RE: LoggingEvent serialization compatibility between 1.2 and 1.3

[Ceki Gülcü <ce...@qos.ch>]

At 09:20 AM 12/5/2004, Scott Deboy wrote:
>I haven't tried sending events from log4j 1.3 to log4j 1.2.8, but I doubt 
>deserialization would work, without more changes to serialization logic.
>
>In 1.3, LoggingEvent contains a new field:
>long sequenceNumber
>The 1.2.8 code wouldn't know what to do with it.
>
>As one of the contributors to an app that can accept 
>SocketAppender-generated events, I would also like to see full serial 
>compatibility between versions.
>
>I'm not aware of a technical reason that would prevent fully supporting 
>serial compatibility between 1.2.9 and 1.3, but such a change requires 
>time and incurs a complexity cost on the code, and that's one area where 
>I'm sure different folks have different views on the cost vs. benefit of 
>such a change, which means such a change is unlikely to be fully supported.

I've been hinting at this for quite some time but apparently not loud
enough.

The reason there was no serious effort to preserve bw compatibility
between 1.2 and 1.3 is performance improvements scheduled for
1.3. Given that we always transmit LoggingEvent objects on the wire, a
protocol specialized for transmitting LoggingEvent objects, and only
LoggingEvent objects, can be made to perform much better than a
generic protocol such as Java serialization. No need to waste time
worrying about MDC or LocationInfo when the whole shebang will be
swept away.

Our new transmission protocol will be simple enough to be implemented
in other languages such as C++ or C#.

Any volunteers to start this work?


>Scott

-- 
Ceki Gülcü

  The complete log4j manual:  http://qos.ch/eclm
  Professional log4j support: http://qos.ch/log4jSupport  



---------------------------------------------------------------------
To unsubscribe, e-mail: log4j-dev-unsubscribe@logging.apache.org
For additional commands, e-mail: log4j-dev-help@logging.apache.org

Re: LoggingEvent serialization compatibility between 1.2 and 1.3

[Niclas Hedhman <ni...@hedhman.org>]

On Tuesday 07 December 2004 05:23, Curt Arnold wrote:

> I'm sure that you could do better with a custom wire-format than the
> current Java default serialization, however I'm not sure that you would
> see any significant advantage over custom serialization.  The benefits
> for using Java custom serialization would be a well defined wire
> format, usage patterns and language support within log4j.  I doubt that
> it would make the implementation of log4net and log4cxx much harder
> than any other protocol.

Well, Serialization is not dead-easy to deserialize in another language, and 
compared to a text/line based protocol it is outright difficult. :o)

The concern, however, may be raised in to outset of a larger system with many 
different Log4J versions, with a single management system.


Cheers
Niclas
-- 
   +------//-------------------+
  / http://www.dpml.net       /
 / http://niclas.hedhman.org / 
+------//-------------------+


---------------------------------------------------------------------
To unsubscribe, e-mail: log4j-dev-unsubscribe@logging.apache.org
For additional commands, e-mail: log4j-dev-help@logging.apache.org

Re: LoggingEvent serialization compatibility between 1.2 and 1.3

[Curt Arnold <ca...@houston.rr.com>]

> I've been hinting at this for quite some time but apparently not loud
> enough.
>
> The reason there was no serious effort to preserve bw compatibility
> between 1.2 and 1.3 is performance improvements scheduled for
> 1.3. Given that we always transmit LoggingEvent objects on the wire, a
> protocol specialized for transmitting LoggingEvent objects, and only
> LoggingEvent objects, can be made to perform much better than a
> generic protocol such as Java serialization. No need to waste time
> worrying about MDC or LocationInfo when the whole shebang will be
> swept away.
>
> Our new transmission protocol will be simple enough to be implemented
> in other languages such as C++ or C#.
>
> Any volunteers to start this work?
>
>

I'm sure that you could do better with a custom wire-format than the 
current Java default serialization, however I'm not sure that you would 
see any significant advantage over custom serialization.  The benefits 
for using Java custom serialization would be a well defined wire 
format, usage patterns and language support within log4j.  I doubt that 
it would make the implementation of log4net and log4cxx much harder 
than any other protocol.


---------------------------------------------------------------------
To unsubscribe, e-mail: log4j-dev-unsubscribe@logging.apache.org
For additional commands, e-mail: log4j-dev-help@logging.apache.org