maria:preview-10.10-uca14

Last commit made on 2022-06-18
Get this branch:
git clone -b preview-10.10-uca14 https://git.launchpad.net/maria

Branch merges

Branch information

Name:
preview-10.10-uca14
Repository:
lp:maria

Recent commits

76875da... by Alexander Barkov

Switching libmariadb to "Libmariadb changes for MDEV-27009 Add UCA-14.0.0 collations"

b641a93... by Alexander Barkov

MDEV-27266 Improve UCA collation performance for utf8mb3 and utf8mb4

Adding two levels of optimization:

1. For every bytes pair [00..FF][00..FF] which:
  a. consists of two ASCII characters or makes a well-formed two-byte character
  b. whose total weight string fits into 4 weights
     (concatenated weight string in case of two ASCII characters,
     or a single weight string in case of a two-byte character)
  c. whose weight is context independent (i.e. does not depend on contractions
     or previous context pairs)
  store weights in a separate array of MY_UCA_2BYTES_ITEM,
  so during scanner_next() we can scan two bytes at a time.
  Byte pairs that do not match the conditions a-c are marked in this array
  as not applicable for optimization and scanned as before.

2. For every byte pair which is applicable for optimization in #1,
   and which produces only one or two weights, store
   weights in one more array of MY_UCA_WEIGHT2. So in the beginning
   of strnncoll*() we can skip equal prefixes using an even more efficient
   loop. This loop consumes two bytes at a time. The loop scans while the
   two bytes on both sides produce weight strings of equal length
   (i.e. one weight on both sides, or two weight on both sides).
   This allows to compare efficiently:
   - Context independent sequences consisting of two ASCII characters
   - Context independent 2-byte characters
   - Contractions consisting of two ASCII characters, e.g. Czech "ch".
   - Some tricky cases: "ss" vs "SHARP S"
     ("ss" produces two weights, 0xC39F also produces two weights)

b8b9525... by Alexander Barkov

MDEV-27265 Improve contraction performance in UCA collations

Adding a hash table for contractions.

The old code iterated through all items in MY_CONTRACTIONS,
and was much slower, especially for those contractions
in the end of the list.

aa5dbdd... by Alexander Barkov

MDEV-27009 Add UCA-14.0.0 collations

- Added one neutral and 22 tailored (language specific) collations based on
  Unicode Collation Algorithm version 14.0.0.

  Collations were added for Unicode character sets
  utf8mb3, utf8mb4, ucs2, utf16, utf32.

  Every tailoring was added with four accent and case
  sensitivity flag combinations, e.g:

  * utf8mb4_uca1400_swedish_as_cs
  * utf8mb4_uca1400_swedish_as_ci
  * utf8mb4_uca1400_swedish_ai_cs
  * utf8mb4_uca1400_swedish_ai_ci

  and their _nopad_ variants:

  * utf8mb4_uca1400_swedish_nopad_as_cs
  * utf8mb4_uca1400_swedish_nopad_as_ci
  * utf8mb4_uca1400_swedish_nopad_ai_cs
  * utf8mb4_uca1400_swedish_nopad_ai_ci

- Introducing a conception of contextually typed named collations:

  CREATE DATABASE db1 CHARACTER SET utf8mb4;
  CREATE TABLE db1.t1 (a CHAR(10) COLLATE uca1400_as_ci);

  The idea is that there is no a need to specify the character set prefix
  in the new collation names. It's enough to type just the suffix
  "uca1400_as_ci". The character set is taken from the context.

  In the above example script the context character set is utf8mb4.
  So the CREATE TABLE will make a column with the collation
  utf8mb4_uca1400_as_ci.

  Short collations names can be used in any parts of the SQL syntax
  where the COLLATE clause is understood.

- New collations are displayed only one time
  (without character set combinations) by these statements:

     SELECT * FROM INFORMATION_SCHEMA.COLLATIONS;
     SHOW COLLATION;

  For example, all these collations:
  - utf8mb3_uca1400_swedish_as_ci
  - utf8mb4_uca1400_swedish_as_ci
  - ucs2_uca1400_swedish_as_ci
  - utf16_uca1400_swedish_as_ci
  - utf32_uca1400_swedish_as_ci
  have just one entry in INFORMATION_SCHEMA.COLLATIONS and SHOW COLLATION,
  with COLLATION_NAME equal to "uca1400_swedish_as_ci", which is the suffix
  without the character set name:

SELECT COLLATION_NAME FROM INFORMATION_SCHEMA.COLLATIONS
WHERE COLLATION_NAME LIKE '%uca1400_swedish_as_ci';

+-----------------------+
| COLLATION_NAME |
+-----------------------+
| uca1400_swedish_as_ci |
+-----------------------+

  Note, the behaviour of old collations did not change.
  Non-unicode collations (e.g. latin1_swedish_ci) and
  old UCA-4.0.0 collations (e.g. utf8mb4_unicode_ci)
  are still displayed with the character set prefix, as before.

- The structure of the table INFORMATION_SCHEMA.COLLATIONS was changed.

  The NOT NULL constraint was removed from these columns:
  - CHARACTER_SET_NAME
  - ID
  - IS_DEFAULT
  and from the corresponding columns in SHOW COLLATION.

  For example:

SELECT COLLATION_NAME, CHARACTER_SET_NAME, ID, IS_DEFAULT
FROM INFORMATION_SCHEMA.COLLATIONS
WHERE COLLATION_NAME LIKE '%uca1400_swedish_as_ci';
+-----------------------+--------------------+------+------------+
| COLLATION_NAME | CHARACTER_SET_NAME | ID | IS_DEFAULT |
+-----------------------+--------------------+------+------------+
| uca1400_swedish_as_ci | NULL | NULL | NULL |
+-----------------------+--------------------+------+------------+

  The NULL value in these columns now means that the collation
  is applicable to multiple character sets.
  The behavioir of old collations did not change.
  Make sure your client programs can handle NULL values in these columns.

- The structure of the table
  INFORMATION_SCHEMA.COLLATION_CHARACTER_SET_APPLICABILITY was changed.

  Three new NOT NULL columns were added:
  - FULL_COLLATION_NAME
  - ID
  - IS_DEFAULT

  New collations have multiple entries in COLLATION_CHARACTER_SET_APPLICABILITY.
  The column COLLATION_NAME contains the collation name without the character
  set prefix. The column FULL_COLLATION_NAME contains the collation name with
  the character set prefix.

  Old collations have full collation name in both FULL_COLLATION_NAME and
  COLLATION_NAME.

SELECT COLLATION_NAME, FULL_COLLATION_NAME, CHARACTER_SET_NAME, ID, IS_DEFAULT
FROM INFORMATION_SCHEMA.COLLATION_CHARACTER_SET_APPLICABILITY
WHERE FULL_COLLATION_NAME RLIKE '^(utf8mb4|latin1).*swedish.*ci$';
+-----------------------------+-------------------------------------+--------------------+------+------------+
| COLLATION_NAME | FULL_COLLATION_NAME | CHARACTER_SET_NAME | ID | IS_DEFAULT |
+-----------------------------+-------------------------------------+--------------------+------+------------+
| latin1_swedish_ci | latin1_swedish_ci | latin1 | 8 | Yes |
| latin1_swedish_nopad_ci | latin1_swedish_nopad_ci | latin1 | 1032 | |
| utf8mb4_swedish_ci | utf8mb4_swedish_ci | utf8mb4 | 232 | |
| uca1400_swedish_ai_ci | utf8mb4_uca1400_swedish_ai_ci | utf8mb4 | 2368 | |
| uca1400_swedish_as_ci | utf8mb4_uca1400_swedish_as_ci | utf8mb4 | 2370 | |
| uca1400_swedish_nopad_ai_ci | utf8mb4_uca1400_swedish_nopad_ai_ci | utf8mb4 | 2372 | |
| uca1400_swedish_nopad_as_ci | utf8mb4_uca1400_swedish_nopad_as_ci | utf8mb4 | 2374 | |
+-----------------------------+-------------------------------------+--------------------+------+------------+

- Other INFORMATION_SCHEMA queries:

  SELECT COLLATION_NAME FROM INFORMATION_SCHEMA.COLUMNS;
  SELECT COLLATION_NAME FROM INFORMATION_SCHEMA.PARAMETERS;
  SELECT TABLE_COLLATION FROM INFORMATION_SCHEMA.TABLES;
  SELECT DEFAULT_COLLATION_NAME FROM INFORMATION_SCHEMA.SCHEMATA;
  SELECT COLLATION_NAME FROM INFORMATION_SCHEMA.ROUTINES;
  SELECT COLLATION_CONNECTION FROM INFORMATION_SCHEMA.EVENTS;
  SELECT DATABASE_COLLATION FROM INFORMATION_SCHEMA.EVENTS;
  SELECT COLLATION_CONNECTION FROM INFORMATION_SCHEMA.ROUTINES;
  SELECT DATABASE_COLLATION FROM INFORMATION_SCHEMA.ROUTINES;
  SELECT COLLATION_CONNECTION FROM INFORMATION_SCHEMA.TRIGGERS;
  SELECT DATABASE_COLLATION FROM INFORMATION_SCHEMA.TRIGGERS;
  SELECT COLLATION_CONNECTION FROM INFORMATION_SCHEMA.VIEWS;

  display full collation names, including character sets prefix,
  for all collations, including new collations.

  Corresponding SHOW commands also display full collation names
  in collation related columns:

  SHOW CREATE TABLE t1;
  SHOW CREATE DATABASE db1;
  SHOW TABLE STATUS;
  SHOW CREATE FUNCTION f1;
  SHOW CREATE PROCEDURE p1;
  SHOW CREATE EVENT ev1;
  SHOW CREATE TRIGGER tr1;
  SHOW CREATE VIEW;

  These INFORMATION_SCHEMA queries and SHOW statements may change in
  the future, to display show collation names.

9d5a0fe... by Alexander Barkov

MDEV-27009 Add UCA-14.0.0 collations - adding version aware implicit weight handling

Implicit weights are now handled according to the Unicode version
(14.0.0 vs earlier versions).

- Adding a new member MY_UCA_INFO::version

- Copy logical positions and the version from "src_uca" to "new_uca"
  in init_weight_level().

- Adding a "const MY_UCA_INFO *" parameter to a few functions
  to know Unicode version to generate implicit weights accordingly:
  - during the collation initialization time, to pages which are
    a mixture of explicit and implicit weights
  - during comparison time, for fully implicit pages

3a8851f... by Alexander Barkov

MDEV-27009 Add UCA-14.0.0 collations - dump logical positions and contractions

- uca-dump can now dump logical positions as a set of "#define" directives.
  Logical positions for 4.0.0 and for 5.2.0 were calculated and put into
  ctype-uca.c manually. That required some efforts by analyzing allkeys.txt
  with help of grep and sort.
  Now when defining a new MY_UCA_INFO it's possible to use the new #define's
  instead of calculating logical positions manually.
  Logical positions also print their weights in DUCET format as a comment
  before the define:

/*
[.0000.0021.0002]
[.0000.0117.0002]
*/

  The comment helps to know weight ranges on various levels,
  which makes it easier to debug the code.

- uca-dump can now dump built-in DUCET contractions

- Adding a new uca-dump command line option --no-contractions, this is useful
  if one needs to re-dump 4.0.0 and 5.2.0 data in ctype-uca.c compatible way.

- Adding a new uca-dump command line options --case-first=upper|level.
  This can be useful if one need to dump with UPPER case first by default.
  It's not yet decided if we'll use --case-first=upper during the dump though.

- Moving parts of the code from the main loop into separate functions
  parse_chars() and parse_weights(). This allows to reuse the code between
  single characters and contractions.

- Adding a new function my_ducet_weight_normalize(), to cut zero weights
  from a weight string, e.g. [AAAA][0000][BBBB] -> [AAAA][BBBB].
  This helps to reuse the code between single characters and contractions.

- Weight normalization is now done before printing, in separate loops inside
  my_ducet_normalize(). Before this change, normalization was done during
  priting, inside the printing loop. This helps to separate steps:
  loading -> normalizing -> printing.
  This makes it easier to follow what's going on, e.g. while debugging.

- Fixing ctype-uca.c to handle built-in contractions of any length.
  Previously we had only built-in contractions in utf8mb4_thai_520_w2,
  which contains only 2-character contractions.

c90574a... by Alexander Barkov

MDEV-27009 Add UCA-14.0.0 collations - Adding implicit weight handling for Unicode-14.0.0

1. Adding separate functions for different Unicode versions
  - my_uca_520_implicit_weight_primary()
   It calculates implicit weights according to the old algorithm
   that we used to dump Unicode-5.2.0 weights.

  - my_uca_1400_implicit_weight_primary()
    It calculates implicit weights according to
    https://unicode.org/reports/tr10/#Values_For_Base_Table
    as of November 2021, Unicode version 14.0.0.

2. Adding the "@version" line recognition when dumping allkeys.txt.
   Implicit weights are dumped according to @version.

3. Dumping the scanned version as a "#define"

4. Removing dumping MY_UCA_NPAGES, MY_UCA_NCHARS, MY_UCA_CMASK, MY_UCA_PSHIFT,
   as they are defined in ctype-uca.c. Removing dumping of "main()", it's not
   needed. The intent is to generate an *.h file which can be put directly
   to the MariaDB source tree.

5. Adding a structure MY_DUCET. It now contains weights for single
   characters and version related members. Later we'll add contractions
   and logical positions in here.

8ba2efc... by Alexander Barkov

MDEV-27009 Add UCA-14.0.0 collations - adding uca-dump into build targets

- Adding uca-dump into build targets
- Adding ctype-uca.h and moving implicit weight related routines there
- Reusing implicit weight routines in ctype-uca.c and uca-dump.c
- Adding handling of command line arguments to uca-dump
- Fixing some compile-time warnings in uca-dump.c

14ecdad... by Sergei Golubchik

remove invalid test

it starts an EXPLAIN of a multi-table join and tries to KILL it.
no sync points.
depending on how fast the hareware is and optimizer development
it might kill EXPLAIN at some random point in time (generally unrelated
to the Bug#28598 it was supposed to test) or EXPLAIN might finish
before the KILL and the test will fail.

d371e35... by Marko Mäkelä

Merge 10.9 into 10.10