JSON parser with nice error messages and little more strict syntax (whitespace-wise). Based on Aeson and Parsec.
JSON being nice readable text-based format seems good candidate for occasionally being created by a user. While Aeson provides really super-optimized parsers, their error messages are not very helpful. Creating larger JSON object by hand can be frustrating (especially) when you make even a small mistake.
While this parser is not optimized for speed, it tries to produce nice and helpful error messages. (This library uses Parsec library.)
Another way to help your user is not allowing him or her to learn wrong habbits. Just look at the following piece of code (be warned - there are trailing spaces there):
{ "name" :
,
"Hal"
}
That (in my opinion) is something one would not like to see in files users of his or hers tool produces. So why not forbid that? This library does not allow such things while still allowing to make the input more airy.
This library was written with re-usability in mind. Parsers it provides do not consume any spaces before of after corresponding values and therefore are more easily reusable for your own projects.
Aeson library is nice to work with with large ecosystem of useful libraries. So why not join them and avoid reinventing the wheel?
{-# LANGUAGE DeriveGeneric #-}
module Main (main) where
import GHC.Generics
import System.Environment (getArgs)
import Data.Aeson hiding (eitherDecode)
import Data.Aeson.Parser.Parsec.Picky (eitherDecode)
import Data.Text.IO as Text (readFile)
data Contact = Contact
{ name :: String
, address :: String
} deriving (Generic, Show)
instance FromJSON Contact
printContacts :: [Contact] -> IO ()
printContacts = mapM_ print
main' :: [String] -> IO ()
main' [filename] = Text.readFile filename
>>= process . eitherDecode filename
where
process = either putStrLn printContacts
main' _ = print "Usage: script [CONTACTS_FILE]"
main :: IO ()
main = getArgs >>= main'
Some examples of possible valid formatting.
Compact.
[{"name":"Alice","address":"Kansas"}]
Haskell-ish style.
[ { "name": "Alice"
, "address": "Kansas"
}
]
More airy version of previous one.
[ { "name": "Alice"
, "address": "Kansas"
}
]
Handing style.
[ { "name": "Alice",
"address": "Kansas"}]
[ { name: "Alice",
"address": "Kansas"}]
"example.json" (line 1, column 5):
unexpected "n"
expecting " ", JSON object key (string), lf new-line or "}"
[ { "name"= "Alice",
"address": "Kansas"}]
"example.json" (line 1, column 11):
unexpected "="
expecting ":"
[ { "name" : "Alice",
"address": "Kansas"}]
"example.json" (line 1, column 11):
unexpected " "
expecting ":"
[ { "name": "Alice" ,
"address": "Kansas"}]
"example.json" (line 1, column 25):
unexpected "\n"
expecting " " or JSON object key (string)
In following example spaces were replaced with plus signs.
[+{+"name":+"Alice",++++
++++"address":+"Kansas"}]
"example.json" (line 1, column 25):
unexpected "\n"
expecting " " or JSON object key (string)
Why another JSON parser? Some internal tool for JSON RPC testing used simple format that re-used JSON parsers. It was already re-written few times and reasons were:
- Bad error messages for people who were writing testing scripts.
- Those people were able to do horrible stuff (trailing spaces, ...).
- Some parsers that used Parsec (and produced helpful error messages) were producing non-aeson data structures and we already use aeson on some places so we had option to be more heterogeneous or make useless conversions.
No parser I was aware of seemed to solve these issues.