This is an automated comment for commit 3c0ccb0 with description of existing statuses. It's updated for the latest CI running

❌ Click here to open a full report in a separate page

Interesting - I needed this feature just yesterday!

One thought - maybe we should make the tuple function to construct a named tuple instead of introducing another function?

But this is only if named tuples support both the .N and .name access.

But this is only if named tuples support both the .N and .name access.

Yeah. It's reasonable to have tuple return named tuples only if we have full-fledged support. Once it's done, we can alias it to namedTuple.

It works:

milovidov-desktop :) SET allow_experimental_analyzer = 1

SET allow_experimental_analyzer = 1

Query id: 8eded140-2f65-4407-a926-93f869739640

Ok.

0 rows in set. Elapsed: 0.000 sec. 

milovidov-desktop :) SELECT CAST(('Hello', 'world') AS Tuple(x String, y String)) AS t, t.1, t.2, t.x, t.y

SELECT
    CAST(('Hello', 'world'), 'Tuple(x String, y String)') AS t,
    t.1,
    t.2,
    t.x,
    t.y

Query id: a26093cb-5b5c-4839-9212-b4f203f68437

┌─t─────────────────┬─tupleElement(t, 1)─┬─tupleElement(t, 2)─┬─t.x───┬─t.y───┐
│ ('Hello','world') │ Hello              │ world              │ Hello │ world │
└───────────────────┴────────────────────┴────────────────────┴───────┴───────┘

1 row in set. Elapsed: 0.011 sec.

Let's anticipate the analyzer enablement, and avoid adding a separate namedTuple function.

Casting to named tuples has been actually working without new analyzer for a long time, there is a test for that: https://github.com/ClickHouse/ClickHouse/blob/master/tests/queries/0_stateless/00547_named_tuples.sql.

But it is really inconvenient to hardcode datatypes during casting. So namedTuple function or new tuple constuctor will be really useful.

One thought - maybe we should make the tuple function to construct a named tuple instead of introducing another function?

But this is only if named tuples support both the .N and .name access.

I'd prefer this way of implementation.

I'd prefer this way of implementation.

It breaks INSERT SELECT query when trying to insert into predefined named tuples from constructed named tuple with different tuple column names. See 01533_multiple_nested, 02475_bson_each_row_format, etc.

It breaks geometry type construction when using named tuple. See 01306_polygons_intersection.

It breaks JOIN analysis because we use tuple types to compare left and right columns. See 02861_join_on_nullsafe_compare, 02911_join_on_nullsafe_optimization, etc.

These are found by fast tests. There might be other issues.

@amosbird We can make it under a setting, enable it by default, and put it under the compatibility setting.

We can make it under a setting, enable it by default, and put it under the compatibility setting.

Sure. I believe I've covered everything. Let's see if isUnquotedIdentifier can help resolve those CI issues.

contrib/orc submodule is changed (fixed) intentionally.

Stateless tests (release, old analyzer, s3, DatabaseReplicated) [2/4]

03036_dynamic_read_subcolumns #65319

Also need to sync with private repo.

Other than these two issues, the CI is green.

@vitlibar Could you please help take another look?

Stateless tests (ubsan) [2/2] — fail: 1, passed: 3322, skipped: 30

01158_zookeeper_log_long #64790

Ok. It is almost ready to merge, but why the ORC submodule was updated? Let's reset it.

Seems like it breaks compatibility for case tuple(NULL):
#66636

Before:

:) select toTypeName(tuple(NULL))

┌─toTypeName(tuple(NULL))──┐
│ Tuple(Nullable(Nothing)) │
└──────────────────────────┘

Now:

:) select toTypeName(tuple(NULL))

┌─toTypeName((NULL))──────────────┐
│ Tuple(`NULL` Nullable(Nothing)) │
└─────────────────────────────────┘

Yes. The isUnquotedIdentifier function needs enhancement to exclude NULL. I suppose that's the only exception.

Can you create a fix for it?

Sure. Will do it now. And we also need to handle true/false.

Hello.
Thank you very much for this great addition. But could you please say from which version of ClickHouse it is present?

@Hubbitus it is in 24.7

Hi @amosbird @vitlibar — while reviewing this PR I found the following:

• Bug (P0): tuple with enable_named_columns_in_function_tuple creates named tuples that silently lose data on CAST/INSERT to differently-named tuple types — tuple with enable_named_columns_in_function_tuple creates named tuples that silently lose data on CAST/INSERT to differently-named tuple types #103527

Happy to discuss — close anything that's wrong or already addressed.