Revise new NullChannel() usage in favor of nullChannel bean

[artembilan]

The problem with a new NullChannel() that we won't get a consistent metrics on the nullChannel bean.

Also consider to change successTimer in the NullChannel into the sendCounter since we don't care about timing on the nullChannel, but rather how many messages have been discarded.

[garyrussell]

We can't change the meter type without adding special code to the graph server.

[artembilan]

OK. I think we can live with 0-timer then. I hope it shows some throughput statistics anyway...

[garyrussell]

I only see these...

$ find . -name "*.java" | grep -v 'src/test' | xargs grep 'new NullChannel'
./spring-integration-file/src/main/java/org/springframework/integration/file/dsl/TailAdapterSpec.java:			this.factoryBean.setOutputChannel(new NullChannel());
./spring-integration-core/src/main/java/org/springframework/integration/transaction/ExpressionEvaluatingTransactionSynchronizationProcessor.java:	private volatile MessageChannel beforeCommitChannel = new NullChannel();
./spring-integration-core/src/main/java/org/springframework/integration/transaction/ExpressionEvaluatingTransactionSynchronizationProcessor.java:	private volatile MessageChannel afterCommitChannel = new NullChannel();
./spring-integration-core/src/main/java/org/springframework/integration/transaction/ExpressionEvaluatingTransactionSynchronizationProcessor.java:	private volatile MessageChannel afterRollbackChannel = new NullChannel();
./spring-integration-core/src/main/java/org/springframework/integration/aggregator/AbstractCorrelatingMessageHandler.java:	this.discardChannel = new NullChannel();

I suppose it was the aggregator where you saw problems?

[artembilan]

We can fix a Spec: the MessageProducerSupport has setOutputChannelName(). Of course, we would need to fix a FileTailInboundChannelAdapterFactoryBean as well, but it is not clear to me why would TailAdapterSpec worry about that NullChannel at all...

Yes, an AbstractCorrelatingMessageHandler is an origin of this issue, but I don't think we should leave all others as is.
Although I would say ExpressionEvaluatingTransactionSynchronizationProcessor should just deal with null value for those channels instead of polluting a NullChannel...