Panic: interface conversion: *hashing.BinaryMemoTable is not hashing.TypedMemoTable[uint16] when using IsIn with FixedSizeBinary(2)

[Patzifist]

Describe the bug, including details regarding any error messages, version, and platform.

The issue seems to be caused by an optimization that treats 2-byte fixed binary data as uint16 for hashing, but fails to provide the correct MemoTable implementation, leading to an interface conversion error.

Request from tpc-ds sf1:
SELECT count(*)
FROM customer_address
WHERE ca_state IN ('MS', 'IN', 'ND', 'OK', 'NM', 'VA');

Actual:
panic: interface conversion: *hashing.BinaryMemoTable is not hashing.TypedMemoTable[uint16]: missing method Exists

goroutine 243 [running]:
://github.com.(*SetLookupState[...]).Init(0x1f76000, {{0x1f7e180, 0x84a9f29e070}, 0x6, {0x84aa34e4340, 0x1, 0x1}, 0x0})
/vendor/://github.com/scalar_set_lookup.go:176 +0x18f
://github.com.CreateSetLookupState({{0x1f7e180, 0x84a9f29e070}, 0x6, {0x84aa34e4340, 0x1, 0x1}, 0x0}, {0x1f6ab50, 0x84a93f74c08})
/vendor/://github.com/scalar_set_lookup.go:144 +0x553
://github.com(0x84a94059cb0, {{0x1f649e0, 0x84a94050a88}, {0x84a943d0360, 0x1, 0x1}, {0x1b9ce40, 0x84a9f31e168}})
/vendor/://github.com +0xe6e
://github.com({0x1f726d8, 0x84a94193d40}, {{0x84aa34e4270?, 0x84a943d00c0?, 0x84aa1bff628?}, 0x48daa5?}, {0x1f8e538, 0x84a946e69b0}, {0x1f83be8, 0x84a94193d10})
/vendor/://github.com +0x1806

Steps to Reproduce:

package main

import (
	"context"
	"fmt"
	"testing"

	"github.com/apache/arrow-go/v18/arrow"
	"github.com/apache/arrow-go/v18/arrow/array"
	"github.com/apache/arrow-go/v18/arrow/compute"
	"github.com/apache/arrow-go/v18/arrow/memory"
)

func TestReproduceArrowPanicWithSetOptions(t *testing.T) {
	mem := memory.NewGoAllocator()
	ctx := context.Background()

	catType := &arrow.FixedSizeBinaryType{ByteWidth: 2}

	numRows := 100
	b := array.NewFixedSizeBinaryBuilder(mem, catType)
	for i := 0; i < numRows; i++ {
		b.Append([]byte("MS"))
	}
	dataArr := b.NewArray()
	defer dataArr.Release()

	vBuilder := array.NewFixedSizeBinaryBuilder(mem, catType)
	vBuilder.Append([]byte("MS"))
	vBuilder.Append([]byte("VA"))
	valueSet := vBuilder.NewArray()
	defer valueSet.Release()

	opts := compute.SetOptions{
		ValueSet:     compute.NewDatumWithoutOwning(valueSet),
		NullBehavior: compute.NullMatchingMatch,
	}

	input := compute.NewDatumWithoutOwning(dataArr)

	defer func() {
		if r := recover(); r != nil {
			fmt.Printf("Caught expected panic: %v\n", r)
		}
	}()

	_, err := compute.IsIn(ctx, opts, input)
	if err != nil {
		fmt.Printf("Got error instead of panic: %v\n", err)
	}
}

Component(s)

Parquet

[SAY-5]

This looks like the same root cause as #785, which was fixed in #797 (merged 2026-05-03). The current main no longer reproduces the panic for FSB widths 1/2/4/8 — TestIsInFixedSizeBinaryFastPaths covers the width=2 case from this report. Could you confirm on a fresh checkout? If still seen, happy to dig further.

[Patzifist]

Yes, confirmed. I ran the tests on the latest main branch, and the panic no longer reproduces for the specified widths. Thanks for pointing that out!