Notes about CI: The unrelated clang failure is fixed in master already (a21e4b3) and the failed stateless test (00910_zookeeper_test_alter_compression_codecs_long) has nothing to do with the changes.

@Mergifyio update

Command update: success

Branch has been successfully updated

Hi, does this need any change or improvement? The CI errors seem to be completely unrelated with the changes.

@Algunenano CI checks looks almost alright. It is waiting for review.

Ah ok, thanks!

@Algunenano @nikitamikhaylov What was the main reason for the bug?
Why it's not possible to calculate the number of distinct aggregate function states?

@Algunenano @nikitamikhaylov What was the main reason for the bug?

I couldn't find where the bug is. I expect the bug (or bugs) to be found in a completely different place of where the crashes are happening and after a couple of days of looking into it I couldn't find it and forbit it instead.

I don't know how many people need to actually calculate the distinct states of distintct states and it's more likely done by mistake rather than an actual need (at least it was in our case). But that's a different topic, if CH wanted to support it then the issues would need to be found and addressed.

how many people need to actually calculate the distinct states of distintct states and it's more likely done by mistake rather than an actual need

Yes, I also expect that :)

I just in doubt that the fix is just covering the culprit instead of making it clear.
I will try to investigate.

The fix is valuable nevertheless, thank you!
The possibility of crash on user provided query is just unbearable and is the №0 priority.

I will try to investigate.

Please let me know if I can help in any way. I'm trying to (slowly) learn the internals of ClickHouse.

The real reason is that the type is wrong:

SELECT toTypeName(uniqExactState(x))
FROM
(
    SELECT quantileState(number) AS x
    FROM numbers(1000)
)

Query id: 3a8dd528-9b14-4506-a5a8-1ab2241ea2b2

┌─toTypeName(uniqExactState(x))───────┐
│ AggregateFunction(quantile, UInt64) │
└─────────────────────────────────────┘

Has to be something like

AggregateFunction(uniqExact, AggregateFunction(quantile, UInt64))

I will check how it is possible.

Another example:

SELECT toTypeName(initializeAggregation('anyState', initializeAggregation('uniqState', 0)))

The bug was introduced in #7109.

This PR does not fix the issue, here is another reproducer:

SELECT anyState(x) FROM (SELECT uniqState(number) AS x FROM numbers(1000))