Well we don't need what's currently provided with those legacy classes, because 
a different approach was taken with this JDBCAppender to handle ID columns.

I have several problems with this approach:

1) The implementation would be far more complicated, because the dialects would 
have to be used to build the prepared statement SQL and bind arguments to it.
2) Performance. I spent considerable time experimenting with ways to improve 
the performance of the JDBCAppender. One thing I discovered, for example, was 
that it was (much ... like 50%) more efficient to have a loop with a bunch of 
if-statements in it than to use polymorphism and method invocations to handle 
the different logic branches. The problem was the overhead of stack management 
(adding/removing method invocations from the stack). Stack management erased 
the alternate approaches I took every single time. If we now have to defer to a 
dialect to perform certain operations, performance will be impacted by stack 
management.
3) Coverage. The link you provided has 6 dialects. Hibernate, on the other 
hand, has 36 dialects. It could easily snowball to that.
4) Testing. Each dialect would either have to lack comprehensive testing, or 
we'd need a bunch of database set up.
5) As you said, this is a "one-off." The problem is in PGSQL's JDBC 
implementation. I'm very resistive to the idea of adding an entire dialect 
feature such as this merely to work around a bug only present in PGSQL's JDBC 
driver.

If someone could demonstrate a compelling need for dialects outside of working 
around this one bug in PGSQL's driver, then I might be convinced to also 
include casting in those dialects. Short of that, I think the JDBCAppender is 
much better off written generically enough that a simple, generic prepared 
statement can handle 99% of cases correctly.

Nick

On Aug 22, 2013, at 5:13 PM, Scott Deboy wrote:

> That doesn't look like much work.  Particularly since you could lift
> what's provided and just extend it to support the one-off we're
> talking about here, yes?
> 
> On 8/22/13, Nick Williams <[email protected]> wrote:
>> Yes, I mean like that. I want to avoid something like that.
>> 
>> N
>> 
>> On Aug 22, 2013, at 4:59 PM, Scott Deboy wrote:
>> 
>>> You mean like:
>>> http://logging.apache.org/log4j/companions/receivers/apidocs/org/apache/log4j/db/dialect/package-summary.html
>>> 
>>> Scott
>>> 
>>> On 8/22/13, Nick Williams <[email protected]> wrote:
>>>> IMO, this is a bug in the PGSQL JDBC driver and not something that Log4j
>>>> can
>>>> easily (or should) address. For a detailed discussion, see the PGSQL
>>>> thread
>>>> about casting text to PGSQL enums [1].
>>>> 
>>>> The problem lies in PGSQL's type handling. With PGSQL, MySQL, and any
>>>> other
>>>> database type that supports enums, you can write a query like this:
>>>> 
>>>> INSERT INTO myTable (enumColumn) VALUES ('enumValue');
>>>> 
>>>> And the database engine implicitly casts the text to enum. If you try
>>>> this
>>>> out on a PGSQL prompt, it works. However, unlike MySQL and other
>>>> vendors,
>>>> PGSQL does NOT allow String parameters in prepared statements to be
>>>> auto-casted to enums. So the following is valid in all vendors that
>>>> support
>>>> enums:
>>>> 
>>>> statement = connection.prepareStatement("INSERT INTO myTable
>>>> (enumColumn)
>>>> VALUES ('enumValue')");
>>>> 
>>>> But the following, while valid with MySQL and everything else, breaks
>>>> with
>>>> PGSQL's JDBC driver:
>>>> 
>>>> statement = connection.prepareStatement("INSERT INTO myTable
>>>> (enumColumn)
>>>> VALUES (?)");
>>>> statement.setString(1, "enumValue");
>>>> 
>>>> IMO, this was a stupid decision on PGSQL's part, and from the thread it
>>>> sounds like others agree with me. It makes it extremely difficult to use
>>>> PGSQL enums with, for example, the Java Persistence API or O/RMs in
>>>> general.
>>>> From the sounds of the thread, the guys over at PGSQL have to intention
>>>> of
>>>> fixing this.
>>>> 
>>>> Now, as for your suggestion. As much as I'd like to accommodate your
>>>> need
>>>> here, it's simply not feasible. Nearly every single database vendor has
>>>> a
>>>> different way to cast data types. Some have a CAST function, others a
>>>> CAST
>>>> keyword, and then PGSQL has that odd double-colon syntax (::). If we were
>>>> to
>>>> add a casting feature, we would then have to have a sense of database
>>>> dialects like an O/RM and learn the syntax of each database, correctly
>>>> applying the syntax based on what type of database we're logging to.
>>>> This
>>>> would be a heavyweight bulky addition to such a simple feature. My
>>>> recommendations, in descending order of how much effort it will require:
>>>> 
>>>> 1) Submit a patch to PGSQL's JDBC driver and lobby for its acceptance.
>>>> 2) Switch to a database vendor that's less [insert insulting word here]
>>>> about its enum casting--like any of the other ones. My favorite is
>>>> MySQL.
>>>> 3) Stop using a database enum column.
>>>> 
>>>> :-/
>>>> 
>>>> Nick
>>>> 
>>>> [1]
>>>> http://www.postgresql.org/message-id/cappfruyta0a0xghg4zh785ss0_fz8gczzcjzgxo2yfphadx...@mail.gmail.com
>>>> 
>>>> On Aug 22, 2013, at 4:12 PM, Peter Rifel wrote:
>>>> 
>>>>> Hello,
>>>>> 
>>>>> I'm wondering if it is at all possible to cast columns to a certain
>>>>> type
>>>>> (a custom enum) with the JDBCAppender.  I have an existing PostgreSQL
>>>>> database that I send all warnings and errors to and the log level is
>>>>> currently being stored as an enum.  When I try and log to the database
>>>>> I
>>>>> get the following message:
>>>>> 
>>>>> org.postgresql.util.PSQLException: ERROR: column "level" is of type
>>>>> log_level but expression is of type character varying  Hint: You will
>>>>> need
>>>>> to rewrite or cast the expression.
>>>>> 
>>>>> Looking through the JDBCDatabaseManager it doesn't appear that this is
>>>>> possible; including the cast in the pattern string results in it
>>>>> getting
>>>>> passed as the value itself rather than the prepared statement.  Does
>>>>> anyone have any other ideas on how to accomplish this?  I wouldn't mind
>>>>> making contributions to the source code to help implement this feature
>>>>> if
>>>>> it is desired.
>>>>> 
>>>>> Thanks,
>>>>> 
>>>>> Peter
>>>> 
>>>> 
>>> 
>>> ---------------------------------------------------------------------
>>> To unsubscribe, e-mail: [email protected]
>>> For additional commands, e-mail: [email protected]
>>> 
>> 
>> 
>> ---------------------------------------------------------------------
>> To unsubscribe, e-mail: [email protected]
>> For additional commands, e-mail: [email protected]
>> 
>> 
> 
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: [email protected]
> For additional commands, e-mail: [email protected]
> 


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to