Hi Ruwan,

thanks for your response. As described earlier it is perfectly clear to me that 
the usage of LinkedHashMaps would not solve the „problem“ entirely as the order 
of top level elements might still change. On the other hand the more annoying 
problem is the order of the elements below the top level elements, e.g. the 
list of hundreds of proxies, sequences and so on in random order.
It was just a thought that changing the implementation to LinkedHashMaps would 
not hurt and at least address this particular issue. I tested it out and it is 
really an improvement.

Rethinking about the serialization process and the object store structure might 
also be a good thought. Enhancing on this part could also ease the process of 
writing custom mediators. Writing factories/serializer pairs gives a lot of 
control but also adds a lot of boilerplate-code and forces users to get in 
touch with rather low level XML-API (Axiom).

Regards,
   Eric

________________________________
From: Ruwan Linton [mailto:[email protected]]
Sent: Monday, November 09, 2009 11:20 PM
To: [email protected]
Subject: Re: Constant/reproducable order of elements in synapse.xml

Hi Eric,

I was on vacation and just got back. :-) Going through my email backlog.

So, First of all I do not understand the issue that you are facing with the 
order. I agree that synapse serializes according to a hard order while builders 
do not expect that order. There is no specific reason to have that hard order, 
it was just the way that the serialization has been implemented. Since XML 
doesn't care about the order in which the elements are present, the 
serialization just goes on the different elements.

Also if we are to preserve the order in which they were present in the 
synapse.xml that is used to build the configuration, making the object 
structure to be stored in LinkedHashMaps is not going to work since we keep 
separate HashMaps for the separate elements, for example proxy services are 
kept in its own HashMap while endpoints, sequences and local entries are on a 
different one.

If we are to do this change we need to rethink about the serilization process 
and the object store structure.

Thanks,
Ruwan

On Mon, Nov 9, 2009 at 11:04 AM, Hubert, Eric 
<[email protected]<mailto:[email protected]>> wrote:
Hi Supun,

If the object model does not preserve the order in which it has been created a 
serializer would need to store the order in some place independent from the 
configuration model itself which is also accessible by the deserializer at any 
later stage. Another implication is, that it is not enough to feed a 
deserializer just with the configuration model alone. It would need the 
additional meta data to be able to preserve the original order.
Also any serializer/factory-deserializer-pair would need to implement this.

Maybe others have already discussed about this at design time? Any input from 
the devs early involved?

Regards,
   Eric
________________________________
From: Supun Kamburugamuva [mailto:[email protected]<mailto:[email protected]>]
Sent: Monday, November 09, 2009 5:46 AM
To: [email protected]<mailto:[email protected]>
Subject: Re: Constant/reproducable order of elements in synapse.xml

Hi Eric,

I thought about this a bit further. Synapse runs on an object model. Not on a 
XML configuration. As far as I understood this was the original goal. Also it 
makes sense. The synapse configuration can be created using XML or may be 
spring or may be pure programming. At the synapse object model layer I think it 
is logical to use a hash map. If a particular serialization and building wants 
to have an order it is the responsibility of the build and serialize method 
writer to impose it.

Isn't there a way to impose this at the serialize and builder level?

Thanks,
Supun..
On Sun, Nov 8, 2009 at 2:22 AM, Hubert, Eric 
<[email protected]<mailto:[email protected]>> wrote:
Hi Supun,

Please see my comments inline!

Synapse language is treated as a programming language at least for the out side 
world. For example sequences can be attributed to functions. In any programming 
language the order in which these symbols occur does not matter as long as the 
symbols can be found by the caller.
Agreed.
So I have doubts in implementing an order for the synapse elements. By order I 
meant something like all the sequences are at the top then the proxy services 
etc.
I share these doubts, but you describe exactly the current implementation. 
Please have a look at: 
SynapseXMLConfigurationSerializer.serializeConfiguration(SynapseConfiguration 
synCfg). It uses a fixed order of top level elements.
Specific element serializers are then using an order depending on the way the 
configuration has been stored in memory. Mostly the XML information will be 
transferred to strongly typed data structures. While deserializing those 
values, currently a fixed order will be used. This might be different, from the 
one originally read in. Some list like structures are stored to unordered 
Map-Implementations for faster key-value access. While deserializing those 
values you end up with a random order. I think that this part could be fixed 
rather easily by using LinkedHashMap preserving the key order in which entries 
have been added to the map.


When a user types in the synapse.xml, we should be able to preserve the order 
in which they have entered the elements in the synapse.xml. If you meant this 
one I'm +1 for implementing it.
I also agree this would be the most desirable option, although the current 
implementation seems to be very far a way from that. Each factory/serializer 
pair would need to be designed for this purpose (e.g. keep a list or list of 
linked maps or any suitable structure of read elements/attributes, store it for 
deserialization purposes only and use this to retrieve the values in the exact 
order from structured object structures). I guess only something like this 
could warrant a working round-trip serialization/deserialization where the user 
may modify any part of the configuration, insert new stuff at any allowed place 
and this exact order will be preserved during later deserializations.
So my concrete suggestion for a start was just to replace all class member 
implementations of SynapseConfiguration which are currently of type HashMap 
with LinkedHashMap. From my point of view this is at least an improvement. No 
pain, big gain? ;-)
Regards,
    Eric



--
Software Engineer, WSO2 Inc
http://wso2.org
supunk.blogspot.com<http://supunk.blogspot.com>



--
Ruwan Linton
Technical Lead & Product Manager; WSO2 ESB; http://wso2.org/esb
WSO2 Inc.; http://wso2.org
email: [email protected]<mailto:[email protected]>; cell: +94 77 341 3097
blog: http://ruwansblog.blogspot.com

Reply via email to