Description

Indexing NUMERIC fields has unchanged behavior.

• A NUMERIC field with value greater than i64::min (-2^63), but less than i64::max (2^63 - 1) is indexed as i64. It does not lose precision. Its type is integer.
• A NUMERIC field with value greater than i64::max (2^63 - 1), but less than u64::max (2^64 - 1) is indexed as u64. It does not lose precision. Its type is integer.
• A NUMERIC field with value greater than u64::max (2^64 - 1), or lower than i64::min (-2^63), or containing a decimal point, is indexed as f64. Its type is number. This may implicate losing precision, but this is not a breaking change.

NUMINCRBY behavior changed to prevent crash.

• i64 + i64 whose result is between i64::min and i64::max has unchanged behavior. Its result is stored as i64, its type is integer.
• i64 + i64 whose result is greater than i64::max or less than i64::min explicitly wraps instead of invoking undefined behavior. Its result is stored as i64, its type is integer.
• i64 + u64 and u64 + u64 behavior is unchanged. Its result is stored as f64, its type is number. This may implicate losing precision, but this is not a breaking change.
• u64 + i64 used to crash. Now it behaves consistently with the prior case.
• If either argument is f64, the result is stored as f64, with type number.

DROP of integers stored as u64 no longer crashes.

This PR only affects NUMERIC fields using vanilla JSON, not those using HDT.

Which issues this PR fixes

1. MOD-6501
2. MOD-4551
3. MOD-4856
4. MOD-5714
5. #875
6. #3726 (Search)

Main objects this PR modified

1. IValue Manager
2. JSON Node