I find myself with a very complicated problem to solve, a problem I suspect many companies have run into in the past. It is going to take me a very long time to solve it, and whilst I am doing so, it occurred to me to share the process on SCN as that way I can hopefully
- provide guidance to the less experienced programmers as to how to solve assorted problems.
- get some guidance from the more experienced programmers as to whether my proposed solutions are a good way of solving the problem
I am going to try to do this in the belief that this is what SCN is all about - the giving and receiving of knowledge, with everyone winning in the end.
The Australian part of my company went live with SAP in the year 2000. Our solution incorporated a large amount of ABAP programs which served as "front ends" to the business users, hiding the underlying complexity of SAP.
Since that time we have added an enormous amount of extra functionality to these programs.
In 2008 the company got taken over, and a decision was made by our new owners to migrate a large chunk of these programs into their existing SAP system, very flattering in a way. Initially only one country would make use of these programs - the United Kingdom - but in the long term all countries in the group (circa 70) would be using them.
So, for the last three years I have been working firstly in the United Kingdom (i.e. where the target users are) and then in Germany (i.e. where the head office is, and thus the development team) to copy over the programs from the Australian system and modify them for use in the UK and beyond.
All is now well, the UK is live, and I am currently en route back to Australia. The good news is that the UK users are happy bunnies, and head office is merrily planning the next rolls out in the USA, India, the rest of Europe etc..
However, being human, I am never satisfied.
The problems that I see looming are as follows:-
- Currently there are two development systems - one in Australia, one for the rest of the group. One day - it could be five years away, but the day will definitely dawn - the desire from above will be to merge the two development systems
- So, as the moment there are two versions of the programs, one in each system. Huge efforts have been made to keep them aligned, but the current feeling is that this is more or less impossible. The divergence will only increase over time.
- When the next countries come on line they will want some changes made for "legal reasons" (which usually translates as "there is no such law, but we have always done it this way"). So "some changes" translates to "an enormous amount of changes per country".
- What I want to avoid at all costs is to have 70 different versions of the program, which all contain the same ten thousand lines of identical code, plus ten thousand lines of country specific code.
Going back to how such a divergence can occur in the first place, some people might say "that's easy to avoid - whenever you make a change in one system, make the same change in the other". That is easy technically, but not politically. The copied program is working fine in a live environment. The target program where loads of changes were being made was not in a live environment. The country with the live environment was very reluctant to have any changes made that might risk the productive system, especially as there was no gain except the nebulous "it is the right thing to do technically" statement. Now we have two live environments, so both sides feel the same way.
As a side issue, when we went live in the year 2000 we were on version 4.5 of SAP, and we did not really take advantage of any of the new technology when we went to 4.7 so a lot of the programs would be deemed "old fashioned" by a lot of the SCN community e.g. very little object orientation as ABAP objects just was not there in version 4.5.
Australia has only just upgraded to ECC6.0 EHP5 from 4.7 and head office is upgrading to the same level at the end of the year. So, when I back in Australia next month it will be my first time programming in an ECC6.0 system, apart from some very minor dabbling in PI.
I have kept up with all the latest information on SDN, attended conferences and inside tracks, and read assorted SAP textbooks about all the new technologies that have come in since 4.7 but I have never had the chance to use any of them for real. This makes me feel somewhat "left behind".
So, when I get back to Australia I will of course have my real job to do, but as a hobby I am going to re-engineer an existing program - a great big incredibly complicated monolithic one. I want to use the :"think big, start small" approach, starting really simple and adding functionality one small bit at a time. As I mentioned earlier there is no time pressure on this so I have the luxury of being able to try out lots of thins in an attempt to get it "done right:, as opposed to "done fast" often a horrible trade off one has to make.
I could have chosen a simple ALV report, but where is the fun it that? I wanted a challenge so i wanted to pick the most complicated thing I can find. I wanted a "Venus De Milo".
Venus De Milo
There is a famous IT proverb about how when a programmer first completes a program it is a thing of beauty, just like the Venus De Milo. The architecture is simple and elegant, it does everything the user wants, everyone is happy. Then as the years go by, extra requirements come in the door one after the other, lots of different programmers work upon your masterpiece, there is never enough time to look at the "big picture" and so workarounds are thrown into the pot, and before you know it the Venus De Milo now has twenty three arms and a Giraffes head sticking out of her chest.
That's the sort of thing I wanted, so I could break it down and totally re-assemble it a la the film "The Fly" and it was easy enough to find an example in our system. The program I have chosen was created in 1997 and has been worked on by dozens of people ever since, and under the hood looks like nothing on Earth. When it was copied to the new system and the German developers took a look, to say they were puzzled was an understatement.
"Why did you do it like that?"
The answer could be, for example, that a certain section of code was done in a certain way to accommodate a complex business requirement which in the end turned out not to be needed. Nothing was ever documented - naturally - and nobody had the time to change the code back, so now we have complicated IF / THEN / ELSE blocks checking for things which once were presumed might occur but actually never did and never will.
So, in conclusion, that is what I am going to do. Shove a very old monolithic program in the matter transmitter, break it down into it's components, and re-assemble it using the latest technology and a design based on OO as opposed to procedural. The only difference is that this time I am starting with something that looks like a human body but with a fly's head on top and a giraffes head sticking out the middle, and hoping the end result is something that looks a bit better.
In an excellent book by Thomas Jung he creates an imaginary situation where a programmer finds himself having to use all the features that have just been added to SAP Netweaver. In a similar way I have made a great big list of things that were not available in SAP back in 1997 and made a provisional mapping to parts of my chosen application.
One small section at a time I will present one or more proposal of how I think I should implement a particular functionality, focusing on a "new" technology and I am hoping to get some feedback saying "you are made - you should do it this way, and here is why" or - not very likely - "yes, you are right" or - best case - a huge debate will ensue online.
Aims for the exercise
Aims - Personal
- To become familiar - in reality as opposed to in theory - to the new features in ABAP introduced after 4.7
- To hopefully provide guidance to less experienced programmers on the SCN as to how to solve assorted problem
- To hopefully receive some guidance from more experienced programmers on the SCN as to whether my proposed solutions are a good way of solving the problem, or the ravings of a madman
- To have made enough progress and got enough feedback by August of this year to be able to give a talk about this concept at the annual SAP conference in Australia
Aims - Application
- To ensure the same program can be run in different countries and respond differently according to local business needs, as opposed to creating 70 half-clones
- To separate the "things that change" from the "things that stay the same". Does that sound familiar?
- To cater for the fact that ten different people are likely to be working on different parts of this program, all at the same time. Forever.
Aims - Technical
- To unit test enable what is clearly legacy code
- To move from a monolithic to an object orientated architecture
- To eventually create both running versions of the program with the new one, with no loss of functionality. Difficult technically, maybe impossible politically, but I can but try.
- As part of the above to be able to prove there is no loss of functionality by a combination of unit tests and HPQC tests (we have a full licence for that, which is a good thing).
The End is Nigh
I will end with two examples of some proposed new ways to achieve a given function in the current application with a method that is better - i.e. one or more of the following
- more readable
- more robust
- more maintainable
- more extendable
Doing a goods issue for a delivery is a technical step and you would expect it to be the same worldwide. Printing a ticket for the customer though, you would expect that printed output to look different in every single country. For example, from looking at what my German colleagues were working on recently some Eastern European countries you have to print out fifty four million different legal paper forms for each delivery, whilst some countries are moving to accept digital documents, where the customer signs on an IPAD or something. Anyway, the point is every country is different.
I thought about this, and I would propose a solution as follows:-
Then the core "model" part of the application can say:-
LO_DELIVERY = ZCL_DELIVERY=>GET_BUSINESS_OBJECT( LD_VBELN ).
The correct sub-class will then be returned based using LIKP-VKORG as a starting point to find out the country. I've already got this working in an actual SAP system.
Then, later, when the core model program says
The correct ticket format gets printed out, even though the method is implemented totally differently in each country. The application model does not need to be changed when you add a new country or change an existing one.
This seemed like a logical solution to me - however the question I would ask is - have I missed something totally obvious? Does the solution I propose make any sense at all, or is there a clearly superior solution which is staring me in the face?
In my application the data in the main internal table is always the same, but the ORDER in which the fields appear in an ALV output, and indeed which fields are suppressed, varies according to the nature of about three or four fields which get input by the user on the selection screen - it would be really confusing if I explained the exact details of how the decision is made, in essence it is to do with the product and what the user is trying to do and it is SO complicated - but how to go about setting up the ALV field catalogue?
In the application as it is now, it works fine, but when one new developer saw how it worked - ten years ago - he was physically sick. It is basically a never ending series of IF / ELSEIF statements.
If (complicated set of conditions)
PERFORM add_field USING 'LIKP' "VBELN' hotspot new_length etc
PERFORM add_field USING something else etc
dozens more lines
ELSEIF (another complicated set of decision)
the same sort of thing, with different fields in different order
As there are loads of fields in every case and one line of code to add each field, the code goes on for ever and is totally unreadable. There has to be a better way. You might wonder why the common fields are not added first before the conditional processing, but the problem is that the display order is totally different in every case. Sometimes the key field is even different. Naturally the display order is also changing all the time. What was vital in 2007 is redundant in 2010 etc. Oh, and in Australia "they" don;t want users to be able to store user-specific variants (though in the UK that is a requirement).
I thought of three ways to do this without a never ending string of IF statements:-
- Define a sub-class for each situation, which redefines a method to say what fields need to be added
- Create one or more Z tables, with key fields based on the fields that influence the decision, and then store the field order
- As the rules are quite complex, I was wondering if BRF plus could be appropriate in such a situation? I have been reading the SAP press book on the subject every lunchtime in the beer garden near work in Germany for the last few weeks, and the "decision table" option described sounds like a possible fit.
So, once again I would ask - are any of these a logical way forward, and if so which one is the best? Or is there a much better, much more obvious way?
That's all Folks
In my next blog I will start with what has already been done so far in order to keep programs in line between the two SAP systems despite the many differences between customising, business process etc, and then move on to comparison between the way certain functionality is implemented currently and some proposed new ways.
I wait with some trepidation for any feedback, I have found in the past it can get quite nasty!
Cheersy Cheers from Moscow,
Subsequent blogs in this series:-