`parseJSON` does not support `null` in JSON

[tvdstorm]

Reproduce:

rascal>import lang::json::IO;
ok
rascal>parseJSON(#map[str,value], "{\"bla\": null}");
|std:///lang/json/IO.rsc|(2599,303,<49,0>,<52,144>): IO("NPE")
        at *** somewhere ***(|std:///lang/json/IO.rsc|(2599,303,<49,0>,<52,144>))
        at parseJSON(|std:///lang/json/IO.rsc|(2895,5,<52,137>,<52,142>)ok

As Rascal does not support null, we need a sentinel value, but this cannot be chosen by the implementation. Suggestion: a keyword parameter specifying a custom value to use in case of null.

[jurgenvinju]

Just to be sure: the parser supports null currently only when the current field is the name of a keyword field of the current data-type.

How about we just return nothing() if no specific type is expected? And we could add support for null and non-null in case we expect Maybe[.]

[jurgenvinju]

rascal>parseJSON(#map[str,value], "{\"bla\": null}")
map[str, value]: ("bla":"null"())
rascal>parseJSON(#map[str,value], "{\"bla\": null}", nulls=(#value:-1))
map[str, value]: ("bla":-1)
rascal>parseJSON(#map[str,Maybe[str]], "{\"bla\": null}")
map[str, Maybe[str]]: ("bla":nothing())
rascal>parseJSON(#map[str,Maybe[str]], "{\"bla\": \"foo\"}")
map[str, Maybe[str]]: ("bla":just("foo"))

[jurgenvinju]

rascal>data Cons = cons(str bla = "null");
ok
rascal>parseJSON(#Cons, "{\"bla\": \"foo\"}")
Cons: cons(bla="foo")
rascal>parseJSON(#Cons, "{\"bla\": null}")
Cons: cons()
rascal>parseJSON(#Cons, "{\"bla\": null}").bla
str: "null"

[jurgenvinju]

@tvdstorm Don't know what to do when writing these values back. Should the null values be written as null again? Or just as they are? Support we have (#int:-1) while reading integer nulls. Should -1 be written as null again when we use the same nulls map?

[jurgenvinju]

Also improved error reporting while we are at it. All internal parse errors and malformedness in the Gson parser is lifted to a Rascal ParseError with the JSON path query to the position, and offset, lengths, line/col, and the diagnostic.