[AVRO-2260] IDL Json Parsing is lossy, and it could be made more accurate.

• Type: Bug
• Status: Open
• Priority: Minor
• Resolution: Unresolved
• Component/s: java
• Assignee: Unassigned
• Reporter: Zoltan Farkas
• Created: 11/Nov/18 13:30
• Updated: 30/Dec/18 04:36

Description

Currently all integers are handled as Long, and all floating point as Double, having basically the following issues:

1) cannot handle numbers larger that MAXLONG.
2) introducing unnecessary precision

JsonNode Json() :
{ String s; Token t; JsonNode n; }
{ 
( s = JsonString() { n = new TextNode(s); }
| (t=<INTEGER_LITERAL> { n = new LongNode(Long.parseLong(t.image)); })
| (t=<FLOATING_POINT_LITERAL> {n=new DoubleNode(Double.parseDouble(t.image));})
| n=JsonObject()
| n=JsonArray()
| ( "true" { n = BooleanNode.TRUE; } )
| ( "false" { n = BooleanNode.FALSE; } )
| ( "null" { n = NullNode.instance; } )
 )
  { return n; }
}

This should be improved to:

JsonNode Json() :
{ String s; Token t; JsonNode n; }
{ 
( s = JsonString() { n = new TextNode(s); }
| (t=<INTEGER_LITERAL> {
           try {
             n = new IntNode(Integer.parseInt(t.image));
           } catch(NumberFormatException  e) {
             try {
                n = new LongNode(Long.parseLong(t.image));
             } catch(NumberFormatException  ex2) {
                n = new BigIntegerNode(new java.math.BigInteger(t.image));
             }
           }
 })
| (t=<FLOATING_POINT_LITERAL> {n=new DecimalNode(new java.math.BigDecimal(t.image));})
| n=JsonObject()
| n=JsonArray()
| ( "true" { n = BooleanNode.TRUE; } )
| ( "false" { n = BooleanNode.FALSE; } )
| ( "null" { n = NullNode.instance; } )
 )
  { return n; }
}