Re: Query cache is not used for tables or databases with dots in their names
So this is effectively a regression then, since an older (upstream) codebase does not exhibit the problem.

Of course it's utterly nasty to have table names with dots in them, but since it's allowed...

Re: Query cache is not used for tables or databases with dots in their names
I've also confirmed the exact same behaviour if a table has a - (dash) in its name.

Launchpad bug id: 671189

I tried this test suite it looks like working:
– source include/have_query_cache.inc

set GLOBAL query_cache_size=1355776;

flush query cache;
flush status;

CREATE DATABASE `foo.bar`;
use `foo.bar`;
CREATE TABLE moocow (a int);
INSERT INTO moocow VALUES (1), (2), (3);
SHOW STATUS LIKE 'Qcache%';
SELECT * FROM moocow;
SHOW STATUS LIKE 'Qcache%';
use test;
drop table `foo.bar`;

The missing bit of info in the description is that the default storage engine in ubuntu packages is InnoDB, so moocow is actually an InnoDB table. With this correction, it's still reproducible on all current versions. The test case should be like this:

--source include/have_query_cache.inc
--source include/have_innodb.inc

set GLOBAL query_cache_size=1355776;

flush query cache;
flush status;

CREATE DATABASE `foo.bar`;
use `foo.bar`;
CREATE TABLE moocow (a int) ENGINE=InnoDB;
INSERT INTO moocow VALUES (1), (2), (3);
SHOW STATUS LIKE 'Qcache%';
SELECT * FROM moocow;
SHOW STATUS LIKE 'Qcache%';
use test;
drop database `foo.bar`;
set GLOBAL query_cache_size = DEFAULT;

And it's reproducible on MySQL 5.1-5.6 too.

As I thought from comment abouve it is bug of innodb engine:

T@5 : | | | | | >Query_cache::ask_handler_allowance
T@5 : | | | | | | >innobase_trx_init
T@5 : | | | | | | <innobase_trx_init
T@5 : | | | | | | qcache: Handler does not allow caching for foo.bar.moocow
T@5 : | | | | | <Query_cache::ask_handler_allowance

The problem is in names escaping:
(gdb) p name
$2 = 0x7ffff7ef54d0 "foo.bar/moocow"
(gdb) p field
$3 = (const unsigned char *) 0x7fffe8524122 "foo@002ebar/moocow"

tablename_to_filename should be called for innodb engine...

It is also repeatable on MySQL 5.1 (with innobase build from sources)

Fixing this bug require niticeble overhead for each table in the query during caching or returnung from the cache, when we have uncached only 1% of queries.

To make the impact minimum we have to change handler interface and makes QC storing both variants of the table/db names (i.e. it can't be done as just a bugfix in old version).

It will not be fixed in 5.* , but in 10.0, because fix will be expensive or need intrusive changes in QC which could be done only in 10.0

it's not a bug in 10.0.2!?

reset query cache;
create database `foo.bar`;
use `foo.bar`;
CREATE TABLE moocow (a int);
INSERT INTO moocow VALUES (1), (2), (3);
SELECT SQL_CACHE * FROM moocow;
select * from information_schema.QUERY_CACHE_QUERIES;

ENTRY_POSITION_IN_CACHE|STATEMENT_SCHEMA|STATEMENT_TEXT|QUERY_ROWS|QUERY_HITS|QUERY_HITS_TOTAL_TIME_US|QUERY_LAST_HIT_TIME|SELECT_EXPEND_TIME_US|SELECT_LOCK_TIME_US|SELECT_ROWS_READ|TABLES|QUERY_INSERT_TIME|RESULT_LENGTH|RESULT_BLOCKS_COUNT|RESULT_BLOCKS_SIZE|RESULT_BLOCKS_SIZE_USED|RESULT_TABLES_TYPE|RESULT_FOUND_ROWS|FLAGS_CLIENT_LONG_FLAG|FLAGS_CLIENT_PROTOCOL_41|FLAGS_PROTOCOL_TYPE|FLAGS_MORE_RESULTS_EXISTS|FLAGS_IN_TRANS|FLAGS_AUTOCOMMIT|FLAGS_PKT_NR|FLAGS_CHARACTER_SET_CLIENT|FLAGS_CHARACTER_SET_RESULTS|FLAGS_COLLATION_CONNECTION|FLAGS_LIMIT|FLAGS_TIME_ZONE|FLAGS_SQL_MODE|FLAGS_MAX_SORT_LENGTH|FLAGS_GROUP_CONCAT_MAX_LEN|FLAGS_DIV_PRECISION_INCREMENT|FLAGS_DEFAULT_WEEK_FORMAT|FLAGS_LC_TIME_NAMES
0|foo.bar|SELECT * FROM moocow|3|0|0|0.000000|208|41|3|`foo.bar`.`moocow`|1370765538.000000|88|1|512|152|0|3|1|1|0|0|0|1|1|utf8|utf8|utf8_general_ci|-1|SYSTEM|PIPES_AS_CONCAT,ALLOW_INVALID_DATES|1024|1024|5|0|en_US

fixed in upcoming 10.0. (now in 10.0-merge)