[ARROW-11495] [RUST] [datafusion] implementation for numerical_coercion is wrong

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Component/s: Rust, Rust - DataFusion
• Labels: pull-request-available
• External issue URL: https://github.com/apache/arrow/issues/18495
• Assignee: Unassigned
• Reporter: Sundy Li
• Created: 04/Feb/21 09:58
• Updated: 11/Jan/23 08:20

Description

Currently `numerical_coercion` did not take consideration for integer overflow, such as 

 
 

//代码占位符
UIn8 + UIn8 should be UInt16, but now it's UInt8
UInt16 * UInt16 should be UInt64, but now it's UInt16

ClickHouse handled it well, maybe we should learn it from ClickHouse

 

//代码占位符 From  https://github.com/ClickHouse/ClickHouse/blob/bd81f43ecb/src/DataTypes/NumberTraits.h

template <bool is_signed, bool is_floating, size_t size>
 struct Construct
{ usingType=Error; }
;
template <> struct Construct<false, false, 1> { using Type = UInt8; };
 template <> struct Construct<false, false, 2> { using Type = UInt16; };
 template <> struct Construct<false, false, 4> { using Type = UInt32; };
 template <> struct Construct<false, false, 8> { using Type = UInt64; };
 template <> struct Construct<false, false, 16> { using Type = UInt256; }; /// TODO: we cannot use our UInt128 here
 template <> struct Construct<false, false, 32> { using Type = UInt256; };
 template <> struct Construct<false, true, 1> { using Type = Float32; };
 template <> struct Construct<false, true, 2> { using Type = Float32; };
 template <> struct Construct<false, true, 4> { using Type = Float32; };
 template <> struct Construct<false, true, 8> { using Type = Float64; };
 template <> struct Construct<true, false, 1> { using Type = Int8; };
 template <> struct Construct<true, false, 2> { using Type = Int16; };
 template <> struct Construct<true, false, 4> { using Type = Int32; };
 template <> struct Construct<true, false, 8> { using Type = Int64; };
 template <> struct Construct<true, false, 16> { using Type = Int128; };
 template <> struct Construct<true, false, 32> { using Type = Int256; };
 template <> struct Construct<true, true, 1> { using Type = Float32; };
 template <> struct Construct<true, true, 2> { using Type = Float32; };
 template <> struct Construct<true, true, 4> { using Type = Float32; };
 template <> struct Construct<true, true, 8> { using Type = Float64; };

Issue Links

• links to: GitHub Pull Request #9596