remove or reconsider the `NumbersTable`

[waynexia]

What type of enhancement is this?

Tech debt reduction

What does the enhancement do?

we have a built-in system table NumbersTable for testing. But it's very bad implemented and brings tons of special logic.

Implementation challenges

No response

[tisonkun]

very bad implemented and brings tons of special logic

What kind of real issues you specify here? From my point of view, except:

} else if schema == DEFAULT_SCHEMA_NAME && table_name == NUMBERS_TABLE_NAME {
            Some(NumbersTable::table(NUMBERS_TABLE_ID))

This table impl is barely an internal testing table for convenient testing.

If we don't want it pollute the main logic but only whitebox table tests, we can exclude its usage in the main process (planning/name resolution) and annotate it with #[cfg(test)]. What do you think?

cc @waynexia

[waynexia]

From my point of view, except:

That's one.

It also breaks a system-level invariant that every table should contain one TIME INDEX column.

If we don't want it pollute the main logic but only whitebox table tests, we can exclude its usage in the main process (planning/name resolution) and annotate it with #[cfg(test)]. What do you think?

Blackbox tests use this table more often. So #[cfg(test)] might not work. But for most cases in blackbox and whitebox tests we can use other ways to workaround NumbersTable. Like creating a temporary real table.

[tisonkun]

It also breaks a system-level invariant that every table should contain one TIME INDEX column.

@waynexia How do we implement it or how can we check this invariant?

I found MemTable::default_numbers_table that may replace the NumbersTable but it doesn't seem to have TIME INDEX also.

Besides, it is expected to remove:

} else if schema == DEFAULT_SCHEMA_NAME && table_name == NUMBERS_TABLE_NAME {
            Some(NumbersTable::table(NUMBERS_TABLE_ID))

? I'm unsure if SELECT * FROM numbers should work or we can drop it.