Add hexFixedWord16#, release 0.3.12.0

Author: andrewthad

CHANGELOG.md

@@ -1,5 +1,11 @@
 # Revision history for bytesmith
 
+## 0.3.20.0 -- 2025-07-11
+
+* Add `hexFixedWord16#`
+* Drop support for base < 4.18, clean up CPP
+* Add more levity-monomorphized variants of bind, pure, and fail
+
 ## 0.3.11.1 -- 2024-02-28
 
 * Update package metadata.

bytesmith.cabal

@@ -1,6 +1,6 @@
 cabal-version:   3.0
 name:            bytesmith
-version:         0.3.11.1
+version:         0.3.12.0
 synopsis:        Nonresumable byte parser
 description:
   Parse bytes as fast as possible. This is a nonresumable parser

src/Data/Bytes/Parser.hs

@@ -106,6 +106,7 @@ module Data.Bytes.Parser
     -- get used in your original source code, GHC will not introduce them.
   , bindFromCharToLifted
   , bindFromCharToByteArrayIntInt
+  , bindFromWordToByteArrayIntInt
   , bindFromLiftedToIntPair
   , bindFromLiftedToInt
   , bindFromIntToIntPair
@@ -524,15 +525,15 @@ peekRemaining = uneffectful $ \b@(Bytes _ off len) ->
 -- | Skip while the predicate is matched. This is always inlined.
 skipWhile :: (Word8 -> Bool) -> Parser e s ()
 {-# INLINE skipWhile #-}
-skipWhile f = go
+skipWhile f = goSkipWhile
  where
-  go =
+  goSkipWhile =
     isEndOfInput >>= \case
       True -> pure ()
       False -> do
         w <- anyUnsafe
         if f w
-          then go
+          then goSkipWhile
           else unconsume 1
 
 {- | The parser @satisfy p@ succeeds for any byte for which the
@@ -697,6 +698,17 @@ bindFromCharToLifted (Parser f) g =
             runParser (g y) (# arr, b, c #) s1
     )
 
+bindFromWordToByteArrayIntInt :: Parser s e Word# -> (Word# -> Parser s e (# ByteArray#, Int#, Int# #)) -> Parser s e (# ByteArray#, Int#, Int# #)
+{-# INLINE bindFromWordToByteArrayIntInt #-}
+bindFromWordToByteArrayIntInt (Parser f) g =
+  Parser
+    ( \x@(# arr, _, _ #) s0 -> case f x s0 of
+        (# s1, r0 #) -> case r0 of
+          (# e | #) -> (# s1, (# e | #) #)
+          (# | (# y, b, c #) #) ->
+            runParser (g y) (# arr, b, c #) s1
+    )
+
 bindFromCharToByteArrayIntInt :: Parser s e Char# -> (Char# -> Parser s e (# ByteArray#, Int#, Int# #)) -> Parser s e (# ByteArray#, Int#, Int# #)
 {-# INLINE bindFromCharToByteArrayIntInt #-}
 bindFromCharToByteArrayIntInt (Parser f) g =

src/Data/Bytes/Parser/Latin.hs

@@ -99,6 +99,7 @@ module Data.Bytes.Parser.Latin
     -- *** Fixed Length
   , hexFixedWord8
   , hexFixedWord16
+  , hexFixedWord16#
   , hexFixedWord32
   , hexFixedWord64
   , hexFixedWord128
@@ -1285,6 +1286,11 @@ hexFixedWord16 e = Parser
         a), b, c #) #) #)
   )
 
+-- | Variant of hexFixedWord16 that returns an unboxed result. The result is
+-- a machine-sized word instead of a 16-bit word, but this function guarantees
+-- that only the low 16 bits may be non-zero. This is helpful because the
+-- result needs to be constructed as a machine-sized word. GHC's primitives
+-- for working with 16-bit words lower to most ISAs poorly.
 hexFixedWord16# :: e -> Parser e s Word#
 {-# noinline hexFixedWord16# #-}
 hexFixedWord16# e = uneffectfulWord# $ \chunk -> if length chunk >= 4
@@ -1295,9 +1301,9 @@ hexFixedWord16# e = uneffectfulWord# $ \chunk -> if length chunk >= 4
         !w3@(W# n3) = oneHex $ PM.indexByteArray (array chunk) (offset chunk + 3)
      in if | w0 .|. w1 .|. w2 .|. w3 /= maxBound ->
              (# |
-                (# (n0 `Exts.timesWord#` 4096##) `Exts.plusWord#`
-                   (n1 `Exts.timesWord#` 256##) `Exts.plusWord#`
-                   (n2 `Exts.timesWord#` 16##) `Exts.plusWord#`
+                (# (n0 `Exts.shiftL#` 12#) `Exts.plusWord#`
+                   (n1 `Exts.shiftL#` 8#) `Exts.plusWord#`
+                   (n2 `Exts.shiftL#` 4#) `Exts.plusWord#`
                    n3
                 ,  unI (offset chunk) +# 4#
                 ,  unI (length chunk) -# 4# #) #)