I agree that an extensible set of black-box stream-decoders (and complementary stream encoders) that handle: zip, encryption, VB, VBS, VS, and the other 19 or so complicated legacy formats, and so forth, is a good and completely acceptable solution to this problem. I think DFDL should be helpful to the person who has to write such a stream-decoder/encoder for describing the physical stream format. This is a pragmatic decision, and for something like zip/unzip I think there is little one could do which would be more efficient than this. The rub with VS format is that most of the data is sitting there in what is very very close to the correct logical data layout. This makes it feel like copying it all to remove that little bit of excess physical structure feels unnecessary, but I agree is probably the right thing to do given the complexity involved in trying to avoid copying it. ...mikeb

-----Original Message----- From: Steve Hanson [mailto:smh@uk.ibm.com] Sent: Monday, November 22, 2004 5:58 AM To: dfdl-wg@gridforum.org Subject: RE: [dfdl-wg] simple way to study hard DFDL example problem - IBMFormat VS rec ords as XML

I wrote my previous mail fairly quickly just before I left on Friday to get something on the table. I've been thinking about this problem over the weekend and have some more thoughts which might help me get across where I am coming from.

The way I view physical rep information is as functions that can be applied to types and fields. Writing the data out to a blocked/segmented format does not fall into this category. It is an orthogonal operation that applies to the whole data and as such is much more akin to encryption and compression. For example, I have a COBOL structure that ends up in an MQSeries queue and in a QSAM file. It has a logical structure, it has a physical representation. In the QSAM case a further transform has taken place to block/segment the structure. I would not expect to see the physical rep properties of the types and elements change.

Mike's idea of a schema level 'stream' rep property sounds ok in principle for parsing, but what other metadata is needed when serialising? How are we informed of the rules for VB blocking or for IMS segmentation? Are they fixed or user-defined? If these rules end up requiring extra metadata at the type/element level then I am not comfortable with this, because we are mixing two sets of physical information.

I think that whatever principles we apply to DFDL including/excluding encryption and compression we should also apply to these formats. What is the current proposal in this area? The cheapest option would be to provide a flexible user-defined transform capability.

We can discuss more on this week's call, but it sounds like this is another of the high-level design issues to be included in the F2F agenda.

Finally a correction. When I said that the broker does not support these 19 or whatever formats, I should have been more specific and said that the broker's message model does not support these. That is, we do not provide physical rep annotation support for such formats, for the reason stated above. The expectation is that is that the decryption/decompression/deblocking has all taken place as a separate transformation elsewhere in the broker.

Regards, Steve

Steve Hanson WebSphere Business Integration Brokers, IBM Hursley, England Internet: smh@uk.ibm.com Phone (+44)/(0) 1962-815848

"Myers, James D"

<jim.myers@pnl.go

v> To Sent by: dfdl-wg@gridforum.org

owner-dfdl-wg@ggf cc .org

Subject RE: [dfdl-wg] simple way to study 19/11/2004 17:04 hard DFDL example problem - IBMFormat VS rec ords as XML

I think we at least agree in practice that there's a limit on how complex a transform you'd want to code in DFDL logic. Not sure if we agree on whether it is possible.

As for LR parsers - I'm not a parser guy, but I just looked at the wikipedia entry :-) :

Seems like a simple enough concept - if you let me have layers, and I can use information in those layers to select choices for further processing, can you stop me from making an LR parser (or doing what an LR parser does)? I've got a stack, and choices let me specify an action table... In the same way that if you give me layers (or variables), addition, and for loops, you can't stop me from doing multiplication. And if you require those things for other reasons but don't need multiplication, you can't really talk about excluding multiplication from the language design. You can say that we won't worry about multiplcation examples or how easy it is to write them down or what performance you'll get trying to run them and suggest that you plug something in to handle them directly though, and this is probably what we need to do in DFDL.

I may still be missing something and there is a piece of functionality that we haven't identified a need for that would be needed for an LR parser/our pathological examples, but I guess I'm getting more convinced that our primitives are sufficiently powerful that they can be used/abused to do all of the complex things that have come up. I'm not sure how we can close the issue - specify the map from DFDL primitives to LR parser as I started to above, or find an example known to require LR parsing and work it? Or?

Jim

...

-----Original Message----- From: owner-dfdl-wg@ggf.org [mailto:owner-dfdl-wg@ggf.org] On Behalf Of mike.beckerle@ascentialsoftware.com Sent: Friday, November 19, 2004 11:36 AM To: smh@uk.ibm.com; dfdl-wg@gridforum.org Subject: RE: [dfdl-wg] simple way to study hard DFDL example problem - IBMFormat VS rec ords as XML

I believe you and Jim are actually disagreeing. Jim is saying he's still optimistic that this transformation, even though complex, can be expressed directly in DFDL. You are saying this would require XSLT or a Java program or whatever to do it.

...

Mike you say you are aware of 19 such legacy formats, and I bet there are more. Well IBM's broker has no specific support

for any of

...

these, nor have we been asked to incorporate them into our message model. Maybe we should play the percentages game - if we see enough different subsystems that use the same cryptic format then it becomes worth building the support into DFDL.

Ascential supports 6 or 7 of these formats today. Batch systems will encounter this more than online. You get them when a mainframe job writes out a tape on a mainframe, and then you read that tape on a unix tape drive either directly or first into a file. Alternatively, you pick up a mainframe file via FTP or some such and directly operate on it on other systems. Mainframe software handles all the VS block and and such stuff in the lower layers as you know (not to mention the tape label) unix software does none of this, you just get the raw bytes.

My point is not as much about these 19 or more particular formats, but the issue of how much complexity we go after.

In the past we've looked at things like logical arrays with run-length-encoded representations and the suggestion has been there that DFDL might be able to directly express this transformation without need to go outside DFDL.

I've come to believe there are certain limits to this complexity and I think perhaps tree-shape compatibility is at the core of them. Building a DFDL description for data that ultimately requires an LR(k) sophistication parser to correctly interpret the data is clearly a non-starter it seems. Where this line is drawn is important.

...mikeb

...mikeb