This looks like it has been fixed for a while already. See:

http://github.com/doctrine/dbal/blob/master/lib/Doctrine/DBAL/Types/SmallIntType.php
http://github.com/doctrine/dbal/blob/master/lib/Doctrine/DBAL/Types/DecimalType.php
http://github.com/doctrine/dbal/blob/master/lib/Doctrine/DBAL/Types/BooleanType.php

I'll take a look at the master next time before submitting a bug report. Anyway I think it would be better to not fix bugs without an existing ticket for the fixed bug

has already been fixed

Linux

Fixed in current master

MySQL Server version: 5.0.51a-24+lenny4 (Debian)

Fixed

Pity that because of this the Doctrine schema-tool update action reports incorrect change list.
It tries to drop/add foreign keys because "ON DELETE CASCADE" option isn't read from the database at all.

5.0.x has no way to export the CASCADE details. Its just not possible to get this data in 5.0

Doctrine - GIT trunk (night build from 2010-07-17), Windows 7, PHP 5.3.2, PostgreSQL 8.4.3, Apache 2.2.15

Both DateTime and DateTimeTz assume that TIMESTAMP(0) is the definition, not the TIMESTAMP that implicitly degrades to TIMESTAMP(6). I am not sure how to handle this, I find this Postgres DateTime stuff rather annoying

Is there a global client side option for Postgres users to configure this?

@Jan: Did you use create the column manually yourself (TIMESTAMP WITHOUT TIMEZONE) or use the Doctrine Schema-Tool which defines TIMESTAMP(0) WITHOUT TIMEZONE.

Fixed in DBAL Trunk, see the following section of the DBAL manual to understand the workaround for PostgreSQL TIMESTAMP( n ) types where n > 0.

ORM always creates TIMESTAMP(0), so this is handled as a legacy database schema.

@Benjamin: Yes, I have created the column manualy directly in database using my own CREATE TABLE definition.

fixed

Linux, PostgreSQL 8.4

fixed

Fixed.

Linux, PostgreSQL 8.4, PHP 5.3.2

fixed, changed on other exceptions also.

maybe there should be a null check in _gatherRowData method, before the value is passed to convertToPHPValue of any type class

This is a bug in the ConversionException code, sorry for that i will fix it tonight.

Btw, you are not using Beta2 but trunk, the ConversionException code was just committed yesterday. Additionally you should use the ORM with the Beta2 not with trunk of DBAL, they don't work together currently in some aspects (SchemaTool)

This was introduced with DBAL-22 and is now fixed

throwing an exception is a no-go in my opinion, it would completely cripple an application if there is some change in the DB.

There are no checks if an integer field is really in int, or if the db has a varchar field making the (int) conversion fail. This should be ignored and is responsibility of the user to get right.

I am looking into the format, however we have several unit-tests for this working correctly, i am a bit confused.

The question is rather, should we deprecate the usage of timezones alltogether? This is a good read on the topic: http://derickrethans.nl/storing-date-time-in-database.html

ok the current state is, we use "TIMESTAMP(0) WITH TIMEZONE", which matches the currently given format "Y-m-d H:i:sO"

The "TIMESTAMP(0) WITH TIMEZONE" is a very UNCOMMON time field

Usually timstamp without time zone is used

Also the difference between integer field is really in int and db has a varchar field making the (int) is different for one reason

The timestamp is returning a datetime object - not a scalar

What happens when you try to use something in PHP that's not an object as an object? It fatal errors

If the integer field is not really an integer when you get your data back from the db and you attempt to use it, PHP is not going to throw a fatal error at you. Your application might be angry, but that's another story.

I would almost rather you simply use strtotime and pass that to the datetime class for postgresql since it will work on all 'versions" of postgresql's timestamp, regardless if it has microseconds or timezone information. Then you're not locking all users of doctrine into using one timestamp format.

$full = '2010-06-11 17:18:39.808397-04';
$no_micro = '2010-06-11 17:18:39-04';
$no_tz = '2010-06-11 17:18:39';
var_dump(DateTime::createFromFormat('Y-m-d H:i:sO', $full));
var_dump(DateTime::createFromFormat('Y-m-d H:i:s.uO', $full));

var_dump(new DateTime($full));
var_dump(new DateTime($no_micro));
var_dump(new DateTime($no_tz));

Quick test - why rely on createFromFormat when the datetime constructor is smart enough to handle all formats?

One reason is probably performance, using the constructor takes double the time:

$full = '2010-06-11 17:18:39.808397-04';
$no_micro = '2010-06-11 17:18:39-04';
$no_tz = '2010-06-11 17:18:39';

$n = 5000;

$ts = microtime(true);
for ($i = 0; $i < $n; $i++) {
    DateTime::createFromFormat('Y-m-d H:i:sO', $no_micro);
    DateTime::createFromFormat('Y-m-d H:i:s.uO', $full);
}
echo "DateTime::createFromFormat: " . number_format(microtime(true) - $ts, 4)."\n";

for ($i = 0; $i < $n; $i++) {
    new DateTime($full);
    new DateTime($no_micro);
}
echo "new DateTime: " . number_format(microtime(true) - $ts, 4)."\n";

DateTime::createFromFormat: 0.1173
new DateTime: 0.2496

Since the DateFormat is pretty static, therefore using createFromFormat() seems an obvious choice. This is of course mainly a good reason for MySQL where you cannot change the format, and Postgres, Oracle and DB2 allow changes in precision and formatting of the date which make this decision a little bit more complicated.

I think everyone BUT mysql allow changes in the precision and formatting of the date - heck SQL Server is the worst for "make the date be whatever you want"

I think this would fall into "premature optimization" - yes it's a great speed improvement for mysql, but it's at the expense of all other DBs...

A Doctrine\DBAL\DBALException due to a wrong date in the database probably leads to a fatal error in any userland code, or would anybody catch all the errors? how would you handle them?

We could think of changing the type to be without timezone, that would be a pretty massive BC though since it affects users database schemas.

Another option would be a configurable format per Postgres/Oracle/DB2 Platform:

$config = new \Doctrine\ORM\Configuration();
$config->setDateTimeGlobalFormat('Y-m-d H:i:s.uO');

However Type instances are flyweight instances, that means it would have to be the format of all Doctrine DateTime typed columns. However this way you would at least have full control over the format, not depending on any Doctrine 2 interpretation.

You can of course add your own datetime type and overwrite the existing or use different ones. Maybe we should supply two DateTime types.

I don't think an exception is a good idea
A PHP warning/notice might be a better solution so you at least know what failed (it shouldn't kill the script though, as a fatal error and exception do)

However a configurable would probably be a good idea...
Or two datetime types - one that allows a format to be used
Or even the ability to define a format for your datetime field (with a fallback to the default format)

For databases that talk to other things, not just PHP via Doctrine, dictating the format of the datetime fields is not an option

I changed the title to reflect the real issue of this ticket, using Postgresql with WITH TIMEZONE is rather problematic with regard to DateTime and DateTimeZone Handling inside PHP, see the following code snippets (http://pastie.org/1009033) and Dericks post (http://derickrethans.nl/storing-date-time-in-database.html).

We should change the default behaviour in Postgres (and Oracle) to be WITHOUT TIMEZONE, since this is the 90% use-case. Additionally it may create less bugs when we don't fiddle with the Timezone used for creation.

Ok we also discussed errors when conversion failed and added a ConversionException. We also implemented a new type DateTimeTz.

This is both currently in my feature branch: http://github.com/doctrine/dbal/tree/DBAL-22

This is a copy of a mail going to the doctrine-dev and doctrine-user lists some minutes ago:

Hello Doctrine 2 + Postgres and Oracle Users,

Both Postgres and Oracle currently save the Date Offset for DateTime
instances they are handling from Doctrine 2. However Date Offsets should
not be confused with Timezones and this can cause considerable issues
with transitions, modifications and comparisons of dates.

As a result we have to change the DateTime type implementations of
Oracle/Postgres for Beta 3 to reduce the risk of users running into date
calculation problems.

Required changes inside Doctrine DBAL Package:

• The column create statement on both platforms will be changed from
  "TIMESTAMP(0) WITH TIME ZONE" to "TIMESTAMP(0) WITHOUT TIME ZONE".
• The supported PHP date format will change from "Y-m-d H:i:sO" to
  "Y-m-d H:i:s".
• Schema-Tool will automatically detect dates with offset retrieved from
  the Database as "DateTimeTz" types.
• Wrongly converted date values will throw a "Doctrine\DBAL\Types
  \ConversionException". This will stop any non-migrated from running, but
  more importantly from corrupting your data.

What does that mean to you as a user?

POSTGRES:

There are two solutions if you use Postgres:

1. The easy way out: We will introduce a new Type "DateTimeTz", which
will keep backwards compability. You will however have to deal with the
Timezone issues yourself. One way is by setting your application
"date.timezone" ini variable to be an offset instead of a timezone on
your server. You have to make transition calculations yourself in this
case.

2. Converting the columns: When upgrading from Beta 2 to the master or
Beta 3 the type will change its behavior, conversions from a Beta 2 WITH
TIME ZONE column to the new type will fail, leading to a
"ConversionException" being thrown. However Schema-Tool will recognize
the changes automatically and ask you to convert the column. First tests
of me showed that converting the TIMESTAMP column from WITH to WITHOUT
timezone works, it even corrects all dates for the offsets to UTC.

ORACLE:

Oracle does not permit changing the types when there is already data in
it (guessing from the preliminary tests I made).
You have to switch all your entity fields using the "DateTime" type to
use the new "DateTimeTz" type otherwise you will experience an
"ConversionException" being thrown.

Planed Schedule for the changes:

• The complete changes are currently in a DBAL feature branch on Github
  http://github.com/doctrine/dbal/tree/DBAL-22
• Merge into the DBAL project this week. This WONT affect you using the
  ORM just yet, its a DBAL change only!
• The ORM Master on Github will still be linked against the DBAL Beta 2
  via a Git Submodule (doctrine2/lib/vendor/doctrine-dbal).
• In the timestamp between the DBAL Beta3 and the ORM Beta3 release we
  will integrate the changes into the ORM package also.

I will send an additional notice to the lists when we will bump the ORM
dependency on the DBAL.

Sorry for the inconvenience regarding this issue, but we feel very
strongly about making this change. This will ultimately solve many
subtle issues that would have popped up here and there.

Thanks goes to Elisabeth Smith for bringing this issue to the table and
to Derick Rethans who helped us understand why timezones/offset in the
database are a pain to work with.

greetings,
Benjamin

Merged into Master now

Hi, on master branch of DBAL package in PostgreSqlPlatform::getDateTimeTzFormatString() I see this:

public function getDateTimeTzFormatString()

But PostgreSQL stores timestamps with microseconds, so format should be more likely:

public function getDateTimeTzFormatString()

Yes, Doctrine converts DateTimeTz from PHP value to database value without microseconds, but if I will use some database function (for example now()) for get timestamp and I will store it to the database directly and than I will read this values through Doctrine, it will cause an error.

Is there any reason for not using default PostgreSQL format of timestamp with microseconds?

Let me understand this, this patch automatically detects the underlying type of a new DOMAIN Type created in postgresql?

So say i create a MONEY type and have it be a decimal or something, your code would make doctrine automatically detect it being a decimal and create the decimal doctrine type?

Indeed, it will detect the "natural" type under MONEY and use this type for doctrine.

Implemented and tested in http://github.com/doctrine/dbal/commit/93ad4c2a24adf93d450c9eefbc966ccf3fe87414

MacOS X 10.5, PHP 5.3.2

is specifying the database name optional during connection to oracle databases? otherwise we could also throw an exception.

It's optional, according to <http://php.net/oci_connect>:

"If not specified, PHP uses environment variables such as TWO_TASK (on Linux) or LOCAL (on Windows) and ORACLE_SID to determine the Oracle instance to connect to."

I just spotted the error, I don't have an Oracle database neither the oci8 extension in my machine.

fixed anyways

Yea, its currently a known limitation that the times do not get logged. However, I have real doubts as to whether that is actually useful as such measurements from PHP are very inaccurate. Its much better to copy+paste a SQL query to a SQL console and run it with ANALYZE/EXPLAIN ... to see its performance.

sfDoctrine2Plugin already uses this logger for its debug toolbar.

Such timings are very useful. The accuracy isn't really the point. If you have N queries executing on a slow request, you have to check N queries to find the problem. With a simple "inaccurate" time measurement you'd be able to quickly identify the problem query(ies) and take further action.

I think timings can be added.

Schedule for BETA 3

Implemented this change, the interface of the SQL Logger changed to:

function logSQL($sql, array $params = null, $executionMS = null);

Additionally the contract of loggers changed, they are now invoked AFTER a query is executed therefore queries that cause failures do not appear on the logs.

This should be fixed now

Introduced a regression

Fixed in current master, scheduled for Beta 3