I’ve been in doubt if I should write this post or not. Read it fast as it may disapear if I regret writing it.
Ever since I started working with Navision, almost 25 years ago, I’ve had my own little stuborn ideas. These ideas got me where I am today, but just as often they got me in big trouble and caused personal relationships to shatter right before my eyes.
I wrote a lot about the legacy of Navision behind Business Central, the good, the bad and the ugly.
Today I want to talk about events, and why they are bad.
Events are bad? But… events are the whole backbone of the extensibility model. How can they be bad.
In order to understand that we first need to talk about Interfaces and Extensible Enums.
Progress in any development language or software framework can cause what was good yesterday to be something to avoid tomorrow, or even today.
Vedbaek, a few years ago…
Let’s rewind the clock a few years. Imagine C/Side with Hooks and Delta Files.
If this does not ring a bell, you are probably not old enough to understand this part of the blog and you can skip to the next paragraph.
A few years ago, and I’ve written many articles about this, Microsoft choose one of the SMB ERP Systems to go to the cloud. They only wanted to invest in one, not in three. Dynamics NAV was the choosen one.
The cloud needed a more mature extensibility model, and NAV had Delta Files and Hooks. This was choosen as the bases for the extension model we have today.
Part of this model was built in C/Side, which ended up being what we now know as “events”. Other parts were built outside C/Side and are what we now know as Table Extensions and Page Extensions. The first version did not offer an IDE for these objects and were tidous to work with.
What happened after that is history. The model grew into a new compiler that is compatible with Visual Studio Code and half a million events were added to a 30 year old application.
1.800 Apps in Microsoft AppSource are built on this model and used everyday.
So why is that bad?
It’s not bad, per se. But it is very tidious and it makes the framework very difficult to work with for junior developers.
Finding your way in thousands and thousands of events require very thourough knowledge of the old Navision application. Since there are only “few” of those it puts high constraignts on the growth of our ecosystem and make salaries for experienced developers go skyrocket.
Events have in between each other no relationship whatsoever. A few weeks ago I was talking to a friend who tried to enhance the Item Tracking module and he had to subscribe to 30+ events accross the system to complete a task.
In another case I was consulting a group of freelancers. They complained that they could never go to AppSource because they had heavily customized the Sales and Purchase posting processes.
My response, as a joke, was that Microsoft has built in the Generic Method pattern to override these posting routines with your own copy. The reason for making it a joke is that I thought (naive a girl as I am) that no sane developer would ever consider doing this.
Their response, to my surprise, was just a “thank you for this great suggestion, we will implement this”.
A third real life story, was a small consultation I did for a partner in Germany that offers a payment add-on that is very succesful. They are on AppSource and started to find out that aparently their App is not compatible with all of the other Apps. In other words, other Apps break their solution.
The reason for this is the “Handled Pattern” which is part of the Generic Method pattern which by itself is also an Anti Pattern but the only solution we had until the introduction of interfaces.
If two Apps subscribe to the same event and one handles it before the other get’s chance… the system fails.
And when someone decides to “override” posting routines the events in the original code are skipped.
Interfaces to the rescue
In my humble opinion, events should be marked as “obsolete pending” in favour of interfaces.
For Example: Sales Post
In Business Central, a Sales Document can have a few types such as quote, order, invoice or credit memo. There are a few more and partners can add new ones.
In my opinion a Sales Document Type should have methods that are implemented as part of an interface, such as “Release”, “Calculate Discount”, “Print” or whatever. Anything that is an action on a page.
If a partner really wants to override how the system works (which is bad enough to start with) they are then required to make their own Document Types. This shows a clear intention that they want the system to behave differently and it also allows other apps to auto-detect if they are compatible with this new implementation.
A Payment System, like the German one, should also replace the Payment System from Microsoft if they think they can do a better job.
Someone making a new Sales Document Type can still call the orriginal Payment Interface in the right places and allow other Payment systems to run nontheless.
Keep on dreaming Marije
A girl can dream right? I fully understand that the above situation will never happen.
Business Central was built on Navision and is it’s own legacy system and events, once our favorite, is now something from the past that is replaced with a better alternative.
Microsoft can never replace the events in the Base App with proper interfaces. The code is simply to old and events are all over the place.
Another problem is that an event publisher is married into it’s caller object. I remember in the very first discussions I had with Thomas Hejlsberg I suggested that an event should be able to move around when refactoring requires without breaking it’s subscribers. Unfortunately this never got implemented.
What about ISV’s?
Microsoft is always ahead of the game when compared to ISV’s. In the last releases of PrintVis we released a total of four interfaces that all serve a functional purpose. If a user or a partner of PrintVis is unhappy with how the interface behaves, they can imlement their own version.
If you have read my thoughts on best practices for Per Tenant Extensions you should have also seen that I don’t recommend anyone other than Microsoft or an ISV to work with Events, Enums or Interfaces.
If I were to do a code review of a Per Tenant Extension for an end-user and I would find any of these three I would put it into my report as “bad unless you have a damn good reason”.
This makes both this blog post and the (Anti) pattern a waste of time I guess.
Back to real life…
I think it’s important to think about the interests of the developers making different choices to understand their motivations.
If you are a narrow ISV selling a product that will maybe only ever be used OnPrem and only modified at most by 1 partner for each customer (and this might even be yourself, acting as VAR as well), then being able to overwrite every single method can be profitable when you are chasing the short term $$$$ because you can cater to all customer needs.
The same way Excel or PowerApps can be profitable for narrow scopes but you would never use them for any robust and scaling software architectures.
So the reason the pattern is so frustrating when Microsoft does it, inside the BaseApp, is that they are NOT building a platform for 1 partner to modify (at least officially). So they should spend more brain power on designing actual good APIs, that allow some extensibility without sacrificing robustness between multiple development partners extending the same parts.
That is much much harder than just following an assembly-line approach of “every function is overrideable, there are no rules” but it should be expected from platform builders.
If you think about the above, you can understand the motivations of both approaches for profit:
The platform builder profits the most long term from a stable platform, and should not touch this pattern at all with a ten-foot pole.
The customer-specific ISV/partner profits the most from constantly addressing short term needs, and they might be able to use this pattern successfully for their day-to-day needs.
Events is something I both love and hate. I love it for making it possible to extend the basic functionality.
But I also hate them so much! In my mind then they almost destroy our efforts to make clean code! Code utilizing events is not clean code, as you cannot see what it actually does just from reading the code. Only way to figure out what is going on is really to debug.
LikeLiked by 1 person