optimize cleanUpBSNulls function

[Paczesiowa]

this uses a lot less memory than the concatMap one. in our app there are queries with length of 10k chars (inserting files to db) and the concatMap one was creating huge amounts of small bytestrings, which would consume additional 1gb of memory and a few extra seconds for gc.

[dcoutts]

Here's a simpler version using the new bytestring builder:

import qualified Data.ByteString as B
import qualified Data.ByteString.Lazy as BL
import qualified Data.ByteString.Builder as B
import qualified Data.ByteString.Builder.Prim as BP
import Data.ByteString.Builder.Prim ((>$<), (>*<))
import Data.Word (Word8)

escapeBSNulls :: B.ByteString -> BL.ByteString
escapeBSNulls = B.toLazyByteString . BP.primMapByteStringBounded conv
  where
    conv :: BP.BoundedPrim Word8
    conv = BP.condB (==0) (BP.liftFixedToBounded replacement)
                          (BP.liftFixedToBounded BP.word8)
    replacement :: BP.FixedPrim a
    replacement = const ('\\', ('0', ('0', '0')))
              >$< BP.char8 >*< BP.char8 >*< BP.char8 >*< BP.char8

In my benchmark on a 13M data file criterion says it is about 20x faster than the low-level version from this pull req.

[dcoutts]

Note that the large factor is likely an effect of using test data with a lot of \0s (an executable file) not from using such a large input (it's actually an even bigger factor for a 10k prefix of the same binary test file).