diff --git a/bytestring-lexing.cabal b/bytestring-lexing.cabal --- a/bytestring-lexing.cabal +++ b/bytestring-lexing.cabal @@ -6,11 +6,11 @@ -- ---------------------------------------------------------------- --- wren gayle romano ~ 2023.11.15 +-- wren gayle romano ~ 2024-04-05 ---------------------------------------------------------------- Name: bytestring-lexing -Version: 0.5.0.11 +Version: 0.5.0.12 Build-Type: Simple Stability: provisional Homepage: https://wrengr.org/software/hackage.html @@ -58,8 +58,9 @@ GHC ==8.10.3, GHC ==9.0.1, GHC ==9.2.4, - GHC ==9.4.4, - GHC ==9.6.1 + GHC ==9.4.8, + GHC ==9.6.4, + GHC ==9.8.1 Source-Repository head Type: git diff --git a/src/Data/ByteString/Lex/Fractional.hs b/src/Data/ByteString/Lex/Fractional.hs --- a/src/Data/ByteString/Lex/Fractional.hs +++ b/src/Data/ByteString/Lex/Fractional.hs @@ -23,20 +23,20 @@ -- packSigned -- * Decimal conversions , readDecimal - -- packDecimal + -- TODO: packDecimal -- TODO: asDecimal -- this will be really hard to make efficient... -- * Hexadecimal conversions , readHexadecimal - -- packHexadecimal - -- asHexadecimal + -- TODO: packHexadecimal + -- TODO: asHexadecimal -- * Octal conversions , readOctal - -- packOctal - -- asOctal -- this will be really hard to make efficient... + -- TODO: packOctal + -- TODO: asOctal -- this will be really hard to make efficient... -- * Exponential conversions , readExponential - -- packExponential - -- asExponential + -- TODO: packExponential + -- TODO: asExponential -- * Precision-limited conversions , decimalPrecision , readDecimalLimited diff --git a/src/Data/ByteString/Lex/Integral.hs b/src/Data/ByteString/Lex/Integral.hs --- a/src/Data/ByteString/Lex/Integral.hs +++ b/src/Data/ByteString/Lex/Integral.hs @@ -33,7 +33,7 @@ -- * Octal conversions , readOctal , packOctal - -- asOctal -- this will be really hard to make efficient... + -- TODO: asOctal -- this will be really hard to make efficient... ) where import Data.ByteString (ByteString) @@ -79,7 +79,8 @@ ----- Decimal {- --- We unroll this definition in order to reduce the number of conversions from native Int to the Integral type. +-- We unroll this definition in order to reduce the number of +-- conversions from native Int to the Integral type. readDecimalSimple :: (Integral a) => ByteString -> Maybe (a, ByteString) readDecimalSimple = start where diff --git a/src/Data/ByteString/Lex/Internal.hs b/src/Data/ByteString/Lex/Internal.hs --- a/src/Data/ByteString/Lex/Internal.hs +++ b/src/Data/ByteString/Lex/Internal.hs @@ -1,10 +1,10 @@ {-# OPTIONS_GHC -Wall -fwarn-tabs #-} {-# LANGUAGE BangPatterns #-} ---------------------------------------------------------------- --- 2021.10.17 +-- 2024-04-11 -- | -- Module : Data.ByteString.Lex.Internal --- Copyright : Copyright (c) 2010--2021 wren gayle romano +-- Copyright : Copyright (c) 2010--2024 wren gayle romano -- License : BSD2 -- Maintainer : wren@cpan.org -- Stability : provisional @@ -62,7 +62,8 @@ ---------------------------------------------------------------- ----- Integral logarithms --- TODO: cf. integer-gmp:GHC.Integer.Logarithms made available in version 0.3.0.0 (ships with GHC 7.2.1). +-- TODO: cf. integer-gmp:GHC.Integer.Logarithms made available in +-- version 0.3.0.0 (ships with GHC 7.2.1). -- @@ -70,7 +71,7 @@ -- -- modified to use 'quot' instead of 'div', to ensure strictness, -- and using more guard notation (but this last one's compiled --- away). See @./test/bench/BenchNumDigits.hs@ for other implementation +-- away). See @./bench/BenchNumDigits.hs@ for other implementation -- choices. -- -- | @numDigits b n@ computes the number of base-@b@ digits required @@ -93,16 +94,27 @@ | b0 <= 1 = error (_numDigits ++ _nonpositiveBase) | n0 < 0 = error (_numDigits ++ _negativeNumber) -- BUG: need to check n0 to be sure we won't overflow Int - | otherwise = 1 + fst (ilog b0 n0) + | otherwise = finish (ilog b0 n0) where + finish (ND e _) = 1 + e ilog !b !n - | n < b = (0, n) - | r < b = ((,) $! 2*e) r - | otherwise = ((,) $! 2*e+1) $! (r `quot` b) + | n < b = ND 0 n + -- TODO(2024-04-11): Check core to see whether these @(2*)@ + -- ops are properly weakened to shifts. + | r < b = ND (2*e) r + | otherwise = ND (2*e+1) (r `quot` b) where - (e, r) = ilog (b*b) n + -- TODO(2024-04-11): Benchmark this lazy-pattern matching, + -- vs using a strict pattern (and alas less guard-notation, + -- to ensure we only evaluate it when needed). + ND e r = ilog (b*b) n +-- TODO(2024-04-11): Benchmark this change in the implementation +-- (relative to using @(,)@ and @($!)@). Also, need to re-run all +-- the benchmarks anyways, to see how things've changed on newer GHC. +data ND = ND {-#UNPACK#-}!Int !Integer + -- | Compute the number of base-@2^p@ digits required to represent a -- number @n@. N.B., this implementation is unsafe and will throw -- errors if the base power is non-positive, or if the number is @@ -134,6 +146,8 @@ | n0 < 0 = error (_numDecimalDigits ++ _negativeNumber) -- Unfortunately this causes significant (1.2x) slowdown since -- GHC can't see it will always fail for types other than Integer... + -- TODO(2024-04-11): See if we can't do more static-analysis + -- code to optimize this path (a~la my C++ safe comparisons) | n0 > limit = numDigits 10 (toInteger n0) | otherwise = go 1 (fromIntegral n0 :: Word64) where