criterion-measurement 0.1.3.0 → 0.1.4.0

raw patch · 4 files changed

+29/−9 lines, 4 filesPVP ok

version bump matches the API change (PVP)

API changes (from Hackage documentation)

Files

• changelog.md +5/−0
• criterion-measurement.cabal +4/−3
• src/Criterion/Measurement/Types.hs +2/−2
• src/Criterion/Measurement/Types/Internal.hs +18/−4

@@ -1,3 +1,8 @@+0.1.4.0++* Fix a bug that occurred with GHC 9.2.4 or later that would cause incorrect+  measurements.+ 0.1.3.0  * Change `criterion_rdtsc` to return `mach_absolute_time` on macOS. This is a

@@ -1,5 +1,5 @@ name:                criterion-measurement-version:             0.1.3.0+version:             0.1.4.0 synopsis:            Criterion measurement functionality and associated types description:         Measurement-related functionality extracted from Criterion, with minimal dependencies. The rationale for this is to enable alternative analysis front-ends. homepage:            https://github.com/haskell/criterion@@ -22,8 +22,9 @@   GHC==8.4.4,   GHC==8.6.5,   GHC==8.8.4,-  GHC==8.10.4,-  GHC==9.0.1+  GHC==8.10.7,+  GHC==9.0.2,+  GHC==9.2.2  flag fast   description: compile without optimizations

@@ -563,7 +563,7 @@     -- ^ Action that creates the environment for a single run.     -> (env -> IO b)     -- ^ Function returning the IO action that should be benchmarked with the-    -- newly genereted environment.+    -- newly generated environment.     -> Benchmarkable perRunEnv alloc = perRunEnvWithCleanup alloc noop @@ -578,7 +578,7 @@     -- ^ Clean up the created environment.     -> (env -> IO b)     -- ^ Function returning the IO action that should be benchmarked with the-    -- newly genereted environment.+    -- newly generated environment.     -> Benchmarkable perRunEnvWithCleanup alloc clean work = bm { perRun = True }   where

@@ -20,6 +20,7 @@ module Criterion.Measurement.Types.Internal (fakeEnvironment, nf', whnf') where  import Data.Int (Int64)+import Control.Exception  -- | A dummy environment that is passed to functions that create benchmarks -- from environments when no concrete environment is available.@@ -45,7 +46,17 @@ -- benchmark code itself could be changed by the user's optimization level. By -- marking them @NOINLINE@, the core benchmark code is always the same. ----- See #183 and #184 for discussion.+-- Finally, it's important that both branches of the loop depend on the state+-- token from the IO action. This is achieved by using `evaluate` rather than `let !y = f x`+-- in order to force the value to whnf. `evaluate` is in the IO monad and therefore the state+-- token needs to be passed through the loop.+--+-- See ghc#21948 where a change in eta-expansion behaviour+-- caused the work to be performed in the wrong place because the otherwise branch+-- did not depend on the state token at all, and the whole loop could be evaluated to+-- a single return function before being run in the IO monad.+--+-- See #183, #184 and #264 for discussion.  -- | Generate a function which applies an argument to a function a -- given number of times, reducing the result to normal form.@@ -53,8 +64,9 @@ nf' reduce f x = go   where     go n | n <= 0    = return ()-         | otherwise = let !y = f x-                       in reduce y `seq` go (n-1)+         | otherwise = do+            y <- evaluate (f x)+            reduce y `seq` go (n-1) {-# NOINLINE nf' #-}  -- | Generate a function which applies an argument to a function a@@ -63,5 +75,7 @@ whnf' f x = go   where     go n | n <= 0    = return ()-         | otherwise = f x `seq` go (n-1)+         | otherwise = do+            _ <- evaluate (f x)+            go (n-1) {-# NOINLINE whnf' #-}