hw-json

Conduits for tokenizing streams.

hw-json is a succinct JSON parsing library. It uses succinct data-structures to allow traversal of large JSON strings with minimal memory overhead.

It is currently considered experimental.

For an example, see app/Main.hs

Prerequisites

• Install haskell-stack.
• Install hlint (eg. stack install hlint)

Building

Run the following in the shell:

git clone [email protected]:haskell-works/hw-json.git
cd hw-json
stack setup
stack build
stack test
stack ghci --ghc-options -XOverloadedStrings \
  --main-is hw-json:exe:hw-json-example

Memory benchmark

Parsing large Json files in Scala with Argonaut

      S0U       EU           OU       MU     CCSU CMD
--------- --------- ----------- -------- -------- ---------------------------------------------------------------
      0.0  80,526.3    76,163.6 72,338.6 13,058.6 sbt console
      0.0 536,660.4    76,163.6 72,338.6 13,058.6 import java.io._, argonaut._, Argonaut._
      0.0 552,389.1    76,163.6 72,338.6 13,058.6 val file = new File("/Users/jky/Downloads/78mbs.json"
      0.0 634,066.5    76,163.6 72,338.6 13,058.6 val array = new Array[Byte](file.length.asInstanceOf[Int])
      0.0 644,552.3    76,163.6 72,338.6 13,058.6 val is = new FileInputStream("/Users/jky/Downloads/78mbs.json")
      0.0 655,038.1    76,163.6 72,338.6 13,058.6 is.read(array)
294,976.0 160,159.7 1,100,365.0 79,310.8 13,748.1 val json = new String(array)
285,182.9 146,392.6 1,956,264.5 82,679.8 14,099.6 val data = Parse.parse(json)
                    ***********

Parsing large Json files in Haskell with Aeson

Mem (MB) CMD
-------- ---------------------------------------------------------
     302 import Data.Aeson
     302 import qualified  Data.ByteString.Lazy as BSL
     302 json78m <- BSL.readFile "/Users/jky/Downloads/78mbs.json"
    1400 let !x = decode json78m :: Maybe Value

Parsing large Json files in Haskell with hw-json

Mem (MB) CMD
-------- ---------------------------------------------------------
     274 import Foreign
     274 import qualified Data.Vector.Storable as DVS
     274 import qualified Data.ByteString as BS
     274 import System.IO.MMap
     274 import Data.Word
     274 (fptr :: ForeignPtr Word8, offset, size) <- mmapFileForeignPtr "/Users/jky/Downloads/78mbs.json" ReadOnly Nothing
     601 cursor <- measure (fromForeignRegion (fptr, offset, size) :: JsonCursor BS.ByteString (BitShown (DVS.Vector Word64)) (SimpleBalancedParens (DVS.Vector Word64)))

Examples

Performance example

import Foreign
import qualified Data.Vector.Storable as DVS
import qualified Data.ByteString as BS
import qualified Data.ByteString.Internal as BSI
import System.IO.MMap
import Data.Word
import System.CPUTime
(fptr :: ForeignPtr Word8, offset, size) <- mmapFileForeignPtr "/Users/jky/Downloads/78mbs.json" ReadOnly Nothing
cursor <- measure (fromForeignRegion (fptr, offset, size) :: JsonCursor BS.ByteString (BitShown (DVS.Vector Word64)) (SimpleBalancedParens (DVS.Vector Word64)))
let !bs = BSI.fromForeignPtr (castForeignPtr fptr) offset size
x <- measure $ jsonBsToInterestBs bs
let !y = runListConduit [bs] (unescape' "")

import Foreign
import qualified Data.Vector.Storable as DVS
import qualified Data.ByteString as BS
import qualified Data.ByteString.Internal as BSI
import System.IO.MMap
import Data.Word
import System.CPUTime
(fptr :: ForeignPtr Word8, offset, size) <- mmapFileForeignPtr "/Users/jky/Downloads/part40.json" ReadOnly Nothing
let !bs = BSI.fromForeignPtr (castForeignPtr fptr) offset size
x <- measure $ BS.concat $ runListConduit [bs] (blankJson =$= blankedJsonToInterestBits)
x <- measure $ jsonBsToInterestBs bs

jsonTokenAt $ J.nextSibling $ J.firstChild $ J.nextSibling $ J.firstChild $ J.firstChild  cursor

Navigation example

$  cabal repl --ghc-option='-package mmap'
λ> :set -XNoMonomorphismRestriction
λ> import qualified Data.ByteString                                            as BS
λ> import           Data.String
λ> import qualified Data.Vector.Storable                                       as DVS
λ> import           Data.Word
λ> import           HaskellWorks.Data.Bits.BitShow
λ> import           HaskellWorks.Data.Bits.BitShown
λ> import           HaskellWorks.Data.FromForeignRegion
λ> import           HaskellWorks.Data.Json.Succinct.Cursor                     as C
λ> import           HaskellWorks.Data.Json.Token
λ> import           HaskellWorks.Data.Succinct.BalancedParens.Internal
λ> import           HaskellWorks.Data.Succinct.BalancedParens.Simple
λ> import           HaskellWorks.Data.Succinct.RankSelect.Binary.Basic.Rank0
λ> import           HaskellWorks.Data.Succinct.RankSelect.Binary.Basic.Rank1
λ> import           HaskellWorks.Data.Succinct.RankSelect.Binary.Basic.Select1
λ> import           HaskellWorks.Data.Succinct.RankSelect.Binary.Poppy512
λ> import qualified HaskellWorks.Data.TreeCursor as TC
λ> import           System.IO.MMap
λ> let fc = TC.firstChild
λ> let ns = TC.nextSibling
λ> let pn = TC.parent
λ> let cd = TC.depth
λ> let ss = TC.subtreeSize
λ> let cursor = "[null, {\"field\": 1}]" :: JsonCursor BS.ByteString (BitShown (DVS.Vector Word64)) (SimpleBalancedParens (DVS.Vector Word64))
λ> cursor

References

• Succinct Data Structures talk by Edward Kmett
• Typed Tagless Final Interpreters
• Conduit Overview

Special mentions

• Sydney Paper Club