Language Design: Complexity, Extensibility, and Intention

Introduction

The object-oriented approach to software is great, and that greatness draws from the power of extensibility.  That we can create our own types, our own abstractions, has opened up worlds of possibilities.  System design is largely focused on this element of development: observing and repeating object-oriented patterns, analyzing their qualities, and adding to our mental toolbox the ones that serve us best.  We also focus on collecting libraries and controls because they encapsulate the patterns we need.

This article explores computer languages as a human-machine interface, the purpose and efficacy of languages, complexity of syntactic structure, and the connection between human and computer languages.  The Archetype project is an on-going effort to incorporate these ideas into language design.  In the same way that some furniture is designed ergonomically, Archetype is an attempt to design a powerful programming language with an ergonomic focus; in other words, with the human element always in mind.

Programming Language as Human-Machine Interface

A programming language is the interface between the human mind and executable code.  The point isn’t to turn human programmers into pure mathematical or machine thinkers, but to leverage the talent that people are born with to manipulate abstract symbols in language.  There is an elite class of computer language experts who have trained themselves to think in terms of purely functional approaches, low-level assembly instructions, or regular, monotonous expression structures—and this is necessary for researchers pushing themselves to understand ever more—but for the every day developer, a more practical approach is required.

Archetype is a series of experiments to build the perfect bridge between the human mind and synthetic computation.  As such, it is based as much as possible on a small core of extensible syntax and maintains a uniformity of expression within each facet of syntax that the human mind can easily keep separate.  At the same time, it honors syntactic variety and is being designed to shift us closer to a balance where all of the elements, blocks, clauses and operation types in a language can be extended or modified equally.  These represent the two most important design tenets of Archetype: the intuitive, natural connection to the human mind, and the maximization of its expressive power.

These forces often seem at odds with each other—at first glance seemingly impossible to resolve—and yet experience has shown that the languages we use are limited in ways we’re often surprised by, indicating that processes such as analogical extension are at work in our minds but not fully leveraged by those languages.

Syntactic Complexity & Extensibility

Most of a programming language’s syntax is highly static, and just a few areas (such as types, members, and sometimes operators) can be extended.  Lisp is the most famous example of a highly extensible language with support for macros which allow the developer to manipulate code as if it were data, and to extend the language to encode data in the form of state machines.  The highly regular, parenthesized syntax is very simple to parse and therefore to extend… so long as you don’t deviate from the parenthesized form.  Therefore Lisp gets away with powerful extensibility at the cost of artificially limiting its structural syntax.

In Lisp we write (+ 4 5) to add two numbers, or (foo 1 2) to call a function with two parameters.  Very uniform.  In C we write 4 + 5 because the infix operator is what we grew up seeing in school, and we vary the syntax for calling the function foo(1, 2) to provide visual cues to the viewer’s brain that the function is qualitatively something different from a basic math operation, and that its name is somehow different from its parameters.

Think about syntax features as visual manifestations of the abstract logical concepts that provide the foundation for all algorithmic expression.  A rich set of fundamental operations can be obscured by a monotony of syntax or confused by a poorly chosen syntactic style.  Archetype involves a lot of research in finding the best features across many existing languages, and exploring the limits, benefits, problems, and other details of each feature and syntactic representation of it.

Syntactic complexity provides greater flexibility, and wider channels with which to convey intent.  This is why people color code file folders and add graphic icons to public signage.  More cues enable faster recognition.  It’s possible to push complexity too far, of course, but we often underestimate what our minds are capable of when augmented by a system of external cues which is carefully designed and supported by good tools.

Imagine if your natural spoken language followed such simple and regular rules as Lisp: although everyone would learn to read and write easily, conversation would be monotonous.  Extend this to semantics, for example with a constructed spoken language like Lojban which is logically pure and provably unambiguous, and it becomes obvious that our human minds aren’t well suited to communicating this way.

Now consider a language like C with its 15 levels of operator precedence which were designed to match programmers’ expectations (although the authors admitted to getting some of this “wrong”, which further proves the point).  This language has given rise to very popular derivatives (C++, C#, Java) and are all easily learned, despite their syntactic complexity.

Natural languages and old world cities have grown with civilization organically, creating winding roads and wonderful linguistic variation.  These complicated structures have been etched into our collective unconscious, stirring within us and giving rise to awareness, thought, and creativity.  Although computers are excellent at processing regular, predictable patterns, it’s the complex interplay of external forces and inner voices that we’re most comfortable with.

Risk, Challenge & Opportunity

There are always trade-offs.  By focusing almost all extensibility in one or two small parts of a language, semantic analysis and code improvement optimizations are easier to develop and faster to execute.  Making other syntactical constructs extensible, if one isn’t careful, can create complexity that quickly spirals out of control, resulting in unverifiable, unpredictable and unsafe logic.

The way this is being managed in Archetype so far isn’t to allow any piece of the syntax tree to be modified, but rather to design regions of syntax with extensibility points built-in.  Outputting C# code as an intermediary (for now) lays a lot of burden on the C# compiler to ensure safety.  It’s also possible to mitigate more computationally expensive semantic analysis and code generation by taking advantage of both multicore and cloud-based processing.  What helps keep things in check is that potential extensibility points are being considered in the context of specific code scenarios and desired outcomes, based on over 25 years of real-world experience, not a disconnected sense of language purity or design ideals.

Creating a language that caters to the irregular texture of thought, while supporting a system of extensions that are both useful and safe, is not a trivial undertaking, but at the same time holds the greatest potential.  The more that computers can accommodate people instead of forcing people to make the effort to cater to machines, the better.  At least to the extent that it enables us to specify our designs unambiguously, which is somewhat unnatural for the human mind and will always require some training.

Summary

So much of the code we write is driven by a set of rituals that, while they achieve their purpose, often beg to be abstracted further away.  Even when good object models exist, they often require intricate or tedious participation to apply (see INotifyPropertyChanged).  Having the ability to incorporate the most common and solid of those patterns into language syntax (or extensions which appear to modify the language) is the ultimate mechanism for abstraction, and goes furthest in minimizing development effort.  By obviating the need to write convoluted yet routine boilerplate code, Archetype aims to filter out the noise and bring one’s intent more clearly into focus.

Problems with RIA Services (Feedback for July 2009 CTP)

RIA Services (new home page) is a collection of tools and libraries for making Rich Internet Applications, especially line of business applications, easier to develop.  Brad Abrams did a great presentation of RIA Services at MIX 2009 that touches on querying, validation, authentication, and how to share logic between the server and client sides.  Brad also has a huge series of articles (26 as I write this) on using Silverlight and RIA Services to build a realistic application.

I love the concept of RIA Services.  Brad and his team have done a fantastic job of identifying the critical issues for LOB systems and have the right idea to simplify those common data access tasks through the whole pipeline from database to UI controls, using libraries, Visual Studio tooling, or whatever it takes to get the job done.

So before I lay down some heavy criticism of RIA Services, take into consideration that it’s still a CTP and that my scenario pushes the boundaries of what was likely conceived of for this product, at least for such an early stage.

Shared Data Model with WPF & Silverlight Clients

The cause of so much of my grief with RIA Services has been my need to share a data model, and access to a shared database, across WPF as well as Silverlight client applications.  Within the constraints of this situation, I keep running into problem after problem while trying to use RIA Services productively.

The intuitive thing to do is: define a single data model project that compiles to a single assembly, and then reference that in my Silverlight and non-Silverlight projects.  This would be a 100% full-fidelity shared data model.  As long as the code I wrote was a subset of both Silverlight and normal .NET Frameworks (an intersection), we could share identical types and write complex validation and model manipulation logic, all without having to constrain ourselves to work within the limitations of a convoluted code generation scheme.  Back when I wrote Compact Framework applications, I did this with great success despite the platform gap, and I didn’t have anything like RIA Services to help.

Incompatible Assemblies

Part of the problem arises because Silverlight assemblies are incompatible with non-Silverlight assemblies.  A lot of what RIA Services is doing is trying to find a way around this limitation: picking up attributes and code files from one project and inserting that code into the Silverlight project with a build action.  This Visual Studio “magic” has been criticized for its weakness in dealing with multiple-solution systems where Visual Studio can’t update the client because it’s not loaded, and I’ve heard there’s work being done to address this, but for my current needs, this magic aspect of it isn’t a problem.  The specifics of how it works, however, are.

Different Data Access APIs

Accessing entities requires a different API in Silverlight via RiaContextBase versus ObjectContext elsewhere.  Complex logic in the model (for validation and other actions against the model) requires access to other entities and therefore access to the current object context, but the context APIs for Silverlight and WPF are very different.  Part of this has to do with Silverlight’s inability to make synchronous calls to the server.

In significantly large systems that I build, I use validation logic such as “this entity is valid if it’s pointing to an entity of a different type that contains a PropertyX value of Y”.  One of my tables stores a tree of data, so I have methods for loading entire subtrees and ensuring that no circular references exist.  For these kinds of tasks, I need access to the data context in basic validation methods.  When I delete nodes from a tree, I need to delete child nodes, so update logic is part of the model that needs to be the same in every client.  I don’t want to define that multiple times for multiple clients.  I like to program very DRY.  In other words, I find myself in need of a shared model.

RIA Services doesn’t provide anything like type equivalence for a shared model, however.  Data model classes in Silverlight inherit from Entity, but EntityObject in WPF.  In the RIA Services domain context, we RaiseDataMemberChanged, but in a normal EF object context, we need to ReportPropertyChanged.  In WPF, I can call MyEntity.Load(MergeOption.PreserveChanges), but in Silverlight there’s no Load method on the entity and no MergeOption enum.  In WPF I can query against context.SomeEntitySet, but in Silverlight you would query against context.GetSomeEntitySetQuery() and then execute the query with another method call.

This chasm of disparity makes all but the simplest shared model logic impractical and frustrating.  The code generation technique, though good in principle, keeps getting in the way.  For example, I have both parameterless and parameterized constructors in my entity classes.  This works great in my WPF client, but when this code is synchronized to my Silverlight client, I get an error because the Silverlight-side entity class is generated in two parts: in the hidden partial class, a parameterless constructor is generated which calls partial method OnCreated; and in the visible partial class, the constructor method I defined on the server is dumped into another file, so I have duplicate constructors.  If I remove the parameterless constructor from the server side, I get an error because my entity class requires a parameterless constructor (and defining a non-default constructor effectively removes the default one from the resulting type unless it’s explicitly defined).  I thought I could define the partial method OnCreated and put my construction logic in there, but the partial method is only defined on the client side.  That means sharing construction logic consists of copying and pasting the OnCreated method across the various clients—far from an ideal solution.

Entity Data Model Required to be in Web Project

Another strategy I attempted was to define the .edmx file and my partial class extensions in a class library, and then reference that from the web project.  I could define the LinqToEntitiesDomainService<MyDataContext>, but sharing entity class code (by generating code in the Silverlight project) isn’t possible unless the .edmx file and partial class extensions are defined in the web project itself.  This would mean that my WPF client would have to reference a web project for data access, which by itself seems wrong.  (Or making a copy of the data model, which is worse.)  It would be better for the WPF client to talk to the same domain service as the Silverlight client, but RIA Services doesn’t give you an option to link that web project to a non-Silverlight project, so again I ran into a brick wall.

So Don’t Do That

The kind of advice I’m getting for this is, “so don’t do that”.  In other words, don’t write complex validation logic in the model or otherwise try to access the data context; don’t write parameterized constructors; don’t aim for 100% type fidelity across all endpoints of a system; don’t try to share data models with Silverlight and non-Silverlight projects, etc.  But I see the potential for RIA Services, so I have to push for these things unless I hear really convincing arguments against them (or compelling alternatives).

Conclusion

The fact that there are different data contexts and data item definitions within those contexts imposes a burden on the developer to use different techniques for each environment, and creates challenges for centralizing data model logic and reusing equivalent logic across different kinds of clients.  My gut feeling is that RIA Services in its current form has some fundamental design flaws that will need to be addressed, taking into consideration systems with a mix of Silverlight, WPF, and other clients, before it becomes a truly robust data access platform.

Better Tool Support for .NET

Productivity Enhancing Tools

Visual Studio has come a long way since its debut in 2002.  With the imminent release of 2010, we’ll see a desperately-needed overhauling of the archaic COM extensibility mechanisms (to support the Managed Package Framework, as well as MEF, the Managed Extensibility Framework) and a redesign of the user interface in WPF that I’ve been pushing for and predicted as inevitable quite some time ago.

For many alpha geeks, the Visual Studio environment has been extended with excellent third-party, productivity-enhancing tools such as CodeRush and Resharper.  I personally feel that the Visual Studio IDE team has been slacking in this area, providing only very weak support for refactorings, code navigation, and better Intellisense.  While I understand their desire to avoid stepping on partners’ toes, this is one area I think makes sense for them to be deeply invested in.  In fact, I think a new charter for a Developer Productivity Team is warranted (or an expansion of their team if it already exists).

It’s unfortunately a minority of .NET developers who know about and use these third-party tools, and the .NET community as a whole would without a doubt be significantly more productive if these tools were installed in the IDE from day one.  It would also help to overcome resistance from development departments in larger organizations that are wary of third-party plug-ins, due perhaps to the unstable nature of many of them.  Microsoft should consider purchasing one or both of them, or paying a licensing fee to include them in every copy of Visual Studio.  Doing so, in my opinion, would make them heroes in the eyes of the overwhelming majority of .NET developers around the world.

It’s not that I mind paying a few hundred dollars for these tools.  Far from it!  The tools pay for themselves very quickly in time saved.  The point is to make them ubiquitous: to make high-productivity coding a standard of .NET development instead of a nice add-on that is only sometimes accepted.

Consider just from the perspective of watching speakers at conferences coding up samples.  How many of them don’t use such a tool in their demonstration simply because they don’t want to confuse their audience with an unfamiliar development interface?  How many more demonstrations could they be completing in the limited time they have available if they felt more comfortable using these tools in front of the masses?  You know you pay good money to attend these conferences.  Wouldn’t you like to cover significantly more ground while you’re there?  This is only likely to happen when the tool’s delivery vehicle is Visual Studio itself.  Damon Payne makes a similar case for the inclusion of the Managed Extensibility Framework in .NET Framework 4.0: build it into the core and people will accept it.

The Gorillas in the Room

CodeRush and Resharper have both received recent mention in the Hanselminutes podcast (episode 196 with Mark Miller) and in the Deep Fried Bytes podcast (episode 35 with Corey Haines).  If you haven’t heard of CodeRush, I recommend watching these videos on their use.

• CodeRush Express (which is free)
• CodeRush with Refactor!
• DXCore (the library you can use to build your own add-ins, and on which CodeRush and Refactor! are built)

For secondary information on CodeRush, DXCore, and the principles with which they were designed, I recommend these episodes of DotNetRocks:

• Tool Development
• Discoverability
• The Science of Good UI

I don’t mean to be so biased toward CodeRush, but this is the tool I’m personally familiar with, has a broader range of functionality, and it seems to get the majority of press coverage.  However, those who do talk about Resharper do speak highly of it, so I recommend you check out both of them to see which one works best for you.  But above all: go check them out!

Refactor – Rename

Refactoring code is something we should all be doing constantly to avoid the accumulation of technical debt as software projects and the requirements on which they are based evolve.  There are many refactorings in Visual Studio for C#, and many more in third-party tools for several languages, but I’m going to focus here on what I consider to be the most important refactoring of them all: Rename.

Why is Rename so important?  Because it’s so commonly used, and it has such far-reaching effects.  It is frequently the case that we give poor names to identifiers before we clearly understand their role in the “finished” system, and even more frequent that an item’s role changes as the software evolves.  Failure to rename items to accurately reflect their current purpose is a recipe for code rot and greater code maintenance costs, developer confusion, and therefore buggy logic (with its associated support costs).

When I rename an identifier with a refactoring tool, all of the references to that identifier are also updated.  There might be hundreds of references.  In the days before refactoring tools, one would accomplish this with Find-and-Replace, but this is dangerous.  Even with options like “match case” and “match whole word”, it’s easy to rename the wrong identifiers, rename pieces of string literals, and so on; and if you forget to set these options, it’s worse.  You can go through each change individually, but that can take a very long time with hundreds of potential updates and is a far cry from a truly intelligent update.

Ultimately, the intelligence of the Rename refactoring provides safety and confidence for making far-reaching changes, encouraging more aggressive refactoring practices on a more regular basis.

Abolishing Magic Strings

I am intensely passionate about any tool or coding practice that encourages refactoring and better code hygiene.  One example of such a coding practice is the use of lambda expressions to select identifiers instead of using evil “magical strings”.  From my article on dynamically sorting Linq queries, the use of “magic strings” would force me to write something like this to dynamically sort a Linq query:

Customers = Customers.Order("LastName").Order("FirstName", SortDirection.Descending);

The problem here is that “LastName” and “FirstName” are oblivious to the Rename refactoring.  Using the refactoring tool might give me a false sense of security in thinking that all of my references to those two fields have been renamed, leading me to The Pit of Despair.  Instead, I can define a function and use it like the following:

public static IOrderedEnumerable<T> Order<T>(this IEnumerable<T> Source, 
    Expression<Func<T, object>> Selector, SortDirection SortDirection)
{
    return Order(Source, (Selector.Body as MemberExpression).Member.Name, SortDirection);
}

Customers = Customers.Order(c => c.LastName).Order(c => c.FirstName, SortDirection.Descending);

This requires a little understanding of the structure of expressions to implement, but the benefit is huge: I can now use the refactoring tool with much greater confidence that I’m not introducing subtle reference bugs into my code.  For such a simple example, the benefit is dubious, but multiply this by hundreds or thousands of magic string references, and the effort involved in refactoring quickly becomes overwhelming.

Coding in this style is most valuable when it’s a solution-wide convention.  So long as you have code that strays from this design philosophy, you’ll find yourself grumbling and reaching for the inefficient and inelegant Find-and-Replace tool.  The only time it really becomes an issue, then, is when accessing libraries that you have no control over, such as the Linq-to-Entities and the Entity Framework, which makes extensive use of magic strings.  In the case of EF, this is mitigated somewhat by your ability to regenerate the code it uses.  In other libraries, it may be possible to write extension methods like the Order method shown above.

It’s my earnest hope that library and framework authors such as the .NET Framework team will seriously consider alternatives to, and an abolition of, “magic strings” and other coding practices that frustrate otherwise-powerful refactoring tools.

Refactoring Across Languages

A tool is only as valuable as it is practical.  The Rename refactoring is more valuable when coding practices don’t frustrate it, as explained above.  Another barrier to the practical use of this tool is the prevalence of multiple languages within and across projects in a Visual Studio solution.  The definition of a project as a single-language container is dubious when you consider that a C# or VB.NET project may also contain HTML, ASP.NET, XAML, or configuration XML markup.  These are all languages with their own parsers and other language services.

So what happens when identifiers are shared across languages and a Rename refactoring is executed?  It depends on the languages involved, unfortunately.

When refactoring a C# class in Visual Studio, the XAML’s x:Class value is also updated.  What we’re seeing here is cross-language refactoring, but unfortunately it only works in one direction.  There is no refactor command to update the x:Class value from the XAML editor, so manually changing it causes my C# class to become sadly out of sync.  Furthermore, this seems to be XAML specific.  If I refactor the name of an .aspx.cs class, the Inherits attribute of the Page directive in the .aspx file doesn’t update.

How frequent do you think it is that someone would want to change a code-behind file for an ASP.NET page, and yet would not want to change the Inherits attribute?  Probably not very common (okay, probably NEVER).  This is a matter of having sensible defaults.  When you change an identifier name in this way, the development environment does not respond in a sensible way by default, forcing the developer to do extra work and waste time.  This is a failure in UI design for the same reason that Intellisense has been such a resounding success: Intellisense anticipates our needs and works with us; the failure to keep identifiers in sync by default is diametrically opposed to this intelligence.  This represents a fragmented and inconsistent design for an IDE to possess, thus my hope that it will be addressed in the near future.

The problem should be recognized as systemic, however, and addressed in a generalized way.  Making individual improvements in the relationships between pairs of languages has been almost adequate, but I think it would behoove us to take a step back and take a look at the future family of languages supported by the IDE, and the circumstances that will quickly be upon us with Microsoft’s Oslo platform, which enables developers to more easily build tool-supported languages (especially DSLs, Domain Specific Languages). 

Even without Oslo, we have seen a proliferation of languages: IronRuby, IronPython, F#, and the list goes on.  A refactoring tool that is hard-coded for specific languages will be unable to keep pace with the growing family of .NET and markup languages, and certainly unable to deal with the demands of every DSL that emerges in the next few years.  If instead we had a way to identify our code identifiers to the refactoring tool, and indicate how they should be bound to identifiers in other languages in other files, or even other projects or solutions, the tools would be able to make some intelligent decisions without understanding each language ahead of time.  Each language’s language service could supply this information.  For more information on Microsoft Oslo and its relationship to a world of many languages, see my article on Why Oslo Is Important.

Without this cross-language identifier binding feature, we’ll remain in refactoring hell.  I offered a feature suggestion to the Oslo team regarding this multi-master synchronization of a model across languages that was rejected, much to my dismay.  I’m not sure if the Oslo team is the right group to address this, or if it’s more appropriate for the Visual Studio IDE team, so I’m not willing to give up on this yet.

A Default of Refactor-Rename

The next idea I’d like to propose here is that the Rename refactoring is, in fact, a sensible default behavior.  In other words, when I edit an identifier in my code, I more often than not want all of the references to that identifier to change as well.  This is based on my experience in invoking the refactoring explicitly countless times, compared to the relatively few times I want to “break away” that identifier from all the code that references.

Think about it: if you have 150 references to variable Foo, and you change Foo to FooBar, you’re going to have 150 broken references.  Are you going to create a new Foo variable to replace them?  That workflow doesn’t make any sense.  Why not just start editing the identifier and have the references update themselves implicitly?  If you want to be aware of the change, it would be trivial for the IDE to indicate the number of references that were updated behind the scenes.  Then, if for some reason you really did want to break the references, you could explicitly launch a refactoring tool to “break references”, allowing you to edit that identifier definition separately.

The challenge that comes to mind with this default behavior concerns code that spans across solutions that aren’t loaded into the IDE at the same time.  In principle, this could be dealt with by logging the refactoring somewhere accessible to all solutions involved, in a location they can all access and which gets checked into source control.  The next time the other solutions are loaded, the log is loaded and the identifiers are renamed as specified.

Language Property Paths

If you’ve done much development with Silverlight or WPF, you’ve probably run into the PropertyPath class when using data binding or animation.  PropertyPath objects represent a traversal path to a property such as “Company.CompanyName.Text”.  The travesty is that they’re always “magic strings”.

My argument is that the property path is such an important construct that it deserves to be an core part of language syntax instead of just a type in some UI-platform-specific library.  I created a data binding library for Windows Forms for which I created my own property path syntax and type, and there are countless non-UI scenarios in which this construct would also be incredibly useful.

The advantage of having a language like C# understand property path syntax is that you avoid a whole class of problems that developers have used “magic strings” to solve.  The compiler can then make intelligent decisions about the correctness of paths, and errors can be identified very early in the cycle.

Imagine being able to pass property paths to methods or return then from functions as first-class citizens.  Instead of writing this:

Binding NameTextBinding = new Binding("Name") { Source = customer1; }

… we could write something like this, have access to the Rename refactoring, and even get Intellisense support when hitting the dot (.) operator:

Binding NameTextBinding = new Binding(@Customer.Name) { Source = customer1; }

In this code example, I use the fictitious @ operator to inform the compiler that I’m specifying a property path and not trying to reference a static property called Name on the Customer class.

With property paths in the language, we could solve our dynamic Linq sort problem cleanly, without using lambda expressions to hack around the problem:

Customers = Customers.Order(@Customer.LastName).Order(@Customer.FirstName, SortDirection.Descending);

That looks and feels right to me.  How about you?

Summary

There are many factors of developer productivity, and I’ve established refactoring as one of them.  In this article I discussed tooling and coding practices that support or frustrate refactoring.  We took a deep look into the most important refactoring we have at our disposal, Rename, and examined how to get the greatest value out of it in terms of personal habits, as well as long-term tooling vision and language innovation.  I proposed including property paths in language syntax due to its general usefulness and its ability to solve a whole class of problems that have traditionally been solved using problematic “magic strings”.

It gives me hope to see the growing popularity of Fluent Interfaces and the use of lambda expressions to provide coding conventions that can be verified by the compiler, and a growing community of bloggers (such as here and here) writing about the abolition of “magic strings” in their code.  We can only hope that Microsoft program managers, architects, and developers on the Visual Studio and .NET Framework teams are listening.

Software Development Methods

Building a House

Building a house is still one of the most common analogies I hear and use for discussing software development with customers.  The problem is that the architecture of the house is left out; and unless you’re a developer, it’s hard to understand why a stack of wireframes isn’t the comprehensive description of the system needed to provide accurate cost and schedule estimates.

These wireframes are like the diagrams a first-time home owner might sketch on the back of a napkin.  In order to know what must be built, and to determine a realistic sense of scope and cost, those rough ideas need to be developed into something much more substantial by a professional.  For our house building scenario, this would involve hiring an architect to transform those ideas into a set of drawings which include such things as plumbing, heat and air conditioning ducts, electrical access panels, drains, the appropriate foundation and structural support elements, and so on.  The customer asks for a “nice room with a window looking to the back yard”, but all of these other hidden structural elements must be taken into consideration as well.

Unfortunately, determining the actual cost and size of a large project with any accuracy requires having this detailed design; and worse, doing this detailed design can take weeks or even months.  By the time development is underway, the customer will inevitably change many of the requirements anyway.  In fact, the very process of evolving such a large and detailed design brings many issues to the surface to discuss, debate, and make decisions about.  It’s easy to see how this process could carry on for quite a while, and there are still changes in the market to consider and adapt to, both during and after requirements definition.

The house building analogy is useful, but it is also misused.  To be fair, a typical waterfall software project put into house building terms would go like this.

1. Customer sends sketches to Architect and asks for plan including schedule and cost.
2. Architect works with Customer, being paid for his services to arrive at detailed design, which provides the project schedule and cost.
3. Builder works with the Architect’s plan to build exactly what’s in the requirements.
4. Days or weeks later, Customer makes changes to requirements.  This requires the Architect to get involved again, update the plan, work with the Builder to make physical adjustments, which may include tearing down existing structures and modifying others.  Plan changes need to be approved by government (verified with tests), documentation updated, and Customer agrees to pay for the changes.
5. * Customer requests to now connect their house to other houses near by, the shapes and layouts of which may change over time.
6. Repeat #4.

However, this is completely at odds with how homes are normally built.  People typically choose a previously-implemented design, and only customize superficial features like countertops, cabinets, floors, and railings.  Houses from this plan have been built before, and the labor and materials cost are known from previous experience.

Building software is usually more like constructing something that’s never been built before: the first sky scraper, the Golden Gate Bridge, or the Hoover Dam.  The requirements are unique, the pieces have never been assembled in such a way before, and there’s an inherent level of risk in creating something new.  When this is the case, the Customer needs the services of an Architect, not just a Builder who stamps out deliverables in a cookie-cutter style.

Waterfall methodologies have been dramatically in decline over the years in favor of so-called Agile methods.  There are several problems with a Waterfall approach:

• Because of requirements churn, attempting to write all the requirements up-front is incredibly wasteful.  Your team will end up doing a lot of design work that is eventually thrown away.  When I hear about churn numbers as high as 80%, I shudder to think of all the time and money spent refining details that will never make it into the production system.
• It tends to take a blind approach to requirements change, working under the assumption that change is bad and should be prevented.  Many waterfall methodologies erect barriers to change, making Customers jump through hoops to push changes through.
• Customers typically don’t want to pay for this architecture phase.  They want to pay for actual work on their system, and the cost of defining the actual scope and cost of their system seems steep.  “Why should I pay you $<large number> just for you to tell me how much you’ll charge me for the final system?”

An Agile Solution: Efficiency and Uncertainty

The Agile community emerged largely as a reaction to the disappointing results of Waterfall projects.  Customers were sick of schedules and budgets that were consistently and grossly overrun (despite all the up-front design), and wanted to more easily change requirements without being put through the wringer to do so.  Developers were tired of spending their valuable time writing detailed design documents that kept changing throughout the project, and the burden to update that documentation as the project goals changed focus or priorities shifted made development a very inefficient and slow process.

Essentially, as is often the case, a trade-off was made.  Some stakeholders realized that the predictability they were attempting to achieve by spelling out everything in detail was delusional: they were aiming at a moving target regardless of their efforts to pin things down.  Project costs rose precipitously because the scope of the system grew.  They needed a process that acknowledged and accepted these dynamic project forces instead of one that denied and suppressed them.  They needed a process that was efficient, and they gave up the illusion of certainty to get it.

The cone of uncertainty is one of the most valuable tools for visualizing and understanding uncertainty in development projects.  From that article:

Research has found that the accuracy of the software estimate depends on the level of refinement of the software’s definition. The more refined the definition, the more accurate the estimate. The reason the estimate contains variability is that the software project itself contains variability. The only way to reduce the variability in the estimate is to reduce the variability in the project itself.

An important—and difficult—concept is that the Cone of Uncertainty represents the best case accuracy it’s possible to have in software estimates at different points in a project. The Cone represents the error in estimates created by skilled estimators [which is up to 16x error at the beginning of a project]. It’s easily possible to do worse. It isn’t possible to be more accurate; it’s only possible to be more lucky.

Commitments made too early in a project undermine predictability, increase risk, increase project inefficiencies, and impair the ability to manage a project to a successful conclusion.

Meaningful commitments are not possible in the early, wide part of the Cone. Effective organizations delay their commitments until they have done the work to force the Cone to narrow. Meaningful commitments in the early-middle of the project (about 30% of the way into the project) are possible and appropriate.

Or as noted by the MSDN documentation for Composite Client Application Guidance:

Application requirements can change over time. New business opportunities and challenges may present themselves, new technologies may become available, or even ongoing customer feedback during the development cycle may significantly affect the requirements of the application. Therefore, it is important to build the application so that it is flexible and can be easily modified or extended over time.

Agile methods have always dealt with small iterations and early, frequent builds to evolve product functionality, but the Lean Development method especially has an explicit focus on making meaningful commitments at the latest possible responsible moment.  From Wikipedia:

As software development is always associated with some uncertainty, better results should be achieved with an options-based approach, delaying decisions as much as possible until they can be made based on facts and not on uncertain assumptions and predictions. The more complex a system is, the more capacity for change should be built into it, thus enabling the delay of important and crucial commitments. The iterative approach promotes this principle – the ability to adapt to changes and correct mistakes, which might be very costly if discovered after the release of the system.

While the methodology pundits debate the relative merits of development processes, Tom DeMarco, author of the fantastic books Peopleware and Controlling Software Projects, challenges us to consider the value of our projects and to re-evaluate elaborate control measures in light of the bigger picture.  From his article:

To understand control’s real role, you need to distinguish between two drastically different kinds of projects:

• Project A will eventually cost about a million dollars and produce value of around $1.1 million.
• Project B will eventually cost about a million dollars and produce value of more than $50 million.

What’s immediately apparent is that control is really important for Project A but almost not at all important for Project B.  This leads us to the odd conclusion that strict control is something that matters a lot on relatively useless projects and much less on useful projects.  It suggests that the more you focus on control, the more likely you’re working on a project that’s striving to deliver something of relatively minor value.

This adds additional credence to the idea of agile development, since the Big Design Up-Front (BDUF) of Waterfall is obsessed with detailed definitions to make accurate schedule promises.  If the business model is sound, lots of control isn’t needed.  It’s better to navigate with a good compass (the product vision) than an incorrect and constantly changing map.

So What Will I Get?

At the end of the day, customers want to know what they’re going to get.  There are tradeshow and conference dates to meet, beta and v1.0 launch targets to hit, release cycles to determine, and so on.  Ultimately, some up-front architecture and design is usually required.  But instead of trying to pin down every detail, it’s actually best in terms of efficiency to share the vision, provide direction, and define the technical scope only in a loose way.  This functional scope definition should reflect the reality of the cone of uncertainty at that point in the development process.

At the beginning of the project, items to include in the project should be broadly defined, having a granularity appropriate for the actual uncertainty.  As the project progresses and work is done, the scope becomes better defined, risks are determined benign or neutralized, and the cadence of that specific team working on that customer’s specific product can be measured.

This is successfully done with Lean and Kanban approaches, providing an empirical method for measuring productivity.  Planning can then be adjusted to match the regular rhythm of production that develops.  This creates a semblance of real predictability.  You can’t predict any one thing with absolute certainty, but if you can measure the general rhythm at which work is done, you will very quickly learn to recognize how long different kinds of tasks actually take.  Because you’re focusing on measured results rather than predictions, it’s easier to normalize task size and estimate larger functional units.

If you’re interested in learning more about Lean development, Mary and Tom Poppendieck’s book Implementing Lean Software Development is a fantastic background and explanation of the Lean process.  They also have a new book coming out this November, exploring the same topics more deeply.

When a good process is followed by hard-working, talented developers, the customer will end up with what they need to satisfy their business goals, often in surprising ways, ways that will be radically different from what they initially asked for.

What Can Be Promised?

The customer has their vision, beginning with the end in mind as they do; and they are asked to break it apart, prioritize individual features, and stretch their end-goal image across a long and expensive tunnel that separates them from their goal.  They usually want some kind of promise of what they’re going to get at the end, even though they’re going to keep adding and changing features throughout its development.

What can be promised?  It would be dishonest to promise customers what they’re asking for, unless they promise not to change their minds as the project progresses, but the only certainty in software development is that requirements will change.

Development isn’t about creating flawless, predictable plans that eliminate business risk.  Development is an unpredictable journey to create tremendous value.  To achieve great reward, risk is a necessary part of the equation.  Instead of coming up with creative ways to hide real project risk, it’s best to have a solid process for detecting and managing that risk.

Why Oslo is Important

Contrary to common misunderstanding and speculation, the point of Oslo is not to put programming in the hands of business analysts who want to write their own business rules.  Do I think some of that will happen?  Architects and engineers will try everything they can imagine.  Some of them will succeed in specific niches or scenarios, but it won’t replace application or system design, and it will probably be very limited for the forseeable future.  Oslo is more about dramatically improving the productivity of designers and developers by generalizing common solution patterns and generating more adaptable tools.

Much of the confusion around Oslo occurs for two reasons:

1. Oslo is designed at a higher level of abstraction than most systems today, so its scope is broad and it will have an impact on virtually every product, solution and service across Microsoft.  It’s difficult to get your head around something that big.
2. Because of its abstract nature, core concepts are defined in terms that are heavily overloaded, like "Model", "Repository", and "Language".  Once you’ve picked up the lingo and can translate Oslo terminology into language you’re already familiar with, both the concept and magnitude of it will become obvious.

Oslo isn’t something completely new; in fact, Oslo borrows from a lot of previous research and even existing model-driven development tools.  Oslo focuses existing technologies and techniques into a coherent and mature vision of development, combining all parts into a more powerful whole, and promises to deliver a supremely adaptable and efficient platform to develop on.

What Is Oslo?

Oslo is a software factory for generating first-class, tool-supported languages out of your declarative specifications.

A factory is a highly organized production facility
that produces members of a product line
using standardized parts, tools and production processes.

-from a review of Software Factories

The product line is analogous to Oslo’s parsers, transform tools, and IDE plugins for new data models and languages (both textual and visual) that you define.  The standardized parts are Oslo’s library components; the tools are the M languages and the Quadrant/Intellipad application; and the processes are shaped by the flow of data through the Oslo tool chain (see the diagram near the end of this article).

With Oslo, you build the custom tools you need to rapidly build or generate software systems.  It’s all about using the right tool for the job, and having a say in how those tools are shaped to obtain the greatest leverage.

As stated at the home page of softwarefactories.com:

We see a capacity crisis looming. The industry continues to hand-stitch applications distributed over multiple platforms housed by multiple businesses located around the planet, automating business processes like health insurance claim processing and international currency arbitrage, using strings, integers and line by line conditional logic. Most developers build every application as though it is the first of its kind anywhere.

In other words, there’s already a huge shortage of experienced, highly-qualified professionals capable of ensuring the success of these increasingly complex systems, and with the need (and complexity) growing exponentially, our current development practices increasingly fall short of the total demand.

Books like Greenfield’s Software Factories have been advocating building at a higher level of abstraction for years, and my initial reaction was to see it as a natural, evolutionary milestone for a highly mature software system.  However, it’s an awful lot of focused development effort to attain such a level of maturity, and not many organizations are able to pull it off given the state of our current development platforms.

It’s therefore fortuitous that Microsoft teams have taken up the challenge of building these abilities into their .NET platform.  After all, that’s where it really belongs: in the framework.

Unexpected Awesomeness

Oslo of course contains a lot of expected awesomeness, but where it will probably have the most impact in terms of developer productivity is with new first-class languages and language tools.  Why?  It first helps to understand the world of data formats and languages.

We’ve had an explosion of data formats–these mini Domain Specific Languages, if you will (especially in the form of complex configuration files).  As systems evolve and scale, and the ways we can configure and compose our application’s behavior continues to grow, at what point do we perceive that configuration graph as the rich language that it becomes?  Or when our user interfaces evolve from Monolithic to Modular to Composite to Granular Composite (or User Composable), at what point does that persistent object graph become our UX DSL (as with XAML in WPF).

Sometimes we set our standards too low, or are slow to raise them when the time has come to do so.  With XML we get extensibility in defining languages and we think, "If we can parse it, then we can build a tool over it."  I don’t know about you, but I’d much rather work with rich client software–some kind of designer–over a textual data format any day.

But you know how things go: some company like Microsoft builds a whole bunch of cool stuff, driven off some XML configuration, or they unleash something like XAML on which WPF, WF, and more are built.  XAML is great for tools to read and write, and although XML and XAML are textual and not binary and therefore human readable in a text editor (the original intention behind that term), it’s simply not as easy to read as C# or VB.NET.  That’s why we aren’t all rushing to program everything in XAML.

Companies like Microsoft, building from the bottom up, release their platforms well in advance of the thick client user experiences that make them enjoyable to use and which encourages mass adoption.  Their models, frameworks, and applications are so large now that they’re released in massively differentiated stages, producing a technology adoption gap.

By giving that language a syntax other than XML, however, we can approach it in the same way we approach our program logic: in the most human readable and aesthetically-pleasant way we can devise, resembling our programming languages of choice.

Sometimes, the density of data and its structure in our model is such that a visual editor fails to represent that model well.  Source code is a case in point.  You could create a visual designer to visualize flow control, branching logic, and even complex expression building (like the iTunes Smart Playlist), but code in text format is more appropriate in this kind of scenario, and ends up being more efficient with the existing tooling available.  Especially with an IDE like Visual Studio, we’re working with human-millenia of effort that have gone into the great code editing tools we use today.  Oslo respects this need for choice by offering support for building both visual and textual DSLs, and recognizes the fluent definition of new formats and languages as the bridge to the next quantum leap in productivity.

If we had an easy way of defining languages in formats that we developers felt comfortable working with–as we’re comfortable with our general purpose languages and their rich tool support–then we’d be much more productive in the transition between a technology first being released and later having rich tool support over it.  WPF has taken quite a while to be adopted as much as it has, partly due to tool availability and maturity.  Before Expression Blend or Cider designers were released and hand-coding XAML was the only way, those who braved the angle brackets struggled with it.  As I play with Silverlight, I realize how much must still be done in XAML, and how we still struggle.  It’s simply not as nice to work with as my C# code.  Not as rich, and not as strongly tool-supported.

That’s one place Oslo provides value.  With the ability to define new textual and visual DSLs, rigorous verification and validation in a rich set of tools, the promise of Intellisense, colorization of keywords, operators, constants, and more, the Oslo architects recognize the ability to enhance our development experience in a language-agnostic way, raising the level of abstraction because, as they say, the way to solve any technical problem is to approach it at one higher level of indirection.  Unfortunately, this makes Oslo so generalized and abstract that it’s difficult to grasp and therefore to appreciate its immensity.  Once you can take a step back and see how it fits in holistically, you’ll see that it has the potential to dramatically transform the landscape of software development.

Currently, it’s a lot of work to implement all the language services in Visual Studio to give them as rich an experience as we’ve come to expect with C#, VB.NET, and others.  This is a serious impediment to doing this kind of work, so solving the problem at the level of Oslo drastically lowers the barrier to entry for implementing tool-supported languages.  The Oslo bits I’ve seen and played with are very early in the lifecycle for this massive scope of technology, but the more I think about its potential, the more impressed I am with the fundamental concept.  As Chris Anderson explained in his PDC session on MGrammar, MGrammar was an implementation detail, but sometime around June 2007, that feature team realized just how much customers wanted direct access to it and decided to release MGrammar to the world.

Modeling & The Repository

That’s all well and good for DSLs and language enthusiasts/geeks, but primarily perhaps, Oslo is about the creation, exploration, relation, and execution of models in an interoperable way.  In other words, all of the models that are currently used to describe a software system, or an entire IT environment, are either not encoded formally enough to verify or execute, or they’re encoded or stored in proprietary ways that don’t allow interoperability with other models.  A diagram in Visio or PowerPoint documenting network topology, for example, knows nothing about the component architecture or deployment model of the software systems installed and running on that network.

When people usually talk about models, they imagine high-level architecture documents, overviews used to visually summarize work that is much more granular in nature.  These models aren’t detailed, and they normally aren’t kept up to date and in sync with the current design as changes are made.  But modeling in Oslo is not an attempt to make these visual models contain all of the necessary detail, or to develop software with visual tools exclusively.  Oslo simply provides the tools, both graphical and textual, to define and relate many models.  It will be up to the development community to decide how all these tools are ultimately used, which parts of our systems will be specified in a mix of general purpose, domain specific, and visual languages.  Ultimately, Oslo will provide the material and glue to fill the gaps between the high and low level specifications, and unite them into a common, connected, and much more useful set of data.

To grasp what Oslo modeling is really all about requires that we expand our definition of "model", to see the models expressed in our configuration and XAML files, in our applications’ database schemas, in our entity classes, and so on.  As software grows in complexity and becomes more composable, we can use various languages to model its behavior, store that in the repository for runtime execution, inspection, or reuse by other systems.

This funny and clever Oslo video (reminiscent of The Hitchhiker’s Guide to the Galaxy) explains modeling in the broader sense alluded to here.

If we had some universal container for the storage of all different kinds of models, and a standardized way of relating entities across models, we’d be able to do things like impact analysis, where we could see the effect on software systems if someone were to alter the network it was running on; or powerful data mining on the IT execution environment of a business.

Many different tools, with different audiences, will be able to connect into this repository to manipulate aspects of the models that they understand and have access to.  This is just the tip of the iceberg.  We already model so much of what we do in the IT and software worlds, and as we begin adopting business process middleware and orchestration software like BizTalk, there’s a huge amount of value in those models converging and connecting.  That’s where the Oslo Repository comes in.

Oslo provides interoperability among models in the same way that SOA provides interoperability among services.  Not unlike the interoperability we have now among many different languages all sharing the same CLR specification.

Bridging data models across repositories or in shared repository is a major step forward.  With Windows Azure and Microsoft’s commitment to their online services platform (and considering the momentum of the SaaS movement with Amazon, Google, and others), shared storage and data sets are the future.  (Check out SQL Data Services if you haven’t already, and watch for some exciting announcements coming later this year!)

The Dichotomy of Data vs. Metadata

Jeff Pinkston from the Oslo team aptly reflects the attitude of the group when he scoffs at the categorical difference between data and metadata.  In terms of storing and querying it, serializing and communicating it, and everything else that matters in enterprise software, data is data and there’s no reason not to treat it the same when it comes to architecting a system.  We have our primary models and our secondary models, our shared models and our protected models, but they’re still just models that shape our software’s behavior, and they share all of the same characteristics when it comes to manipulation and access.  It’s their ultimate effect that differs.

It’s worth noting, I think, the line that’s been drawn between code and data in some programming languages and not in others (C# vs. LISP).  A division has been made for the sake of security rather than necessity.  Machine instruction codes are represented in the same sort of binary data and realized in the same digital circuitry as traditional user data.  It’s tempting to keep things locked down and divided, but as languages evolve to become more late bound and dynamic (and as the tools evolve to make this feasible), there will be more need for the manipulation of expression trees and ASTs.  I strongly suspect the lines will blur until they disappear.

Schema and Object Instance Languages

In order to define models, we need a tool.  In Oslo, this is a textual language called MShema and an editor called Intellipad.  I personally think it’s odd to talk people’s ears off about "model, model, model", and then to use the synonym "schema" to name the language, but all of these names could change before they’re shipped for all we know.

This is a simple example of an MSchema document:

module MyModel
{
    type Person
    {
        LastName : Text;
        FirstName : Text;
    }

    People : Person*;
}

By running this through the "M Compiler", a SQL script is generated that will create the appropriate database objects.  Intellipad is able to verify the correctness of your schema, and what’s really nice is that you don’t even have to specify data types when you start sketching out your model.  Defaults are assumed, and you can get more specific as your model evolves.

MGraph is a language for defining instances of objects, constrained by an MSchema and similar in format.  So MSchema is to MGraph what XSD is to XML.

In this article, Lars Corneliussen explains Microsoft’s vision to make MGraph as common as XML is today.  Take a look at his article to see a side-by-side comparison of the same object represented as XML (POX), JSON, and MGraph, and decide for yourself which you like best (or see below).

MSchema and MGraph are easier and more efficient to read and write than XML.  Their message format resembles typical structured programming languages, and developers are already familiar with these formats.  XML is a fine format for a tool; it’s human readable but not human-friendly.  A C-style language, on the other hand, is much more human-friendly than all of the angle brackets and the redundancy (and verbosity) of tag text.  That narrows down our choice to JSON and MGraph.

In JSON, the property/field/attribute names are delimited by quotation marks, suggesting that the whole structure is a dumb property bag.

{
    "LastName" : "Vanderboom",
    "FirstName" : "Dan"
}

MGraph has a very similar syntax, but its attribute property names are recognized and validated by the parser generated from MSchema, so the quotation marks are unnecessary.  It ends up looking more natural, and a little more concise.

{
    LastName : "Vanderboom",
    FirstName : "Dan"
}

Because MGraph is just a message format, and Microsoft’s service offerings already support multiple message formats (SOAP/POX/JSON/etc.), it wouldn’t disrupt any of their architecture to add an MGraph adapter, and I’ll be shocked if I don’t hear about one in their next release.

Meta-Languages and MGrammar

In the same way that Oslo includes a meta-model because it allows us to define models, it also includes a meta-language because it allows us to define languages (as YACC and ANTLR have done).  However, just as Pinkston doesn’t think data and metadata should be treated different, it makes sense to think of a language that defines languages as just another language.  There is something Zen about that, where the tools somehow seem to bend back upon themselves like one of Escher‘s drawings.

Here is an example language defined by MGrammar in a great article on MSDN called MGrammar in a Nutshell:

module SongSample
{
    language Song
    {
        // Notes
        token Rest = "-";
        token Note = "A".."G";
        token Sharp = "#";
        token Flat = "b";
        token RestOrNote = Rest | Note (Sharp | Flat)?;

        syntax Bar = RestOrNote RestOrNote RestOrNote RestOrNote;
        syntax List(element)
          = e:element => [e]
          | es:List(element) e:element => [valuesof(es), e];

        // One or more bars (recursive technique)
        syntax Bars = bs:List(Bar) => Bars[valuesof(bs)];
        syntax ASong = Music bs:Bars => Song[Bars[valuesof(bs)]];
        syntax Songs = ss:List(ASong) => Songs[valuesof(ss)];

        // Main rule
        syntax Main = Album ss:Songs => Album[ss];

        // Keywords
        syntax Music = "Music";
        syntax Album = "Album";

        // Ignore whitespace
        syntax LF = "\u000A";
        syntax CR = "\u000D";
        syntax Space = "\u0020";

        interleave Whitespace = LF | CR | Space;
    }
}

This is a pretty straight forward way to define a language and generate a parser.  Aside from the obvious keywords to define syntax rules and token patterns (with an alternative and more readable format for regular expressions), the => projection operator allows you to shape the MGraph output according to your needs.

I created two simple languages with MGrammar on the plane trip back to Milwaukee from the PDC in November.  The majority of my time was spent fussing with the editor, Intellipad, and for the last half hour I found it very easy to create a language on the fly, extending and changing it through experimentation quickly and easily.  Projections, which are functional expressions in MGrammar used to shape MGraph output, are the most challenging part.  There are a number of techniques that shape the output graph, so it will be good to see how this is approached in future reference examples.

Surreptitiously announced just before I wrote this, Mike Weinhardt at Microsoft indicated that a gallery of example grammars for MGrammar is being put together, to point to the sample grammars for various languages in addition to grammars that the community develops, and it should be available by the end of this month.  These examples demonstrating how to define languages and write sensible projections, coming from the developers who are putting MGrammar together, will be an invaluable tool for teaching you how to use common patterns (just as 101 LINQ Samples did for LINQ).

As Doug Purdy explained on .NET Rocks: "People who are building a domain specific language, and they don’t want to understand how to build a parser, or they’re not language designers.  Actually, they are language designers.  They design a language, but they actually don’t do the whole thing.  They don’t build a parser.  What they do, they just leverage the XML parser.  And what we’re trying to do is provide a toolset for folks where they don’t have to resort to XML in order to do DSLs."

From the same episode, Don Box said of the DSL session at PDC: "I’ve never seen a session with more geek porn in it."

Don: "It’s like crack for developers.  It’s kind of addictive; it takes over your life."

Doug: "If you want the power of Anders in your hand…"

The Tool Chain

Now that we have a better sense of what’s included in Oslo in terms of languages, editors, and the shared repository, we can look at the relationship among the other pieces, which are manifested in the CTP as a set of command-line tools.  In the future, these will integrate into an IDE, most likely Visual Studio.  (I’d expect Intellipad and Quadrant to merge with Visual Studio, but there’s no guaranty this will happen.)

When you create your model with MSchema, you’ll use m to validate that model and generate a SQL script to create a SQL Server 2008 database schema (yes, it only works right now with SQL Server 2008).  You’ll also use the m command to validate your object graph (written in MGraph) against your schema, and translate that into a set of SQL commands to perform inserts and updates against tables.

With enough models, there’ll be huge value in adding yours to the repository.  If you don’t mind writing MGraph or you generate it automatically with something like an MGraphSerializer class in your code, this may be all you need.

If, on the other hand, you decide you could really benefit by defining your own textual language to use instead of MGraph, you can use MGrammar to define a new language.  This language gets compiled by the mg compiler to create your parser, and the mgx command translates code in your new language into an MGraph, which can then be pulled into your database using m.

This diagram depicts the process:

Other than these command-line tools, Quadrant is the highly extensible visual tool for exploring models graphically, and Intellipad is a different face on the same shell for defining DSLs with MGrammar and writing DSL code, as well as writing and verifying MSchema and MGraph code.

We should see fairly soon the convergence of these three languages (MGraph, MSchema, and MGrammar) into a single M language.  This makes sense, since what you want to project in your DSL should be something within your model, verified by your schema.  This may ultimately make these projections much easier to write.

We’ll also see this tool chain absorbed into multiple development environments, eventually with rich binding across multiple representations of our model, although this will take longer in Visual Studio.

Languages and Nested Languages

I looked at some MService examples, and I can understand Damon’s concern that although it’s nice to have "operation" as a keyword in a service-oriented language, with more keywords giving you the ability to specify aspects of each endpoint and the communications patterns required, that enclosing the business logic within that service language is probably not a good idea.  I took this from Dennis van der Stelt’s blog:

service Service
{
  operation PhotoUpload(stream : Stream) : Text
  {
    .PostUriTemplate = "upload";

    index : Text = invoke DateTime.Now.Ticks.ToString();
    filename : Text = "d:\\demo\\photo\\" + index + ".jpg";
    invoke MService.ServiceHelper.StoreInFile(stream, filename);

    return index;
  }
}

Why not?  You’re defining a general purpose language within the curley braces, one capable of defining variables, assigning values, referencing .NET objects, and calling methods.  But why do you want to learn a new language to write services when the language you’re using right now is already supremely capable of that?  Don’t you already know a good syntax for invoking methods (other than "invoke %mehthod%")?  If instead you simply referenced an assembly, type, and method from an MService script, you could externally turn any .NET method with serializable parameters and return value into a service operation by feeding it this kind of file, without having to recompile, and without having to reinvent the wheel.

The possible exception would be if MGrammar adds the ability (as discussed by speakers at the PDC) of supporting multiple layers of enclosing languages within other languages.  In other words, you could use MService to define operations and their attributes using its own syntax, and within the curly braces that follow, use the C# or VB.NET parsers to process the logic with the comprehension of a separate language.  There are some neat possibilities here, but I expect the development community to be conservative and hesitent about mixing layers of semantics, as there is an awful lot of room for confusion and complexity.  It may be better to leave different language blocks in separate files or containers, and to allow them to reference each other as .NET assemblies and XML files reference each other today.

However, I wouldn’t get too hung up on the early versions of these new languages, or any one language specifically.  The useful, sensible ones that take real developer needs into account and provide the most value will be adopted, and many more will quickly fall into disuse.  But the overall pattern will be for the emergence of an amazing amount of leverage in terms of improving human comprehension and taking advantage of our ability to manipulate structured, symbolic object graphs to build and verify software systems.

Resources

After a few months of research and many hours of writing, I don’t feel like I’ve even scratched the surface.  But instead of giving you an absolutely comprehensive picture, I’m going to stop here and continue in future articles.  In the meantime, check out the following resources.

For an overview of the development paradigm, look for information on language-oriented programming, including an article I wrote that alludes to how "we will have to raise the level of abstraction to a point that may be hard for us to imagine with our existing tools and languages" due to the "precipitious growth of software complexity".  The "community of abstractions" is the model in Oslo-speak.

For Microsoft specific content: there were some great sessions at the PDC (watch the recorded videos).  It was covered (with much confusion) on the .NET Rocks! podcast (here and here) as well as on Software Engineering Radio; and there are lots of bloggers talking about their initial experiences with it, such as Shawn Wildermuth, Lars Corneliussen, and of course Chris Sells and Jeff Pinkston.  The most clear and coherent explanation I’ve heard was from an interview with Ron Jacobs and David Chappell (Ron gave the keynote at MSDN Dev Con, hosted the ARCast podcast for years).  MSDN has at least 29 videos on the Oslo Developer Center, where there’s a good amount of information. including a FAQ.  There’s also the online guide for MGrammar, MGrammar in a Nutshell, and the Oslo team blog.

If you’re interested in creating DSLs, make sure to keep a look out for details about the upcoming DSL Developers Conference, which is tentatively planned for April 16-17, immediately following the Lang.NET conference (on general purpose languages) on April 14-16.  I’m hoping to be at both this year.  And in case you haven’t heard, Microsoft is planning another PDC Conference for 2009, the first time ever these conferences have run for two consecutive years!  There will no doubt be much more Oslo news and conference material to cover it at the PDC in November.

Pluralsight, an instructor-led training company, now teaches a two-day "Oslo" Fundamentals course (and Don Box’s blog is hosted there).

The best way to learn about Oslo, however, is to dive in and use it.  That’s what I’m doing with my newest system, which needs to be modeled from scratch.  So if you haven’t done so already, download the Oslo SDK (link updated to January 2009 SDK) and introduce yourself to the future of modeling and development!

[Click here for the next article in this Oslo series, on common misconceptions and fallacies about Oslo.]

The Future of Programming Languages

Two of the best sessions at the PDC this year were Anders Hejlsberg’s The Future of C# and a panel on The Future of Programming.

A lot has been said and written about dynamic programming, metaprogramming, and language syntax extensions–not just academically over the past few decades, but also as a recently growing buzz among the designers and users of mainstream object-oriented languages.

Dynamic Programming

After a scene-setting tour through the history and evolution of C#, Anders addressed how C# 4.0 would allow much simpler interoperation between C# and dynamic languages.  I’ve been following Charlie Calvert’s Language Futures website, where they’ve been discussing these features early on with the development community.  It’s nice to see how seriously they take the feedback they’re getting, and I really think it’s going to have a positive impact on the language as a whole.  Initial thoughts revolved around creating a new block of code with code like dynamic { DynamicStuff.SomeUndefinedProperty = “whatever”; }.

But at the PDC we saw that instead dynamic will be a type for our dynamic objects, and so dynamic lookup of members will only be allowed for those variables.  Anders’ demo showed off interactions with JavaScript and Python, as well as Office via COM, all without the ugly Type.Missing parameters (optional parameter support also played a part in that).  Other ideas revolved around easing Reflection access, and XML document access for Xml nodes dynamically.

Meta-Programming

At the end of his talk, Anders showed a stunning demo of metaprogramming working within C#.  It was an early prototype, so all language features were not supported, but it worked similar to Eval where the code was constructed inside a string and then compiled at runtime.  But it was flexible and powerful enough that he could create delegates to functions that he Eval’ed up into existence.  Someone in the audience asked how this was different from Lisp macros, to which Anders replied: “This is basically Lisp macros.”

Before you get too excited (or worried) about this significant bit of news, Anders made no promises about when metaprogramming would be available, and he subtly suggested that it may very well be a post-4.0 feature.  As he said in the Future of Programming Panel, however: “We’re rewriting the compiler in managed code, and I’d say one of the big motivators there is to make it a better metaprogramming system, sort of open up the black box and allow people to actually use the compiler as a service…”

Regardless of when it arrives, I hope they will give serious consideration to providing syntax checking of this macro or meta code, instead of treating it blindly at compile-time as a “magic string”, as has so long plagued the realm of data access.  After all, one of the primary advantages of Linq is to enable compile-time checking of queries, to enforce not only strict type checking, but to also more fundamentally ensure that data sources and their members are valid.  The irregularity of C#’s syntax, as opposed to Lisp, will make that more difficult (thanks to Paul for pointing this out), but I think most developers will eventually agree it’s a worthwhile cause.  Perhaps support for nested grammars in the generic sense will set the stage for enabling this feature.

Language Syntax Extensions

If metaprogramming is about making the compiler available as a service, language extensions are about making the compiler service transparent and extensible.

The majority (but not all) of the language design panel stressed caution in evolving and customizing language syntax and discussed the importance of syntax at length, but they’ve been considering the demands of the development community seriously.  At times Anders vacillated between trying to offer alternatives and admitting that, in the end, customization of language syntax by developers would prevail; and that what’s important is how we go about enabling those scenarios without destroying our ability to evolve languages usefully, avoiding their collapse from an excess of ambiguity and inconsistency in the grammar.

“Another interesting pattern that I’m very fond of right now in terms of language evolution is this notion that our static languages, and our programming languages in general, are getting to be powerful enough, that with all of these things we’re picking up from functional programming languages and metaprogramming, that you can–in the language itself–build these little internal DSLs, where you use fluent interface style, and you dot together operators, and you have deferred execution… where you can, in a sense, create little mini languages, except for the syntax.

If you look at parallel extensions for .NET, they have a Parallel.For, where you give the start and how many times you want to go around, and a lambda which is the body you want to execute.  And boy, if you squint, that looks like a Parallel For statement.

But it allows API designers to experiment with different styles of programming.  And then, as they become popular, we can pick them up and put syntactic veneers on top of them, or we can work to make languages maybe even richer and have extensible syntax like we talked about, but I’m encouraged by the fact that our languages have gotten rich enough that you do a lot of these things without even having to have syntax.” – Anders Hejlsberg

On one hand, I agree with him: the introduction of lambda expressions and extension methods can create some startling new syntax-like patterns of coding that simply weren’t feasible before.  I’ve written articles demonstrating some of this, such as New Spin on Spawning Threads and especially The Visitor Design Pattern in C# 3.0.  And he’s right: if you squint, it almost looks like new syntax.  The problem is that programmers don’t want to squint at their code.  As Chris Anderson has noted at the PDC and elsewhere, developers are very particular about how they want their code to look.  This is one of the big reasons behind Oslo’s support for authoring textual DSLs with the new MGrammar language.

One idea that came up several times (and which I alluded to above) is the idea of allowing nested languages, in a similar way that Linq comprehensions live inside an isolated syntactic context.  C++ developers can redefine many operators in flexible ways, and this can lead to code that’s very difficult to read.  This can perhaps be blamed on the inability of the C++ language to provide alternative and more comprehensive syntactic extensibility points.  Operators are what they have to work with, so operators are what get used for all kinds of things, which change per type.  But their meaning gets so overloaded, literally, that they lose any obvious (context-free) meaning.

But operators don’t have to be non-alphabetic tokens, and the addition of new keywords or symbols could be introduced in limited contexts, such as a modifier for a member definition in a type (to appear alongside visibility, overload, override, and shadowing keywords), or within a delimited block of code such as an r-value, or a curly-brace block for new flow control constructs (one of my favorite ideas and an area most in need of extensions).  Language extensions might also be limited in scope to specific assemblies, only importing extensions explicitly, giving library authors the ability to customize their own syntax without imposing a mess on consumers of the library.

Another idea would be to allow the final Action delegate parameter of a function to be expressed as a curly-brace-delimited code block following the function call, in lieu of specifying the parameter within parentheses, and removing the need for a semicolon.  For example, with a method defined like this:

public static class Parallel
{
    // Action delegate defined last, to take advantage of C# syntactic sugar
    public static void For(long Start, long Count, Action Action)
    {
        // TODO: implement
    }
}

…a future C# compiler might allow you to write code like this:

Parallel.For(0, 10)
{
    // add code here for the Action delegate parameter
}

As Dr. T points out to me, however, the tricky part will consist of supporting local returns: in other words, when you call return inside that delegate’s code block, you really expect it to return from the enclosing method, not the one defined by the delegate parameter.  Support for continue or break would also make for a more intuitive fit.  If there’s one thing Microsoft does right, it’s language design, and I have a lot of confidence that issues like this will continue to be recognized and ultimately implemented correctly.  In reading their blogs and occasionally sharing ideas with them, it’s obvious they’re as passionate about the language and syntax as I am.

The key for language extensions, I believe, will be to provide more structured extensibility points for syntax (such as control flow blocks), instead of opening up the entire language for arbitrary modification.  As each language opens up some new aspect of its syntax for extension, a number of challenges will surface that will need to be dealt with, and it will be critical to solve these problems before continuing on with further evolution of the language.  Think of all we’ve gained from generics, and the challenges of dealing with a more complex type system we’ve incurred as a result.  We’re still getting updates in C# 4.0 to address shortcomings of generics, such as issues regarding covariance and contravariance.  Ultimately, though, generics were well worth it, and I believe the same will be said of metaprogramming and language extensions.

Looking Forward

I’ll have much more to say on this topic when I talk about Oslo and MGrammar.  The important points to take away from this are that mainstream language designers are taking these ideas to heart now, and there are so many ideas and options out there that we can and will experiment to find the right combination (or combinations) of both techniques and limitations to make metaprogramming and language syntax extensions useful, viable, and sustainable.

Windows Azure: Internet Operating System

As I write this during the keynote of PDC 2008, Ray Ozzie is announcing (with a great deal of excitement) a new Windows-branded Internet operating system called Windows Azure.  Tipping their hats to work done by Amazon.com with their EC2 (Elastic Cloud Computing) service, Microsoft is finally revealing their vision for a world-wide, scalable, cloud services platform that looks forward toward the next fifty years.

Windows Azure, presumably named for the color of the sky, is being released today as a CTP in the United States, and will be rolled out soon across the rest of the planet.  This new Internet platform came together over the past few years after a common set of scalable technologies emerged behind the scenes across MSN, Live, MSDN, XBox Live, Zune, and other global online services.  Azure is the vehicle for delivering that expertise as a platform to Microsoft developers around the world.

SQL Server Data Services, now called SQL Services (storage, data mining, reporting), .NET Services (federated identity, workflows, etc.), SharePoint Services, Live Services (synchronization), and Dynamics/CRM Services are provided as examples of services moving immediately to Windows Azure, with many more on their way, both from Microsoft and from third parties such as Bluehoo.com, a social networking technology using Bluetooth on Windows Mobile cell phones to discover and interact with others nearby.

The promise of a globally distributed collection of data centers offering unlimited scalability, automatic load-balancing, intelligent data synchronization, and model-driven service lifecycle management is very exciting.  Hints of this new offering could be seen from podcasts, articles, and TechEd sessions earlier this year on cloud services in general, talk of Software-As-Service (SAS), Software+Services, the Internet Service Bus, and the ill-named BizTalk Services.

The development experience looks pretty solid, even this early in the game.  Testing can be done on your local development machine, without having to upload your service to the cloud, and all of the cloud services APIs will be available, all while leveraging existing skills with ASP.NET, web services, WCF and WF, and more.  Publishing to the cloud involves the same Publish context menu item you’re used to in deploying your web applications today, except that you’ll end up at a web page where you’ll upload two files.  Actual service start time may take a few minutes.

Overall, this is a very exciting offering.  Because of Microsoft’s commitment to building and supporting open platforms, it’s hard to imagine what kinds of products and services this will lead to, how this will catalyze the revolution to cloud services that has already begun.

Functional Programming as Intensity of Expression

On my long drive home last night, I was thinking about the .NET Rocks episode with Ted Neward and Amanda Laucher on F# and functional programming.  Though they’re writing a book on F# together, it seems even they have a hard time clearly articulating what functional programming is all about, and where it’s all headed in terms of mainstream commercial use… aside from scientific and data transformation algorithms, that is (as with the canonical logging example when people explain AOP).

I think the basic error is in thinking that Functional is a Style of programming.  Yet, to say that so-called Imperative-based languages are non-functional is ridiculous.  Not in the sense that they “don’t work”, but that they’re based on Objects “instead of” Functions.

This isn’t much different from the chicken-and-egg problem.  Though the chicken-and-egg conundrum has a simple (but unobvious) answer, it doesn’t really matter whether the root of program logic is a type or a function.  If I write a C# program with a Program class, the Main static function gets called.  Some action is the beginning of a program, so one might argue that functions should be the root-most logical construct.  However, you’d then have to deal with functions containing types as well as types containing functions, and as types can get very large (especially with deep inheritance relationships), you’d have to account for functions being huge, spanning multiple code files, and so on.  There’s also the issue of types being organizational containers for functions (and other members).  Just as we use namespaces to organize our types, so we use types to organize functions.  This doesn’t prevent us from starting execution with a function or thinking of the program’s purpose functionally; it just means that we organize it inside a logical container that we think of as a “thing”.

Does this limit us from thinking of business processes as functional units?  Ted Neward suggests that we’ve been trained to look for the objects in a system, and base our whole design process on that. But this isn’t our only option for how to think about design, even in our so-called imperative languages.  If we’re thinking about it wrong, we can and should change the process; we don’t need to blame our design deficiencies on the trivial fact of which programming construct is the root one.  In fact, there’s no reason we should use any one design principle to the exclusion of others.  Looking for the things in the system is and will remain a valuable approach for discovering and defining database schemas and object models.  The very fact that “functional languages” aren’t perceived as especially useful for stateful components isn’t a fault of a style of programming, but is rather a natural consequence of functions being an incomplete aspect of a general purpose programming language.  Functional is a subset of expressive capability.

Where “functional languages” have demonstrated real value is not in considering functions as root-level constructs (this may ultimately be a mistake), but rather in increasing the flexibility of a language to be much more expressive when defining functions.  Making functions first-class citizens that can be passed as parameters, returned as function values, and stitched together with metaprogramming techniques, is a huge step in the right direction.  The use of simple constructs such as operators to match patterns, reverse the evaluation of functions and the flow of values with piping, and perform complex set- and list-based operations, all increase the expressive intensity and density of the functions in a language.  This can only add to the richness of our existing object models.

Sticking objects together in extensible and arbitrarily complex structures is routine for us, but now we’re seeing a trend toward the same kind of composability in functions.  Of course, even this isn’t new, per se; the environmental forces that demand this power just haven’t become significant enough to require that level of power in mainstream languages, because technology evolution (like evolution in general) tends to work by adapting solutions that are “good enough”.

It’s common to hear how F# is successfully incorporating “both functional and imperative” styles into one language, and this is important because what we need is not so much the transition to a functional style, as I’ve mentioned already, but a growth of greater functional expressiveness and power in existing, successful, object-oriented languages.

So let our best and favorite languages grow, and add greater expressive powers to them, not only for defining functions, but also in declaring data structures, compile-time constraints and guarantees, and anything else that will help to raise the level of abstraction and therefore the productivity with which we can naturally express and fulfill our business needs.

Ultimately, “functional programming” is not a revolutionary idea, but rather an evolutionary step forward.  Even though it’s impact is great, there’s no need to start from scratch, to throw out our old models.  Incompatibility between functional and imperative is an illusion perpetuated by an unclear understanding of their relationship and each aspect’s purpose.

Observations on the Evolution of Software Development

Neoteny in the Growth of Software Flexibility and Power

Neoteny is a biological phenomenon of an organism’s development observed across multiple generations of a species.  According to Wikipedia, neoteny is “the retention, by adults in a species, of traits previously seen only in juveniles”, and accounts for many evolutionary shifts, including the human brain’s ability to remain elastic and malleable later in life than those of our distant ancestors.

So how does this relate to software?  Software is a great deal like an organic species.  The species emerged (not long ago), incubated in a more or less fragile state for a number of decades, and continues to evolve today.  Each software application or system built is a new member of the species, and over the generations they have become more robust, intelligent, and useful.  We’ve even formed a symbiotic relationship with software.

Consider the fact that software running on computers was at one time compiled to machine language code for a specific processor.  With the invention of platform-independent instruction sets and their associated runtimes performing just-in-time compilation (Java’s JVM and .NET Framework’s CLR), we’ve delayed the actual production of machine language code until it’s actually needed on the target machine.  The compiler produces a slightly more abstract representation of the program logic, and an extra translation step at installation or runtime is needed to complete the process to make the software usable.

With the growing popularity of dynamic languages such as Lisp, Python, and the .NET Framework’s upcoming release of its Dynamic Language Runtime (DLR), we’re taking another step of neoteny.  Instead of a compiler generating instruction byte codes, a “compiler for any dynamic language implemented on top of the DLR has to generate DLR abstract trees, and hand it over to the DLR libraries” (per Wikipedia).  These abstract syntax trees (AST), normally an intermediate artifact created deep within the bowels of a traditional compiler (and eventually discarded), are now persisted as compiler output.

Traits previously seen only in juveniles… now retained by adults.  Not too much of a metaphorical stretch!  The question is: how far can we go?  And I think the answer depends on the ability of hardware to support the additional “just in time” processing that needs to occur, executing more of the compiler’s tail-end tasks within the execution runtime itself, providing programming languages with greater flexibility and power until the compilation stages we currently execute at design-time almost entirely disappear (to be replaced, perhaps, by new pre-processing tasks.)

I remember my Turbo Pascal compiler running on a 33 MHz processor with 1 MB of RAM, and now my cell phone runs at 620 MHz (with a graphics accelerator) and has gigabytes of memory and storage.  And yet with the state of things today, the inclusion of language-specific compilers within the runtime is still quite infeasible.  In the .NET Framework, there are too many potential languages that people might attempt to include in such a runtime: C#, F#, VB, Boo, IronPython, etc.  Trying to cram all of those compilers into a universal runtime that would fit (and perform well) on a cell phone or other mobile device isn’t yet feasible, which is why we have technologies with approaches like System.Reflection.Emit (on the full .NET Framework), and Mono.Cecil (which works on Compact Framework as well).  These work at the platform-independent CIL level, and so can interpret and generate programs generically, interact with each others’ components, and so on.  One metaprogramming mechanism can therefore be reused across all .NET languages, and this metalinguistic programming trend is being discussed on the C# and other language design teams.

I’ve just started using Mono.Cecil, chosen because it is cross-platform friendly (and open source).  The API isn’t very intuitive, but because the source is available, and because extension methods can go a long way to making it more accessible, it’s a great option.  The documentation is sparse, and assembly generation has some performance issues, but it’s a work-in-progress with tremendous potential.  If you’re doing any kind of static analysis or have any need to dynamically generate and consume types and assemblies (to get around language limitations, for example), I’d encourage you to check it out.  A comparison of Mono.Cecil to System.Reflection can be found here.  Another library called LinFu, which performs lots of mind-bending magic and actually uses Mono.Cecil, is also worth exploring.

VB10 will supposedly be moving to the DLR to become a truly dynamic language, which considering their history of support for late binding, makes a lot of sense.  With a dynamic language person on the C# 4.0 team (Jim Hugunin from IronPython), one wonders if C# won’t eventually go the same route, while keeping its strongly-typed feel and IDE feedback mechanisms.  You might laugh at the idea of C# supporting late binding (dynamic lookup), but this is being planned regardless of the language being static or dynamic.

As the DLR evolves, performance optimizations are being discovered and implemented that may close the gap between pre-compiled and dynamically interpreted languages.  Combine this with manageable concurrent execution, and the advantages we normally attribute to static languages may soon disappear altogether.

The Precipitous Growth of Software System Complexity

We’re truly on the cusp of a precipitous period of growth for software complexity, as an exploding array of devices and diverse platforms around the world connect in an ever-more immersive Internet.  Taking full advantage of parallel and distributed computing environments by solving the challenges of concurrency and coordination, as well as following the trend toward increased integration among software components, is pushing software complexity into new orders of magnitude.  The strategies we come up with for organizing these systems will have to take several key factors into consideration, and we will have to raise the level of abstraction to a point that may be hard for us to imagine with our existing tools and languages.

One aspect that’s clear is the rise of declarative or intention-based syntax, whether represented as XML, Domain Specific Langauges (DSL), attribute decoration, or a suite of new visual modeling editors.  This is in part a consequence of raising the abstraction level, as lower-level libraries are entrusted to solve common problems and take advantage of common opportunities.

Another is the use of Inversion of Control (IoC) containers and dependency injection in component based architectures, thereby standardizing the lifecycle of the application and its components, and providing a common environment or ecosystem for all of its components, as well as introducing a common protocol for component location, creation, access, and disposal.  This level of consistency is valuable for sharing a common understanding of how to troubleshoot software components.  The more predictable a component’s interaction with the rest of the system, the easier it is to debug and modify; conversely, the more unique it and its communication system is, the more disparity there is among components, and the more difficult to understand and modify without introducing errors.  If software is a species and applications are individuals, then components are the cells of a system.

Even the introduction of functional programming languages into the mainstream over the past couple years is due, in part, to the ability of those languages to provide more declarative support, more syntactic flexibility, and new ways of dealing with concurrency and coordination issues (such as immutable values) and light-weight, ad hoc data structures (tuples).

Balancing the Forces of Coupling, Cohesion, and Modularity

On a fundamental level, the more that components are independent, the less coupled and the more modular and flexible they are.  But the more they can communicate with and are allowed to benefit from each other, the more interdependent they become.  This adds to cohesiveness and synergy, but also stronger coupling to a community of abstractions.

A composition of services has layers and segments of interdependence, and while there are dependencies, these should be dependencies on abstractions (interfaces and not implementations).  Since there will be at least one implementation of each service, and the extensibility exists to build others as needed, dependency is only a liability when the means for fulfilling it are not extensible.  Both sides of a contract need to be fulfilled regardless; service-oriented or component-based designs merely provide a mechanism for each side to implement and fulfill its part of the contract, and ideally the system also provides a discovery mechanism for the service provider to publish its availability for other components to discover and consume it.

If you think about software components as a hierarchy or tree of services, with services of one layer depending on more root services, it’s easy to see how this simplifies the perpetual task of adding new and revising existing functionality.  You’re essentially editing an outline, and you have opportunities to move services around, reorganize dependencies easily, and have many of the details of the software’s complexity absorbed into this easy-to-use outline structure (and its supporting infrastructure).  Systems of arbitrary complexity become feasible, and then relatively routine.  There’s a somewhat steep learning curve to get to this point, but once you’ve crossed it, your opportunities extend endlessly for no additional mental cost.  At least not in terms of how to compose your system out of individual parts.

Absorbing Complexity into Frameworks

The final thing I want to mention is that a rise in overall complexity doesn’t mean that the job of software developers necessarily has to become more difficult than it is currently.  With the proper design of components that abstract away the complexity into reusable frameworks with intuitive interfaces, developers at the business logic level don’t need to be aware of the inner complexity, in the same way that software developers are largely absolved of the responsibility of thinking about the processor’s inner workings.  As we build our technology stack higher and higher, like the famed Tower of Babel, we must make sure that it’s organized and structured in a way to support that upward growth and the load imposed upon it… so it doesn’t come crashing down.

The requirements for building components tomorrow will not be the same as they were yesterday.  As illustrated in this account of the effort involved in a feature change at Microsoft, in the future, we will also want to consider issues such as tool-assisted refactorability (and patterns that frustrate this, such as “magic strings”), and due to an explosion of component libraries, discoverability of types, members, and their use.

A processor can handle any complexity of instruction and data flow.  The trick is in organizing all of this in a way that other developers can understand and work with.

Introduction to Port-Based Asynchronous Messaging

…using terminology and examples from the Concurrency & Coordination Runtime in C#

 

Port-Based Messaging

Port-based messaging has been in practice for a long time.  In the realm of low-level electronics, components have always been operating in parallel, hardware interface ports are designed around standards, and messages are posted to those ports, queuing up somewhere until the receiver is ready to process them.  This pattern has worked extremely well in multiple domains for a long time, and its key characteristic is the decoupling of data flow from control flow.

In sequential programs, one statement runs after another each time it’s run, and the behavior is predictable and repeatable. Concurrency is difficult because you have to consider all possible interleavings of multiple simultaneous tasks, so the overall behavior is nondeterministic. Depending on the relative timings of concurrent tasks, you could get different results each time if you’re not careful to set the appropriate locks on shared resources.  Port-based messaging architectures isolate islands of state across different execution contexts, and connect them with safe, atomic messages delivered through pre-defined ports.

The posting of a message to a port, as shown in Figure 1, is followed by some handler method that is receiving and processing messages.  What’s not evident in the diagram, however, is that while data flows into the port, that posting is a non-blocking call.  The sender continues on doing other work, taking the time only to queue up a message somewhere.

Queuing is important because, even with large thread pools, we can’t guaranty that a receiver will be listening at the very moment a message arrives.  Letting them queue up means the receiver doesn’t have to block on a thread to wait.  Instead, the data waits and the thread checks messages on the port when it can.

What is a Port?

So what exactly is a port?  A port is a communication end point, but not in the sense of “a web service on a physical server”.  Think much more fine grained than that, even more fine-grained than methods.  With sequential programming, we commonly use try-catch blocks and handle both the exceptional and non-exceptional results of operations within a single method.  In port-based programming, you post a message to a port, which results in some handler method running on the receiving end, and different results can be sent back on different callback ports depending on the type of message.  Instead of calling a method that returns back to you when it ends, port-based programming is about always moving forward, and unwinding a call stack has very little meaning here.

We can see in the sequence diagram above (Figure 3) a collection of services that communicate with and depend on each other.  Starting from the top, the left-most service posts a message to port 1, and then goes on to do other work or surrenders its thread back to the dispatcher for other tasks that are waiting to run.  A registered method on port 1 runs, and the logic there needs another service to complete it’s task, so it posts a message on port 2, and also continues processing without waiting.  The path of solid blue arrow lines traces the message path for normal execution.  If anything goes wrong, an exception can be posted to a different callback port, shown with a red outline in the diagram.

This diagram shows one possible composition of services and data flow.  Port sets, which are simply a collection of related ports, are shown as callback receivers in pairs, but they can consist of any number of ports with any mixture of messages types, depending on the needs of the system being coordinated.  In this example, if anything goes wrong in the handler methods at ports 2, 5, or 6, an exception message will be routed to port 6, where another handler method can compensate for or report on the error.  Also note that while during startup this system may have to process data at port 4 before the logic at ports 5, 7, and 8 can run… once it gets going, there could be activity operating at many ports concurrently (not just one port per service).

Arbiters, Dispatchers, and DispatcherQueues

Now it’s time to peel away some of the layers of simplification presented so far.  (It may help to have a beer or glass of wine at this point.)

An arbiter is a rule (or set of rules) about when and how to process messages for a specific port (or set of ports).  (It is helpful to think of arbiter as a data flow or message flow coordinator.)  Should messages be pulled off the queue as soon as they arrive?  Should the software wait until 5 messages have arrived before processing them all as a group?  Should messages be checked according to a timer firing every 20 seconds?  Should logic be run only when two ports have messages waiting (a join)?  What logic should be run when one of these conditions occurs?  Can method handlers on three specific ports run concurrently until a message arrives on a fourth port, whose handler must run exclusively, and when done the other three can run again (interleave)?  These are just a few of the many coordination patterns that can be expressed with different types of arbiters (and hierarchically nested arbiters, which are ingenious).

Figure 4 illustrates that an arbiter is associated with a port to monitor and a method to execute under the right conditions.  The logic of the arbiter, depending on its type and configuration, determines whether to handle the message.  It gets its thread from a thread dispatcher, which contains a thread pool.  (Not the same as System.Threading.ThreadPool, though, as there can only be one of those per process.)

The next diagram (figure 5) could represent a join coordination.  An arbiter waits for messages on two ports, and depending on how it’s defined, it may process messages from one port repeatedly, but as soon as it receives a message on the second port (it may be an exception port, for example), the whole arbiter might tear itself down so that no more handling on those port will occur.  As you are probably starting to see, composition and attachment of arbiters are key to controlling specific patterns of coordination in arbitrarily powerful and flexible ways.

In the Concurrency & Coordination Runtime, we can attach and detach these arbiters during runtime; we don’t have to define them statically at compile time.  There has been some criticism towards this approach because dynamic arbitration rules are much more difficult to verify formally with analysis, and are therefore difficult to optimize compilation and thread management for, but the advantages of this flexibility are enormous and the performance (see this paper by Chrystanthakopoulos and Singh) has been very impressive compared to conventional multithreading approaches.  Ultimately, it’s not about whether we can guaranty 100% that nothing will go wrong using only the mathematical models currently in our repertoire, but whether we can be productive with these techniques to release software that meets acceptable quality standards across a broad range of application domains.

I don’t think we’re going to find a better set of strategies to work with anytime soon, and when we’ve pushed this technology hard enough, the tactics will be fine tuned and we can start absorbing some of these coordination concepts into languages themselves (without sacrificing the dynamism that a library of composable parts provides).  People are going to attempt concurrent programming whether it’s safe or not, and using a library such as the CCR significantly reduces the risk of ad hoc multi-threading code.

Cooperative Multitasking

When mainstream operating systems like Windows took their first steps to support multi-tasking, cooperative versus preemptive multi-tasking was a common topic.  The idea of an operating system depending on applications to surrender control in a well-behaved way was generally and rightfully considered a bad idea.  Any kind of error or instability in software could easily bring down the entire operating system, and enforcing a quickly growing community of software vendors to share nicely wasn’t a realistic option.  Being preemptive meant that the OS could forcefully stop an application from running after giving it a small, measured slice of time, and then switch the thread to a new context where another application could run for another time slice.  Regardless of how poorly applications ran, as benevolent dictator, the OS could ensure a fair scheduling of processor time.

The solution encapsulated in the Concurrency & Coordination Runtime is, on the other hand, a cooperative multi-tasking strategy.  However, because it operates within the local scope of an individual process, and is isolated from other processes in the same OS, its risk of destabilizing the system is nonexistent.  This deep level of cooperation, in fact, is what gives the CCR its great performance.  When used correctly, which George Chrysanthakopoulos (in this video) and his colleagues have brilliantly put within our reach in the CCR library, threads don’t sit around waiting on some resource or for long-running operations to complete; instead, control is freely surrendered back to the thread pool, where it is quickly assigned to a new task.

Finally, by surrendering threads freely instead of holding onto them, a continuous flow of threads through the different activities of the system is maintained, and there is therefore always an abundance of them to handle new messages waiting on ports.  Existing threads are not wasted.  As the Tao Te Ching says:

If you want to accord with the Tao,
just do your job, then let go.

Control & Data Flow: Sequential vs. Concurrent

In sequential programs, stacks are used to unwind method calls and provide return values (return messages), and threads follow the data flow; whereas in port-based programming, threads are managed by one or more thread dispatchers that are capable of maximizing the use of that thread by making it available in a pool and sharing it with with many other (potentially unrelated) tasks.  Data flows orthogonally and according to a different coordination strategy than control flow.  This task-thread agnosticism (the opposite of thread-affinity) is similar to the statelessness of a web server such as IIS; one or more threads from a large pool are injected into the tasks of processing, rendering, and serving up huge numbers of web pages, after which those threads are recycled back into the thread pool for execution of other tasks for a highly concurrent and scalable service platform.

So herein lies the trick: in order to split this coupling between data flow and control flow, a different means is needed to compose the two coordination strategies.  In C# and other popular imperative programming languages, methods implicitly pass thread control along with data arguments (the message), and the use of the stack for method calls asserts constraints on control flow, so making the CCR work involves some interesting patterns.

That’s why port-based programming is hard to get your head around.  It’s such a large shift from common sequential logic and it requires some additional planning (and good visualizations).  It’s obviously important to have a good set of tools for expressing that coordination, a simple set of conceptual primitives that allows us to compose arbitrarily-complex coordination patterns.  These primitives, including Message, Port, PortSet, Dispatcher (thread pool), and others provide the building blocks that we can use to define these relationships.  Once we define what we want to happen with these building blocks, the CCR can make it all happen.

This level of coordination is a level beyond the strategies used by most concurrent applications and frameworks in software today, primarily because there hasn’t been a pressing need for it until recently–processors had been growing phenomenally in speed for many years.  Now, however, we’re told that processor speed has plateaued, that we now have to scale out to scale up, spreading the processing load across multiple cores.  We are very fortunate that the work being done by researchers in fields like robotics can be applied in other service oriented architectures, and is a testament to the widespread use of the .NET Framework and the fantastic efforts of some very bright individuals.

Where to Find Microsoft Robotics Studio

Robotics Studio is a free download and can be found here, and while it is full of other good stuff, it’s worth checking out for the Concurrency and Coordination Runtime alone.