/Basics.hs

@@ -42,7 +42,7 @@ import qualified Data.Foldable    as F
 import           Control.Exception (Exception, SomeException)
 import           Control.LVish.Internal as I
 import           Control.LVish.DeepFrz.Internal (Frzn, Trvrsbl)
-import qualified Internal.Control.LVish.SchedIdempotent as L
+import qualified Internal.Control.LVish.Sched as L
 import qualified Control.LVish.Logging as Lg
 import           Control.LVish.Types
 import           Control.Par.EffectSigs

haskell/lvish/Control/LVish/DeepFrz.hs

@@ -55,7 +55,7 @@ import GHC.Prim (unsafeCoerce#)
 import Control.LVish.DeepFrz.Internal (DeepFrz(..), NonFrzn, Frzn, Trvrsbl)
 import Control.LVish.Internal (Par(WrapPar))
 import Control.Par.EffectSigs
-import Internal.Control.LVish.SchedIdempotent (runPar, runParIO)
+import Internal.Control.LVish.Sched (runPar, runParIO)
 --------------------------------------------------------------------------------
 
 -- | Under normal conditions, calling a `freeze` operation inside a

haskell/lvish/Control/LVish/Internal.hs

@@ -34,7 +34,7 @@ module Control.LVish.Internal
 
 import           Control.LVish.MonadToss
 import           Control.Applicative
-import qualified Internal.Control.LVish.SchedIdempotent as L
+import qualified Internal.Control.LVish.Sched as L
 import           Control.LVish.DeepFrz.Internal (Frzn, Trvrsbl)
 import           Control.Par.EffectSigs
 
@@ -74,7 +74,7 @@ newtype Par :: EffectSig -> * -> * -> * where
 -- to.
 
 -- LK: I don't care if we use `a` and `d` or `all` and `delt`, but why
--- not be consistent between here and SchedIdempotent.hs?  Also, what
+-- not be consistent between here and Sched.hs?  Also, what
 -- does `all` mean?
 newtype LVar s all delt = WrapLVar { unWrapLVar :: L.LVar all delt }
 

haskell/lvish/Control/LVish/Logical.hs

@@ -7,7 +7,7 @@ module Control.LVish.Logical (asyncAnd, asyncOr, andMap, orMap) where
 import Control.Par.EffectSigs
 import Control.LVish.Basics
 import Control.LVish.Internal (Par(WrapPar), unsafeDet)
-import Internal.Control.LVish.SchedIdempotent (liftIO, HandlerPool)
+import Internal.Control.LVish.Sched    (liftIO, HandlerPool)
 import Data.LVar.IVar    as IV
 
 import qualified Data.Atomics.Counter as C

haskell/lvish/Control/LVish/SchedIdempotentInternal.hs → haskell/lvish/Control/LVish/SchedQueue.hs

@@ -3,9 +3,10 @@
 {-# LANGUAGE NamedFieldPuns, BangPatterns #-}
 {-# LANGUAGE RecursiveDo #-}
 
-module Control.LVish.SchedIdempotentInternal (
-  State(logger, no), 
-  new, number, next, pushWork, nullQ, yieldWork, currentCPU, setStatus, await, prng
+
+module Control.LVish.SchedQueue (
+  State(logger, no),
+  new, number, idemp, next, pushWork, yieldWork, currentCPU, setStatus, await, prng
   ) where
 
 
@@ -85,6 +86,7 @@ popOther = popMine
 -- All the state relevant to a single worker thread
 data State a s = State
     { no       :: {-# UNPACK #-} !Int, -- ^ The number of this worker
+      idemp    :: Bool,                -- ^ are we assuming task idempotence?
       numWorkers :: Int,               -- ^ Total number of workers in this runPar
       prng     :: IORef StdGen,        -- ^ core-local random number generation
       status   :: IORef s,             -- ^ A thread-local flag
@@ -189,7 +191,13 @@ new DbgCfg{dbgDests,dbgRange,dbgScheduling} numWorkers s = do
         workpool <- newDeque
         status   <- newIORef s
         prng     <- newIORef $ mkStdGen i
-        return State { no = i, workpool, idle, status, states, prng, logger, numWorkers }
+        return State { no = i,
+#ifdef LVISH_DEDUP
+                       idemp = False,
+#else    
+                       idemp = True,
+#endif
+                       workpool, idle, status, states, prng, logger, numWorkers }
   rec states <- forM [0..(numWorkers-1)] $ mkState states
   return (logger,states)
 

haskell/lvish/Control/LVish/Unsafe.hs

@@ -7,7 +7,7 @@ module Control.LVish.Unsafe() where
 
 import Control.LVish.Internal
 import Control.Monad.IO.Class
-import qualified Internal.Control.LVish.SchedIdempotent as L
+import qualified Internal.Control.LVish.Sched as L
 
 instance MonadIO (Par e s) where
   liftIO = WrapPar . L.liftIO   

haskell/lvish/Control/Reagent.hs

@@ -39,4 +39,4 @@ postCommit :: Reagent a -> (a -> IO b) -> Reagent b
 postCommit r f succ fail = r (\x -> f x >>= succ) fail
 
 choice :: Reagent a -> Reagent a -> Reagent a
-choice = error "TODO"
+choice = error "TODO"