Math operations for int values fail for big values

[harshil-goel]

What version of Dgraph are you using?

v1.1.0-134-gd9767791

Have you tried reproducing the issue with the latest release?

Yes

What is the hardware spec (RAM, OS)?

32gb ram, ubuntu 19.04

Steps to reproduce the issue (command/config used to run Dgraph).

{
  set {
    _:user1 <money> "48038396025285290" .
  }
}

{
  q(func: has(money)) {
    f as money
    g: math(2+f)
  }
}

Expected behaviour and actual result.

Expected Behaviour

"q": [
      {
        "money": 48038396025285290,
        "g": 48038396025285292
      }
    ]

Actual Result:

"q": [
      {
        "money": 48038396025285290,
        "g": 48038396025285290
      }
    ]

This happens because all the math operations happen in float64 (https://github.com/dgraph-io/dgraph/blob/master/query/aggregator.go#L124), which only has a precision of 53 bits. So int, being 64 bits can overflow when put into float64 and hence the error.

[harshil-goel]

We are going to change the current implementation in which we typecast int to float, to an implementation in which we keep the highest information type among the corresponding arguments.
Potentially it would allow for queries described in the issue, and if one wants a float result of their queries, the user can ask for math(f + 2.0).