hw-json-0.0.0.1: README.md
# hw-json
[](https://circleci.com/gh/haskell-works/hw-json/tree/v0.0-branch)
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`](../master/app/Main.hs)
## Prerequisites
* Install `haskell-stack`.
* Install `hlint` (eg. `stack install hlint`)
## Building
Run the following in the shell:
git clone git@github.com: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](https://www.youtube.com/watch?v=uA0Z7_4J7u8)
* [Typed Tagless Final Interpreters](http://okmij.org/ftp/tagless-final/course/lecture.pdf)
* [Conduit Overview](https://www.schoolofhaskell.com/school/to-infinity-and-beyond/pick-of-the-week/conduit-overview)
## Special mentions
* [Sydney Paper Club](http://www.meetup.com/Sydney-Paper-Club/)