Tiny Workflow Language

From: "andrew cooke" <andrew@...>

Date: Wed, 8 Mar 2006 08:54:18 -0300 (CLST)

I think I may finally have understood this "business process" thing at
work today.  If the following still makes sense in the cold light of
tomorrow morning then I will implement it at work.

tiny.package-info: "A tiny workflow"
tiny.Test: interface DomXml->boolean
tiny.Sink: interface Annotated->void
tiny.Flow: interface extends Test, Sink
tiny.Process: class Filter Flow Object->void
tiny.sink.package-info: "Simple sinks for error handling"
tiny.sink.Throw: class<Sink Class<Exception>
tiny.sink.Silent: class<Sink
tiny.sink.LogMsg: class<Sink String (or XPath?)
tiny.sink.Log: class<Sink String (or XPath?)
tiny.test.package-info: "Tests using XML annotation"
tiny.test.Always: class<Test
tiny.test.Never: class<Test
tiny.test.XPath: class<Test XPathExpression
tiny.flow.package-info: "Build Flow-based sequences"
tiny.flow.All: class<Flow Test List<Flow>
tiny.flow.First: class<Flow Test List<Flow> Sink
tiny.flow.Pair: class<Flow Test Sink
tiny.input.package-info: "Conversion to the internal format"
tiny.input.Normalize: interface Object->boolean Object->Annotated
tiny.input.Filter: class List<Normalize> Sink
tiny.input.Name: class<Normalize String

Notes
- the Sinks in First and Filter are exception handlers
- assembly via constructors or setters (Spring)
- All + First give basic logic, sequencing, etc
- Support for single elements/arrays as well as lists, for easy config

The basic idea is that you define a Process that accepts an Object and
calls an appropriate Sink.  The input Object is processed by a matching
Filter to give an Annotated which combines the original Object with an Xml
annotation.  Subsequent processing decisions are based on the annotation,
which is passed to various Tests.

Realising that Flow was the only necessary component for constructing more
complex logic (ie that Test could be Always if the test were irrelevant)
was critical in simplifying the types.  Without that, classes in the flow
package had to manage various combinations of Test and Sink.

I think the set of operations above is sufficient for quite a range of
tasks.  It'll be interesting to see if (1) this is anything like the right
idea and (2) if so, how it compares to something like BEPL.

The most obvious things that are missing at the moment are:
- the ability to modify the annotation (state)
- sophisticated error handling

An example in Java might be:

new Process(
new Filter(
new IngestRequestNormalize(),
new Throw("Cannot normalize")),
new First(
new Flow[] {
new TestSinkPair(
new XPath("a/b/c/Mosaic"),
new MosaicIngest())
new TestSinkPair(
new XPath("p/q/r/Echelle"),
new EchelleIngest())},
new Throw("Cannot ingest")))

That's a process that converts incoming data if it's an IngestRequest.
Converted data are then tested to see if they are from a Mosaic and, if
so, ingested.  If not, a test is made for Echelle data.  If that fails
too, an exception is raised.

In Spring it would be similar, but with properties.  I think it will look
pretty good.

Now I need to sleep.  Want to run tomorrow before work.

Andrew