{-# LANGUAGE ImplicitParams #-} {-# LANGUAGE LambdaCase #-} {-# LANGUAGE MagicHash #-} module PossehlAnalyticsPrelude ( -- * Text conversions Text, ByteString, Word8, fmt, textToString, stringToText, showToText, textToBytesUtf8, textToBytesUtf8Lazy, bytesToTextUtf8, bytesToTextUtf8Lazy, bytesToTextUtf8Lenient, bytesToTextUtf8LenientLazy, bytesToTextUtf8Unsafe, bytesToTextUtf8UnsafeLazy, toStrict, toLazy, toStrictBytes, toLazyBytes, charToWordUnsafe, -- * IO putStrLn, putStderrLn, exitWithMessage, -- * WIP code todo, -- * Records HasField, -- * Control flow doAs, (&), (<&>), (<|>), foldMap1, foldMap', join, when, unless, guard, ExceptT (..), runExceptT, MonadThrow, throwM, MonadIO, liftIO, MonadReader, asks, Bifunctor, first, second, bimap, both, foldMap, fold, foldl', fromMaybe, mapMaybe, findMaybe, Traversable, for, for_, traverse, traverse_, traverseFold, traverseFold1, traverseFoldDefault, MonadTrans, lift, -- * Data types Coercible, coerce, Proxy (Proxy), Map, annotate, Validation (Success, Failure), failure, successes, failures, traverseValidate, traverseValidateM, traverseValidateM_, eitherToValidation, eitherToListValidation, validationToEither, These (This, That, These), eitherToThese, eitherToListThese, validationToThese, thenThese, thenValidate, NonEmpty ((:|)), singleton, nonEmpty, nonEmptyDef, toList, toNonEmptyDefault, maximum1, minimum1, Generic, Semigroup, sconcat, Monoid, mconcat, ifTrue, ifExists, Void, absurd, Identity (Identity, runIdentity), Natural, intToNatural, Contravariant, contramap, (>$<), (>&<), Profunctor, dimap, lmap, rmap, Semigroupoid, Category, (>>>), (&>>), Any, -- * Enum definition inverseFunction, inverseMap, enumerateAll, -- * Map helpers mapFromListOn, mapFromListOnMerge, -- * Error handling HasCallStack, module Data.Error, ) where import Control.Applicative ((<|>)) import Control.Category (Category, (>>>)) import Control.Monad (guard, join, unless, when) import Control.Monad.Catch (MonadThrow (throwM)) import Control.Monad.Except ( ExceptT (..), runExceptT, ) import Control.Monad.IO.Class (MonadIO, liftIO) import Control.Monad.Identity (Identity (Identity)) import Control.Monad.Reader (MonadReader, asks) import Control.Monad.Trans (MonadTrans (lift)) import Data.Bifunctor (Bifunctor, bimap, first, second) import Data.ByteString ( ByteString, ) import Data.ByteString.Lazy qualified import Data.Char qualified import Data.Coerce (Coercible, coerce) import Data.Data (Proxy (Proxy)) import Data.Error import Data.Foldable (Foldable (foldMap', toList), fold, foldl', for_, sequenceA_, traverse_) import Data.Foldable qualified as Foldable import Data.Function ((&)) import Data.Functor ((<&>)) import Data.Functor.Contravariant (Contravariant (contramap), (>$<)) import Data.Functor.Identity (Identity (runIdentity)) import Data.List.NonEmpty (NonEmpty ((:|)), nonEmpty) import Data.Map.Strict ( Map, ) import Data.Map.Strict qualified as Map import Data.Maybe (fromMaybe, mapMaybe) import Data.Maybe qualified as Maybe import Data.Profunctor (Profunctor, dimap, lmap, rmap) import Data.Semigroup (Max (Max, getMax), Min (Min, getMin), sconcat) import Data.Semigroup.Foldable (Foldable1 (fold1), foldMap1) import Data.Semigroup.Traversable (Traversable1) import Data.Semigroupoid (Semigroupoid (o)) import Data.Text ( Text, ) import Data.Text qualified import Data.Text.Encoding qualified import Data.Text.Encoding.Error qualified import Data.Text.Lazy qualified import Data.Text.Lazy.Encoding qualified import Data.These (These (That, These, This)) import Data.Traversable (for) import Data.Void (Void, absurd) import Data.Word (Word8) import GHC.Exception (errorCallWithCallStackException) import GHC.Exts (Any, RuntimeRep, TYPE, raise#) import GHC.Generics (Generic) import GHC.Natural (Natural) import GHC.Records (HasField) import GHC.Stack (HasCallStack) import PyF (fmt) import System.Exit qualified import System.IO qualified import Validation ( Validation (Failure, Success), eitherToValidation, failure, failures, successes, validationToEither, ) -- | Mark a `do`-block with the type of the Monad/Applicativ it uses. -- Only intended for reading ease and making code easier to understand, -- especially do-blocks that use unconventional monads (like Maybe or List). -- -- Example: -- -- @ -- doAs @Maybe $ do -- a <- Just 'a' -- b <- Just 'b' -- pure (a, b) -- @ doAs :: forall m a. m a -> m a doAs = id -- | Forward-applying 'contramap', like '&'/'$' and '<&>'/'<$>' but for '>$<'. (>&<) :: Contravariant f => f b -> (a -> b) -> f a (>&<) = flip contramap infixl 5 >&< -- | Forward semigroupoid application. The same as '(>>>)', but 'Semigroupoid' is not a superclass of 'Category' (yet). -- -- Specialized examples: -- -- @@ -- for functions : (a -> b) -> (b -> c) -> (a -> c) -- for Folds: Fold a b -> Fold b c -> Fold a c -- @@ (&>>) :: Semigroupoid s => s a b -> s b c -> s a c (&>>) = flip Data.Semigroupoid.o -- like >>> infixr 1 &>> -- | encode a Text to a UTF-8 encoded Bytestring textToBytesUtf8 :: Text -> ByteString textToBytesUtf8 = Data.Text.Encoding.encodeUtf8 -- | encode a lazy Text to a UTF-8 encoded lazy Bytestring textToBytesUtf8Lazy :: Data.Text.Lazy.Text -> Data.ByteString.Lazy.ByteString textToBytesUtf8Lazy = Data.Text.Lazy.Encoding.encodeUtf8 bytesToTextUtf8 :: ByteString -> Either Error Text bytesToTextUtf8 = first exceptionToError . Data.Text.Encoding.decodeUtf8' bytesToTextUtf8Lazy :: Data.ByteString.Lazy.ByteString -> Either Error Data.Text.Lazy.Text bytesToTextUtf8Lazy = first exceptionToError . Data.Text.Lazy.Encoding.decodeUtf8' -- | decode a Text from a ByteString that is assumed to be UTF-8 (crash if that is not the case) bytesToTextUtf8Unsafe :: ByteString -> Text bytesToTextUtf8Unsafe = Data.Text.Encoding.decodeUtf8 -- | decode a Text from a ByteString that is assumed to be UTF-8 (crash if that is not the case) bytesToTextUtf8UnsafeLazy :: Data.ByteString.Lazy.ByteString -> Data.Text.Lazy.Text bytesToTextUtf8UnsafeLazy = Data.Text.Lazy.Encoding.decodeUtf8 -- | decode a Text from a ByteString that is assumed to be UTF-8, -- replace non-UTF-8 characters with the replacment char U+FFFD. bytesToTextUtf8Lenient :: Data.ByteString.ByteString -> Data.Text.Text bytesToTextUtf8Lenient = Data.Text.Encoding.decodeUtf8With Data.Text.Encoding.Error.lenientDecode -- | decode a lazy Text from a lazy ByteString that is assumed to be UTF-8, -- replace non-UTF-8 characters with the replacment char U+FFFD. bytesToTextUtf8LenientLazy :: Data.ByteString.Lazy.ByteString -> Data.Text.Lazy.Text bytesToTextUtf8LenientLazy = Data.Text.Lazy.Encoding.decodeUtf8With Data.Text.Encoding.Error.lenientDecode -- | Make a lazy text strict toStrict :: Data.Text.Lazy.Text -> Text toStrict = Data.Text.Lazy.toStrict -- | Make a strict text lazy toLazy :: Text -> Data.Text.Lazy.Text toLazy = Data.Text.Lazy.fromStrict toStrictBytes :: Data.ByteString.Lazy.ByteString -> ByteString toStrictBytes = Data.ByteString.Lazy.toStrict toLazyBytes :: ByteString -> Data.ByteString.Lazy.ByteString toLazyBytes = Data.ByteString.Lazy.fromStrict textToString :: Text -> String textToString = Data.Text.unpack stringToText :: String -> Text stringToText = Data.Text.pack showToText :: (Show a) => a -> Text showToText = stringToText . show -- | Unsafe conversion between 'Char' and 'Word8'. This is a no-op and -- silently truncates to 8 bits Chars > '\255'. It is provided as -- convenience for ByteString construction. -- -- Use if you want to get the 'Word8' representation of a character literal. -- Don’t use on arbitrary characters! -- -- >>> charToWordUnsafe ',' -- 44 charToWordUnsafe :: Char -> Word8 charToWordUnsafe = fromIntegral . Data.Char.ord {-# INLINE charToWordUnsafe #-} -- | Single element in a (non-empty) list. singleton :: a -> NonEmpty a singleton a = a :| [] -- | If the given list is empty, use the given default element and return a non-empty list. nonEmptyDef :: a -> [a] -> NonEmpty a nonEmptyDef def xs = xs & nonEmpty & \case Nothing -> def :| [] Just ne -> ne -- | Construct a non-empty list, given a default value if the ist list was empty. toNonEmptyDefault :: a -> [a] -> NonEmpty a toNonEmptyDefault def xs = case xs of [] -> def :| [] (x : xs') -> x :| xs' -- | @O(n)@. Get the maximum element from a non-empty structure. maximum1 :: (Foldable1 f, Ord a) => f a -> a maximum1 xs = xs & foldMap1 Max & getMax -- | @O(n)@. Get the minimum element from a non-empty structure. minimum1 :: (Foldable1 f, Ord a) => f a -> a minimum1 xs = xs & foldMap1 Min & getMin -- | Annotate a 'Maybe' with an error message and turn it into an 'Either'. annotate :: err -> Maybe a -> Either err a annotate err = \case Nothing -> Left err Just a -> Right a -- | Map the same function over both sides of a Bifunctor (e.g. a tuple). both :: Bifunctor bi => (a -> b) -> bi a a -> bi b b both f = bimap f f -- | Find the first element for which pred returns `Just a`, and return the `a`. -- -- Example: -- @ -- >>> :set -XTypeApplications -- >>> import qualified Text.Read -- -- >>> findMaybe (Text.Read.readMaybe @Int) ["foo"] -- Nothing -- >>> findMaybe (Text.Read.readMaybe @Int) ["foo", "34.40", "34", "abc"] -- Just 34 findMaybe :: Foldable t => (a -> Maybe b) -> t a -> Maybe b findMaybe mPred list = let pred' x = Maybe.isJust $ mPred x in case Foldable.find pred' list of Just a -> mPred a Nothing -> Nothing -- | 'traverse' with a function returning 'Either' and collect all errors that happen, if they happen. -- -- Does not shortcut on error, so will always traverse the whole list/'Traversable' structure. -- -- This is a useful error handling function in many circumstances, -- because it won’t only return the first error that happens, but rather all of them. traverseValidate :: forall t a err b. Traversable t => (a -> Either err b) -> t a -> Either (NonEmpty err) (t b) traverseValidate f as = as & traverse @t @(Validation _) (eitherToListValidation . f) & validationToEither -- | 'traverse' with a function returning 'm Either' and collect all errors that happen, if they happen. -- -- Does not shortcut on error, so will always traverse the whole list/'Traversable' structure. -- -- This is a useful error handling function in many circumstances, -- because it won’t only return the first error that happens, but rather all of them. traverseValidateM :: forall t m a err b. (Traversable t, Applicative m) => (a -> m (Either err b)) -> t a -> m (Either (NonEmpty err) (t b)) traverseValidateM f as = as & traverse @t @m (\a -> a & f <&> eitherToListValidation) <&> sequenceA @t @(Validation _) <&> validationToEither -- | 'traverse_' with a function returning 'm Either' and collect all errors that happen, if they happen. -- -- Does not shortcut on error, so will always traverse the whole list/'Traversable' structure. -- -- This is a useful error handling function in many circumstances, -- because it won’t only return the first error that happens, but rather all of them. traverseValidateM_ :: forall t m a err. (Traversable t, Applicative m) => (a -> m (Either err ())) -> t a -> m (Either (NonEmpty err) ()) traverseValidateM_ f as = as & traverse @t @m (\a -> a & f <&> eitherToListValidation) <&> sequenceA_ @t @(Validation _) <&> validationToEither -- | Like 'eitherToValidation', but puts the Error side into a NonEmpty list -- to make it combine with other validations. -- -- See also 'validateEithers', if you have a list of Either and want to collect all errors. eitherToListValidation :: Either a c -> Validation (NonEmpty a) c eitherToListValidation = first singleton . eitherToValidation -- | Convert an 'Either' to a 'These'. eitherToThese :: Either err a -> These err a eitherToThese (Left err) = This err eitherToThese (Right a) = That a -- | Like 'eitherToThese', but puts the Error side into a NonEmpty list -- to make it combine with other theses. eitherToListThese :: Either err a -> These (NonEmpty err) a eitherToListThese (Left e) = This (singleton e) eitherToListThese (Right a) = That a -- | Convert a 'Validation' to a 'These'. validationToThese :: Validation err a -> These err a validationToThese (Failure err) = This err validationToThese (Success a) = That a -- | Nested '>>=' of a These inside some other @m@. -- -- Use if you want to collect errors and successes, and want to chain multiple function returning 'These'. thenThese :: (Monad m, Semigroup err) => (a -> m (These err b)) -> m (These err a) -> m (These err b) thenThese f x = do th <- x join <$> traverse f th -- | Nested validating bind-like combinator inside some other @m@. -- -- Use if you want to collect errors, and want to chain multiple functions returning 'Validation'. thenValidate :: (Monad m) => (a -> m (Validation err b)) -> m (Validation err a) -> m (Validation err b) thenValidate f x = eitherToValidation <$> do x' <- validationToEither <$> x case x' of Left err -> pure $ Left err Right a -> validationToEither <$> f a -- | Put the text to @stderr@. putStderrLn :: Text -> IO () putStderrLn msg = System.IO.hPutStrLn System.IO.stderr $ textToString msg exitWithMessage :: Text -> IO a exitWithMessage msg = do putStderrLn msg System.Exit.exitWith $ System.Exit.ExitFailure (-1) -- | Run some function producing applicative over a traversable data structure, -- then collect the results in a Monoid. -- -- Very helpful with side-effecting functions returning @(Validation err a)@: -- -- @ -- let -- f :: Text -> IO (Validation (NonEmpty Error) Text) -- f t = pure $ if t == "foo" then Success t else Failure (singleton ("not foo: " <> t)) -- -- in traverseFold f [ "foo", "bar", "baz" ] -- == Failure ("not foo bar" :| ["not foo baz"]) -- @ -- -- … since @(Semigroup err => Validation err a)@ is a @Semigroup@/@Monoid@ itself. traverseFold :: (Applicative ap, Traversable t, Monoid m) => (a -> ap m) -> t a -> ap m traverseFold f xs = -- note: could be weakened to (Foldable t) via `getAp . foldMap (Ap . f)` fold <$> traverse f xs {-# INLINE traverseFold #-} -- | Like 'traverseFold', but fold over a semigroup instead of a Monoid, by providing a starting element. traverseFoldDefault :: (Applicative ap, Traversable t, Semigroup m) => m -> (a -> ap m) -> t a -> ap m traverseFoldDefault def f xs = foldDef def <$> traverse f xs where foldDef = foldr (<>) {-# INLINE traverseFoldDefault #-} -- | Same as 'traverseFold', but with a 'Semigroup' and 'Traversable1' restriction. traverseFold1 :: (Applicative ap, Traversable1 t, Semigroup s) => (a -> ap s) -> t a -> ap s -- note: cannot be weakened to (Foldable1 t) because there is no `Ap` for Semigroup (No `Apply` typeclass) traverseFold1 f xs = fold1 <$> traverse f xs {-# INLINE traverseFold1 #-} -- | Use this in places where the code is still to be implemented. -- -- It always type-checks and will show a warning at compile time if it was forgotten in the code. -- -- Use instead of 'error' and 'undefined' for code that hasn’t been written. -- -- Uses the same trick as https://hackage.haskell.org/package/protolude-0.3.0/docs/src/Protolude.Error.html#error {-# WARNING todo "'todo' (undefined code) remains in code" #-} todo :: forall (r :: RuntimeRep). forall (a :: TYPE r). HasCallStack => a todo = raise# (errorCallWithCallStackException "This code was not yet implemented: TODO" ?callStack) -- | Convert an integer to a 'Natural' if possible -- -- Named the same as the function from "GHC.Natural", but does not crash. intToNatural :: Integral a => a -> Maybe Natural intToNatural i = if i < 0 then Nothing else Just $ fromIntegral i -- | @inverseFunction f@ creates a function that is the inverse of a given function -- @f@. It does so by constructing 'M.Map' internally for each value @f a@. The -- implementation makes sure that the 'M.Map' is constructed only once and then -- shared for every call. -- -- __Memory usage note:__ don't inverse functions that have types like 'Int' -- as their result. In this case the created 'M.Map' will have huge size. -- -- The complexity of reversed mapping is \(\mathcal{O}(\log n)\). -- -- __Performance note:__ make sure to specialize monomorphic type of your functions -- that use 'inverseFunction' to avoid 'M.Map' reconstruction. -- -- One of the common 'inverseFunction' use-case is inverting the 'show' or a 'show'-like -- function. -- -- >>> data Color = Red | Green | Blue deriving (Show, Enum, Bounded) -- >>> parse = inverseFunction show :: String -> Maybe Color -- >>> parse "Red" -- Just Red -- >>> parse "Black" -- Nothing -- -- __Correctness note:__ 'inverseFunction' expects /injective function/ as its argument, -- i.e. the function must map distinct arguments to distinct values. -- -- Typical usage of this function looks like this: -- -- @ -- __data__ GhcVer -- = Ghc802 -- | Ghc822 -- | Ghc844 -- | Ghc865 -- | Ghc881 -- __deriving__ ('Eq', 'Ord', 'Show', 'Enum', 'Bounded') -- -- showGhcVer :: GhcVer -> 'Text' -- showGhcVer = \\__case__ -- Ghc802 -> "8.0.2" -- Ghc822 -> "8.2.2" -- Ghc844 -> "8.4.4" -- Ghc865 -> "8.6.5" -- Ghc881 -> "8.8.1" -- -- parseGhcVer :: 'Text' -> 'Maybe' GhcVer -- parseGhcVer = 'inverseFunction' showGhcVer -- -- Taken from relude’s @Relude.Extra.Enum@. inverseFunction :: forall a k. (Bounded a, Enum a, Ord k) => (a -> k) -> (k -> Maybe a) inverseFunction f k = Map.lookup k $ inverseMap f -- | Like `inverseFunction`, but instead of returning the function -- it returns a mapping from all possible outputs to their possible inputs. -- -- This has the same restrictions of 'inverseFunction'. inverseMap :: forall a k. (Bounded a, Enum a, Ord k) => (a -> k) -> Map k a inverseMap f = enumerateAll <&> (\a -> (f a, a)) & Map.fromList -- | All possible values in this enum. enumerateAll :: (Enum a, Bounded a) => [a] enumerateAll = [minBound .. maxBound] -- | Create a 'Map' from a list of values, extracting the map key from each value. Like 'Map.fromList'. -- -- Attention: if the key is not unique, the earliest value with the key will be in the map. mapFromListOn :: Ord key => (a -> key) -> [a] -> Map key a mapFromListOn f xs = xs <&> (\x -> (f x, x)) & Map.fromList -- | Create a 'Map' from a list of values, merging multiple values at the same key with '<>' (left-to-right) -- -- `f` has to extract the key and value. Value must be mergable. -- -- Attention: if the key is not unique, the earliest value with the key will be in the map. mapFromListOnMerge :: (Ord key, Semigroup s) => (a -> (key, s)) -> [a] -> Map key s mapFromListOnMerge f xs = xs <&> (\x -> f x) & Map.fromListWith -- we have to flip (`<>`) because `Map.fromListWith` merges its values “the other way around” (flip (<>)) -- | If the predicate is true, return the @m@, else 'mempty'. -- -- This can be used (together with `ifExists`) to e.g. create lists with optional elements: -- -- >>> import Data.Monoid (Sum(..)) -- -- >>> :{ mconcat [ -- ifTrue (1 == 1) [1], -- [2, 3, 4], -- ifTrue False [5], -- ] -- :} -- [1,2,3,4] -- -- Or any other Monoid: -- -- >>> mconcat [ Sum 1, ifTrue (1 == 1) (Sum 2), Sum 3 ] -- Sum {getSum = 6} ifTrue :: Monoid m => Bool -> m -> m ifTrue pred' m = if pred' then m else mempty -- | If the given @Maybe@ is @Just@, return the @m@, else return mempty. -- This can be used (together with `ifTrue`) to e.g. create lists with optional elements: -- -- >>> import Data.Monoid (Sum(..)) -- -- >>> :{ mconcat [ -- ifExists (Just [1]), -- [2, 3, 4], -- ifExists Nothing, -- ] -- :} -- [1,2,3,4] -- -- Or any other Monoid: -- -- >>> mconcat [ Sum 1, ifExists (Just (Sum 2)), Sum 3 ] -- Sum {getSum = 6} ifExists :: Monoid m => Maybe m -> m ifExists = fold