Add a function to split text at the first occurrence of a character #625

Activity

1. Bodigrim commented on Jan 13, 2025

   

   Bodigrim

   on Jan 13, 2025

   Contributor

   More actions

   IMHO split1 c = fmap (drop 1) . break (== c) is short enough and does not suffer from any significant performance penalty: drop 1 is constant-time, negligible in comparison to the linear time of break.

   Note that the function from byteslice has a different signature, with Maybe.
2. sol commented on Apr 5, 2025

   

   sol

   on Apr 5, 2025

   Member

   More actions

   @Bodigrim comparing

   split1 c = fmap (drop 1) . (breakOn $ singleton c)

   vs

   split1 c = fmap (drop 1) . break (== c)

   The first option, using breakOn, should be significantly faster than the second option, no?

   Rational:

   breakOn can look at the needle and choose the optimal strategy (specifically if c is in the ASCII range, it can just do a sequential scan).

   break on the other hand always needs to decode to Char.
3. sol commented on Apr 5, 2025

   

   sol

   on Apr 5, 2025

   Member

   More actions

   Corresponding code:
   

   text/src/Data/Text/Internal/Search.hs

   Lines 49 to 56 in 5e57460

   	indices :: Text -- ^ Substring to search for (@needle@)
   	-> Text -- ^ Text to search in (@haystack@)
   	-> [Int]
   	indices needle@(Text narr noff nlen)
   	| nlen == 1 = scanOne (A.unsafeIndex narr noff)
   	| nlen <= 0 = const []
   	| otherwise = indices' needle
   	{-# INLINE indices #-}

   text/src/Data/Text/Internal/Search.hs

   Lines 93 to 100 in 5e57460

   	scanOne :: Word8 -> Text -> [Int]
   	scanOne c (Text harr hoff hlen) = loop 0
   	where
   	loop !i
   	| i >= hlen = []
   	| A.unsafeIndex harr (hoff+i) == c = i : loop (i+1)
   	| otherwise = loop (i+1)
   	{-# INLINE scanOne #-}
4. Bodigrim commented on Apr 5, 2025

   

   Bodigrim

   on Apr 5, 2025

   Contributor

   More actions

   @sol yes, I'd expect breakOn to be faster.