Initial version of the second item done: https://github.com/MariaDB/server/commit/84675ed91f4, where for consistency we

1. truncate not only decimals, but also ulonglong, -ulonglong, LONGLONG_MAX and LONGLONG_MIN
2. truncate not only maxvalue but also minvalue (obviously)

Now working on the first item.

Initial version of the first item, part 1: https://github.com/MariaDB/server/commit/8a96330b07a. It supports signed datatypes for sequence values (unsigned datatypes will be in a separate patch). A pending task for this patch is to choose a proper representation of the the value datatypes in the sequence table. Currently I am lazily using the number representation of the C++ enum, but need to find out whether there is a builtin sql enum already in the server code, or we have to use something else, like string (Type_handler::name()).

I could not find any built-in sql enums corresponding to the c++ enum_field_types, so I updated my patch to use string instead.

However, the change from tinyint to string fails some tests. AFAIK, there are two types of failure:

• "ER_SEQUENCE_INVALID_TABLE_STRUCTURE (4086): Sequence 'test.t1' table structure is invalid (value_type)", which is caused by the varchar field defaulting to my_charset_bin, which adds a binary flag to the field.
• aria table removing the row in the sequence when cycling back (sequence.next testcase):

  CREATE or REPLACE SEQUENCE t1 minvalue 1 maxvalue 2 increment by -1 cache 2 cycle engine=aria;

  select next value for t1;

  select * from t1;

  CREATE or REPLACE SEQUENCE t1 minvalue 1 maxvalue 2 increment by -1 cache 2 cycle engine=aria;

  select next value for t1;

  select * from t1;

  The first select * from t1 (with myisam) returns the correct result:

  next_not_cached_value minimum_value maximum_value start_value

  increment cache_size cycle_option cycle_count value_type

  0 1 2 2 -1 2 1 0 bigint

  But the second one (with aria) returns empty:

  next_not_cached_value minimum_value maximum_value start_value

  increment cache_size cycle_option cycle_count value_type

I have therefore kept this change in a separate commit (https://github.com/MariaDB/server/commit/f9cb7cbb7dc).

I will continue debugging these problems. Meanwhile, sanja could you take a look at the three patches and let me know what you think (e.g. whether you agree with the general approach)? After that I will continue with adding support for unsigned value types and the information schema tables.

• https://github.com/MariaDB/server/commit/84675ed91f4
• https://github.com/MariaDB/server/commit/8a96330b07a (the commit message mentions changing the tinyint to something else, which I did in the next commit below, tests pass)
• https://github.com/MariaDB/server/commit/f9cb7cbb7dc (some tests do not pass as mentioned above, may need to add more tests for the new feature)

One concern with adding a column for value types is migration: how do we handle schema changes caused by server upgrade?

After discussions with sanja, I have revised the commits:

• https://github.com/MariaDB/server/commit/84675ed91f4 (item 2)
• https://github.com/MariaDB/server/commit/51815a7d9ff (item 1)

Following up on discussions, the parsing is fixed - new non-terminals have been added and there's no longer newly introduced conflicts. sanja igor.

Current progress: Truncation of maxvalue/minvalue (item 2) done. Introducing types (item 1) still wip, mainly pending implementing ALTER SEQUENCE ... AS <new_type> (commit fd5a6414f43b2a3 below), but also need to fix a couple tests (see commit msg of 79d4283490cee) and maybe add more tests. information_schema.sequences (item 3) not started.

Commits in order (will squash when no longer wip):

• https://github.com/MariaDB/server/commit/f69ea459ca66aff8228a1b08c11c3ea78b3c0758
• https://github.com/MariaDB/server/commit/86de48ae1f6f03527eba65ebf35293bf906cf5b0
• https://github.com/MariaDB/server/commit/79d4283490cee04db6bf34e2b9c3a3bbf605ea96
• https://github.com/MariaDB/server/commit/fd5a6414f43b2a33493fc44762a05a0d46e607df

As an update, Item 3 is implemented, and all default and sql_sequence tests pass except compat/oracle.sequence which probably requires backward compatibility with oracle that does not support the new AS <TYPE> syntax[1]. The remaining non-trivial issue and blocker is the implementation of ALTER SEQUENCE ... AS <NEW_TYPE>.

Commits in order (the "alter as new_type" wip commit is in a different branch [2] and has not changed much since the last update):

• https://github.com/MariaDB/server/commit/f69ea459ca66aff8228a1b08c11c3ea78b3c0758 (truncating max/minvalue)
• https://github.com/MariaDB/server/commit/86de48ae1f6f03527eba65ebf35293bf906cf5b0 (signed types)
• https://github.com/MariaDB/server/commit/3758c111e89685f0fc87ea01c234d2cffd84977b (unsigned types)
• https://github.com/MariaDB/server/commit/3594bf4e1c813c1dff6449995b4eae7c244afcad (information_schema)

[1] https://docs.oracle.com/en/database/oracle/oracle-database/21/sqlrf/CREATE-SEQUENCE.html
[2] https://github.com/MariaDB/server/compare/bb-11.0-mdev-28152-alter-value-type

As discussed with serg and sanja, we implement an incomplete alter feature so that it is more likely to be done during this development cycle. The missing feature is: in ALTER SEQUENCE, AS <new_type> cannot be combined with any other alter options. That is, we can do alter sequence s as int, or alter sequence s maxvalue 100 minvalue 20, but not alter sequence s as int maxvalue 100 minvalue 20.

Patches in order:

• https://github.com/MariaDB/server/commit/b779f00af30858a5c41378c0e72381cd3d5875a4 (truncating maxvalue/minvalue)
• https://github.com/MariaDB/server/commit/cbeb2bbfe8827d7b9a8790b671fbd94fb3ef92d2 (supporting AS <type>)
• https://github.com/MariaDB/server/commit/b0caadedace1b2056454c3e17f02270dea0eca3a (information schema table)

Passing all tests / CI is green tick. sanja can you take a look or assign to someone else if you don't have time?

I realised I need to update the implementation of setval() too for unsigned types. Working on it.

Meanwhile I have addressed all the comments in the first round review at https://lists.launchpad.net/maria-developers/msg13297.html

New patch:
https://github.com/MariaDB/server/commit/eef42da6b59

Diff with the previous (squashed) patch:
https://github.com/MariaDB/server/commit/8c2738b21da

sanja In case you haven't had a chance to look at the patches in the previous comment yet, here is the updated patch with setval() and auto_increment() updated for unsigned types:

https://github.com/MariaDB/server/commit/30425c431c5

The diff with the previous patch that you have reviewed:

https://github.com/MariaDB/server/commit/bd4a2fac119

Thank you looks OK now

rebased to 11.4:

9548c168fb26083852118402bf6d809a77ae57b8 upstream/bb-11.4-mdev-28152 MDEV-28152 Features for sequences

Locally tests pass on the sql_sequence suite.

waiting for CI...

Hi lstartseva, ptal at the following commit, thanks

445ce402b01 upstream/bb-11.4-mdev-28152 MDEV-28152 Features for sequences

Testing is completed but MDEV-33012 needs to be fixed before pushing

Hi lstartseva, could you test the patch on top of the MDEV-33169 patch that fixes MDEV-33012?

4d2224702fc upstream/bb-11.4-mdev-28152 MDEV-28152 Features for sequences

09a710036b6 MDEV-33169 Reset sequence used fields after check in alter sequence

Really, it looks like patch for MDEV-33169 fixes problems MDEV-33012. Tests passed.
Feature is ok to push with this changes.

Thanks lstartseva. The fixversion was changed to 11.5 and the feature was not included in 11.4-preview, so we have to wait for an 11.5 branch to be created first.

11.5 branch has been created with the MDEV-33169 patch included. I have verified that MDEV-33012 is fixed by it. So I have applied the patch on top as 374783c3d9a87170b313a05240dadaa813ec6034 and pushed it to 11.5.