{-# LANGUAGE OverloadedStrings #-} {-# LANGUAGE TypeFamilies #-} module Main (main) where import Control.DeepSeq import Criterion.Main import Data.List.NonEmpty (NonEmpty (..)) import Data.Semigroup ((<>)) import Data.Text (Text) import Data.Void import Text.Megaparsec import Text.Megaparsec.Char import qualified Data.List.NonEmpty as NE import qualified Data.Set as E import qualified Data.Text as T import qualified Text.Megaparsec.Char.Lexer as L -- | The type of parser that consumes 'String's. type Parser = Parsec Void Text main :: IO () main = defaultMain [ bparser "string" manyAs (string . fst) , bparser "string'" manyAs (string' . fst) , bparser "many" manyAs (const $ many (char 'a')) , bparser "some" manyAs (const $ some (char 'a')) , bparser "choice" (const "b") (choice . fmap char . manyAsB' . snd) , bparser "count" manyAs (\(_,n) -> count n (char 'a')) , bparser "count'" manyAs (\(_,n) -> count' 1 n (char 'a')) , bparser "endBy" manyAbs' (const $ endBy (char 'a') (char 'b')) , bparser "endBy1" manyAbs' (const $ endBy1 (char 'a') (char 'b')) , bparser "manyTill" manyAsB (const $ manyTill (char 'a') (char 'b')) , bparser "someTill" manyAsB (const $ someTill (char 'a') (char 'b')) , bparser "sepBy" manyAbs (const $ sepBy (char 'a') (char 'b')) , bparser "sepBy1" manyAbs (const $ sepBy1 (char 'a') (char 'b')) , bparser "sepEndBy" manyAbs' (const $ sepEndBy (char 'a') (char 'b')) , bparser "sepEndBy1" manyAbs' (const $ sepEndBy1 (char 'a') (char 'b')) , bparser "skipMany" manyAs (const $ skipMany (char 'a')) , bparser "skipSome" manyAs (const $ skipSome (char 'a')) , bparser "skipCount" manyAs (\(_,n) -> skipCount n (char 'a')) , bparser "skipManyTill" manyAsB (const $ skipManyTill (char 'a') (char 'b')) , bparser "skipSomeTill" manyAsB (const $ skipSomeTill (char 'a') (char 'b')) , bparser "takeWhileP" manyAs (const $ takeWhileP Nothing (== 'a')) , bparser "takeWhile1P" manyAs (const $ takeWhile1P Nothing (== 'a')) , bparser "decimal" mkInt (const (L.decimal :: Parser Integer)) , bparser "octal" mkInt (const (L.octal :: Parser Integer)) , bparser "hexadecimal" mkInt (const (L.hexadecimal :: Parser Integer)) , bparser "scientific" mkInt (const L.scientific) , bgroup "" [bbundle "single error" n [n] | n <- stdSeries] , bbundle "2 errors" 1000 [1, 1000] , bbundle "4 errors" 1000 [1, 500, 1000] , bbundle "100 errors" 1000 [10,20..1000] , breachOffset 0 1000 , breachOffset 0 2000 , breachOffset 0 4000 , breachOffset 1000 1000 , breachOffsetNoLine 0 1000 , breachOffsetNoLine 0 2000 , breachOffsetNoLine 0 4000 , breachOffsetNoLine 1000 1000 ] -- | Perform a series to measurements with the same parser. bparser :: NFData a => String -- ^ Name of the benchmark group -> (Int -> Text) -- ^ How to construct input -> ((Text, Int) -> Parser a) -- ^ The parser receiving its future input -> Benchmark -- ^ The benchmark bparser name f p = bgroup name (bs <$> stdSeries) where bs n = env (return (f n, n)) (bench (show n) . nf p') p' (s,n) = parse (p (s,n)) "" s -- | Bench the 'errorBundlePretty' function. bbundle :: String -- ^ Name of the benchmark -> Int -- ^ Number of lines in input stream -> [Int] -- ^ Lines with parse errors -> Benchmark bbundle name totalLines sps = let s = take (totalLines * 80) (cycle as) as = replicate 79 'a' ++ "\n" f l = TrivialError (20 + l * 80) (Just $ Tokens ('a' :| "")) (E.singleton $ Tokens ('b' :| "")) bundle :: ParseErrorBundle String Void bundle = ParseErrorBundle { bundleErrors = f <$> NE.fromList sps , bundlePosState = PosState { pstateInput = s , pstateOffset = 0 , pstateSourcePos = initialPos "" , pstateTabWidth = defaultTabWidth , pstateLinePrefix = "" } } in bench ("errorBundlePretty-" ++ show totalLines ++ "-" ++ name) (nf errorBundlePretty bundle) -- | Bench the 'reachOffset' function. breachOffset :: Int -- ^ Starting offset in 'PosState' -> Int -- ^ Offset to reach -> Benchmark breachOffset o0 o1 = bench ("reachOffset-" ++ show o0 ++ "-" ++ show o1) (nf f (o0 * 80, o1 * 80)) where f :: (Int, Int) -> (SourcePos, PosState Text) f (startOffset, targetOffset) = let (x, _, y) = reachOffset targetOffset PosState { pstateInput = manyAs (targetOffset - startOffset) , pstateOffset = startOffset , pstateSourcePos = initialPos "" , pstateTabWidth = defaultTabWidth , pstateLinePrefix = "" } in (x, y) -- | Bench the 'reachOffsetNoLine' function. breachOffsetNoLine :: Int -- ^ Starting offset in 'PosState' -> Int -- ^ Offset to reach -> Benchmark breachOffsetNoLine o0 o1 = bench ("reachOffsetNoLine-" ++ show o0 ++ "-" ++ show o1) (nf f (o0 * 80, o1 * 80)) where f :: (Int, Int) -> (SourcePos, PosState Text) f (startOffset, targetOffset) = reachOffsetNoLine targetOffset PosState { pstateInput = manyAs (targetOffset - startOffset) , pstateOffset = startOffset , pstateSourcePos = initialPos "" , pstateTabWidth = defaultTabWidth , pstateLinePrefix = "" } -- | The series of sizes to try as part of 'bparser'. stdSeries :: [Int] stdSeries = [500,1000,2000,4000] ---------------------------------------------------------------------------- -- Helpers -- | Generate that many \'a\' characters. manyAs :: Int -> Text manyAs n = T.replicate n "a" -- | Like 'manyAs', but interspersed with \'b\'s. manyAbs :: Int -> Text manyAbs n = T.take (if even n then n + 1 else n) (T.replicate n "ab") -- | Like 'manyAs', but with a \'b\' added to the end. manyAsB :: Int -> Text manyAsB n = manyAs n <> "b" -- | Like 'manyAsB', but returns a 'String'. manyAsB' :: Int -> String manyAsB' n = replicate n 'a' ++ "b" -- | Like 'manyAbs', but ends in a \'b\'. manyAbs' :: Int -> Text manyAbs' n = T.take (if even n then n else n + 1) (T.replicate n "ab") -- | Render an 'Integer' with the number of digits linearly dependent on the -- argument. mkInt :: Int -> Text mkInt n = (T.pack . show) ((10 :: Integer) ^ (n `quot` 100))