Default Nullable for type or not?

[wangzhen11aaa]

Context

I am doing the task that supports the nullable group-by, running into some bewilderments. Our database makes the type nullable default, quote
create table t(a UInt64, b UInt32) Engine = Fuse; insert into t(a,b) select if (number % 3 = 1, null, number) as a, number + 3 as b from numbers(10); from #4079
When we do the following query search like select a from t group by a,we will use 8 bytes + 1 byte size to store this hash_key. I mean from this simple query, we can infer that all the types will be nullable and cost 1 byte more in the hash key when used as a group by key. Actually, we may not have any non-nullable type any more, unless that we have the type to explictly declare it as nonnullable(type), we lost some flexibility.

The other database design (mainly clickhosue):

1. We know that the nullable is explictly declared in Clickhouse, It will makes it more economy in space while lose some flexibility.

For me

I just think that we lost some basic non-nullable value type may be not symmetric and it will drag all the hash_methods into the nullable desert. (serielize and deserilize with all the following bitmap).

Quote from sundyli

doris，mysql，snowflake : nullable default,
clickhouse, firebolt. non-nullable default.

So I launched this disscussion to get more thought into it. Thank you.

[sundy-li]

For now on:

SqlType eg: c Int8 is mapping to datavalue's Nullable(Int8Type), c Int8 not null is mapping to Int8Type.

This may be a little confusing.

So I would vote for defaults to be non-nullable data_type, which is the style of ClickHouse & firebolt, then SqlType is compatible with datavalue's datatype.

What's your opinion? cc @zhang2014 @BohuTANG @Xuanwo

[Xuanwo]

I prefer non-nullable data_type personally (without technical consideration)

[BohuTANG]

I think this is a user experience problem, not a technical problem.
For most user, they don't care what the NULL is, non-nullable default data type will lower the user's barrier to use.

[sundy-li]

#4435 Defaults to not null now.

[ZeaLoVe]

I prefer nullable default. :)

[sundy-li]

Default nullable could be more flexible when we migrate other DMBS tests.

Should we add a setting datatype_default_nullable to control the behavior? cc @leiysky @BohuTANG @zhang2014 @Xuanwo

[Xuanwo]

• Migrating other DMBS tests' goal is to improve the quality of databend.
• Adding new settings like datatype_default_nullable makes our logic and behavior more complex, maybe worse for our code quality.

From my first look, it's not a good tradeoff. I prefer to postpone this feature until we have strong requests from our real users.

[leiysky]

Agree with @Xuanwo